guychuk commited on
Commit
4121761
·
verified ·
1 Parent(s): cd5ca95

Publish release artefacts + model card + MANIFEST

Browse files
Files changed (7) hide show
  1. MANIFEST.json +22 -0
  2. README.md +274 -0
  3. best.pt +3 -0
  4. eval_wf_full.json +3292 -0
  5. jevbench_summary.json +2824 -0
  6. probe_results.json +155 -0
  7. results.json +2248 -0
MANIFEST.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "release": "typical-small",
3
+ "public_repo": "OzLabs/typical-small",
4
+ "public_url": "https://huggingface.co/OzLabs/typical-small",
5
+ "date": "2026-09-21",
6
+ "publish_date": "2026-09-21",
7
+ "source_git_commit": "1ff61f4c40b84ba28983d2258e96ffc759bda162",
8
+ "source_checkpoint": "ts1b",
9
+ "parent": "typical-small-preview",
10
+ "parent_public_repo": "OzLabs/typical-small-preview",
11
+ "lineage": "typical-small-preview -> typical-small",
12
+ "source_hf_repo": "guychuk/pcdm-runs",
13
+ "source_hf_path": "typical-small/",
14
+ "files": {
15
+ "best.pt": "sha256:f62d6e083ea80a08716d901f8409510b6a97cfe00bc65db25a37691ed555c46e",
16
+ "results.json": null,
17
+ "eval_wf_full.json": null,
18
+ "probe_results.json": null,
19
+ "jevbench_summary.json": null
20
+ },
21
+ "notes": "Public mirror of guychuk/pcdm-runs/typical-small/{best.pt,results.json,eval_wf_full.json,probe_results.json,jevbench_summary.json}, downloaded and re-uploaded byte-identical (best.pt sha256 verified against releases/MANIFEST.json before and after upload). results_nomm.json (the pre-eval-only-rerun results file) present in the private source repo was not published here. README.md is a newly authored public model card, not copied from the source repo."
22
+ }
README.md ADDED
@@ -0,0 +1,274 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: Qwen/Qwen3-1.7B-Base
4
+ library_name: pytorch
5
+ tags:
6
+ - decision-model
7
+ - system-one
8
+ - calibrated-probabilities
9
+ - typed-decisions
10
+ - jev-like
11
+ pipeline_tag: text-classification
12
+ datasets:
13
+ - stanfordnlp/snli
14
+ - nyu-mll/multi_nli
15
+ - facebook/anli
16
+ - google/boolq
17
+ - cais/mmlu
18
+ - allenai/ai2_arc
19
+ - allenai/openbookqa
20
+ - tau/commonsense_qa
21
+ - allenai/sciq
22
+ - allenai/qasc
23
+ - jeggers/logiqa2_formatted
24
+ - deepmind/aqua_rat
25
+ - openlifescienceai/medmcqa
26
+ - cua-ai/cua-s1-forms
27
+ - dwidlee/systemone-lite-general
28
+ - MagaBitmex/jev-4b-distill-data
29
+ - Zhengping/UNLI
30
+ - metaeval/ambient
31
+ ---
32
+
33
+ # typical-small
34
+
35
+ **Credit: OzLabs.**
36
+
37
+ ## What this is
38
+
39
+ `typical-small` is a Jev-style System-One decision model: given a **state** (context), a
40
+ **question** (instructions + rubric/criteria), and a **runtime-defined label set** (not a fixed
41
+ vocabulary — the candidates are passed in at inference time), it returns a calibrated probability
42
+ distribution over those labels plus an explicit abstain (∅) option. It implements three typed
43
+ decision primitives directly: **Choice** (categorical, N-way, native contextual readout),
44
+ **Score** (ordinal, K-way Choice trained with ordinal-smoothed targets), and **Noul** (yes/no, a
45
+ dedicated Bernoulli head routed per row). The model does not generate text; it produces a
46
+ probability vector (plus abstain mass) over whatever options the caller supplies for that decision.
47
+
48
+ Release 1 (PLAN7 Tracks A–D): a clean retrain of the 1.7B checkpoint, `ts1b`, uploaded verbatim to
49
+ `guychuk/pcdm-runs` under `typical-small/`. This is the frozen successor to `typical-small-preview`
50
+ (`nc_v3_tap20_wf`) — REPORT.md §3ae's verdict, unchanged. Parent release: `typical-small-preview`.
51
+
52
+ ## What this model is (architecture)
53
+
54
+ - **Backbone:** `Qwen/Qwen3-1.7B-Base`, truncated at layer 20 of 28 — same tap as the preview.
55
+ - **Adaptation:** LoRA, rank 16, on the top 8 layers (13–20) of the truncated backbone.
56
+ - **Readout:** native contextual decision head (`nc_head=n3`), `letters_nonull` rendering, *factored*
57
+ null (∅ is a separate head decision, not a rendered candidate) — identical shape to the preview.
58
+ - **Typed primitives** (PLAN7 Track C, new in this release):
59
+ - **Choice** — the N3 contextual readout above, unchanged.
60
+ - **Noul** (yes/no) — a Bernoulli head, `P(yes) = σ(w·h_D)`, on the decision state with a
61
+ query-only suffix (no rendered candidates). Routed **per row** (commit `7b9d520`): a row only
62
+ goes through the Bernoulli head if its candidate set is exactly `{yes, no}`; every other 2-way
63
+ label set (e.g. `{true, false}`, `{approve, reject}`) is scored as ordinary Choice. The first
64
+ `ts1` run collapsed to chance because this routing was global instead of per-row (REPORT.md
65
+ §3ae) — `ts1b` is the fix, gated by 140 tests.
66
+ - **Score** — K-way Choice trained with ordinal-smoothed targets (τ = 0.7, §3ac's adopted recipe).
67
+ Note: `results.json["args"]` was regenerated by an `--eval_only` rerun (to restore missing MMLU
68
+ sets) and therefore records the *eval* command's flags (`ordinal_smooth: 0.0`); the training
69
+ args stored inside `best.pt` are authoritative: `ordinal_smooth 0.7`, `noul_head bern`,
70
+ `family_weights E:0.40,K:0.15,W:0.35,U:0.10`, `null_aug W:0.20`, `max_state 1024`, `bs 64`,
71
+ `grad_accum 4`, `steps 12000`. The original results file is preserved as `results_nomm.json`
72
+ (not included in this public release).
73
+
74
+ ## Training args (from `results.json["args"]` of the eval-only rerun — see the Score note above; `best.pt` holds the training args)
75
+
76
+ ```
77
+ backbone: Qwen/Qwen3-1.7B-Base
78
+ lora_layers: 8 lora_r: 16 lora_lr: 0.0001
79
+ tap_layer: 20 zscore: false
80
+ readout: native nc_head: n3 nc_render: letters_nonull
81
+ null: factored noul_head: bern score_head: choice ordinal_smooth: 0.0 # eval-only rerun flag; training used 0.7 (see above)
82
+ steps: 12000 bs: 64 grad_accum: 1
83
+ eval_every: 4000 val_every: 1000 ckpt_every: 2000
84
+ data: data_v5 extra_data: data_kb,data_wf,data_wf_hf,data_wf_long,data_wh,data_u
85
+ family_weights: E:0.40,K:0.15,W:0.35,U:0.10
86
+ bucket_map: data_wf_long=W,data_wh=W,data_u=U
87
+ null_aug: W:0.20 max_state: 1024
88
+ tower_d: 512 tower_layers: 2 tower_heads: 8
89
+ seed: 0
90
+ ```
91
+
92
+ Final val NLL 0.376, temperature T = 1.124, null_offset 0.0 (fit post-hoc on the `val` split, applied
93
+ at eval).
94
+
95
+ ## Data
96
+
97
+ Family-balanced sampler, E .40 / K .15 / W .35 / U .10 of every training batch, plus `--null_aug
98
+ W:0.20` (15–20% of W rows get their gold candidate removed so the target is ∅) and `--max_state 1024`
99
+ (vs. the preview's 256):
100
+
101
+ - **`data_v5`** — evidence tasks `stanfordnlp/snli`, `nyu-mll/multi_nli`, `facebook/anli`,
102
+ `google/boolq` plus a wide span of classification label vocabularies at low rows-per-source, so
103
+ the model can't memorize a fixed set of label vocabularies.
104
+ - **`data_kb`** — closed-book knowledge-MCQ distillation from `cais/mmlu` (auxiliary_train),
105
+ `allenai/ai2_arc`, `allenai/openbookqa`, `tau/commonsense_qa`, `allenai/sciq`, `allenai/qasc`,
106
+ `jeggers/logiqa2_formatted`, `deepmind/aqua_rat`, `openlifescienceai/medmcqa`.
107
+ - **`data_wf` / `data_wf_hf`** — unchanged from the preview: our own rubric-conditioned
108
+ typed-workflow corpus (Choice/Noul/Score, held-out rubric styles), plus `cua-ai/cua-s1-forms`
109
+ (60k, MIT), `dwidlee/systemone-lite-general` (20k, MIT), `MagaBitmex/jev-4b-distill-data` (6.6k,
110
+ Apache-2.0). See `releases/typical-small-preview.md` (this org's preview release card) for the
111
+ per-source breakdown.
112
+ - **`data_wf_long`** — 1,024-token-state variant of the workflow corpus (PLAN7 §3z/§3aa): the same
113
+ in-repo rubric-conditioned generator at longer state length, added because long-state families
114
+ (policy documents, multi-hop) were untrained at 256 tokens. No external license constraints.
115
+ - **`data_wh`** (`scripts/decisionmix_v2.py`, DecisionMix v2, PLAN7 Track D): 60,605 train / 3,000
116
+ val / 6 eval files, a programmatic rule engine (12 domains × 6 boolean conditions, levels 1–7),
117
+ 100% of train rows in counterfactual rubric groups (same state+candidates, ≥2 distinct golds).
118
+ Generated in-repo, no external license constraints. Held out: 1 domain/level
119
+ (`wh_heldout_grammar`), 3 whole domains (`wh_heldout_family`), 2 rubric styles
120
+ (`wh_heldout_style`), and level 7 — temporal/numeric/probability/trade-off composition — which is
121
+ eval-only, never trained on. 0/231 JevBench leak hits.
122
+ - **`data_u`** (same script, DecisionMix v2): 31,029 train / 2,000 val / 3 eval files, 100%
123
+ soft-target rows. Real sources: `Zhengping/UNLI`'s *validation* split (train/test already
124
+ consumed elsewhere) and `metaeval/ambient`'s ambiguous rows (uniform target over listed labels).
125
+ `metaeval/chaos-mnli-ambiguity` is **eval-only** (MNLI-only mirror of `chaos-nli`, which 404s),
126
+ never trained on. The remaining 92% is four synthetic generators with exact closed-form targets
127
+ (partial evidence, noisy-sensor Bayes posterior, ordinal confusion, conflicting-sources log-odds),
128
+ each with a held-out parameter range, generated in-repo. 0/231 JevBench leak hits.
129
+
130
+ See PLAN7.md's "Track D — DecisionMix v2 build notes" for the full corpus notes, and REPORT.md
131
+ §3z–§3ae for the mixture-sweep and Release-1-candidate history.
132
+
133
+ ## Results
134
+
135
+ ### §3ae — Release-1 candidate at 1.7B (`ts1b`) vs the preview and the intermediate `r1_dmv2`
136
+
137
+ | | `typical-small-preview` (§3t/§3w) | `r1_dmv2` (§3ad) | **`ts1b`** | 1.7B base |
138
+ |---|---|---|---|---|
139
+ | CLINC-150 / TREC-fine / HWU64 / 20NG | .734 / .372 / .735 / .522 | .820 / .474 / .726 / .501 | .804 / **.508** / .761 / .540 | .845 / .468 / .757 / .515 |
140
+ | SNLI / MNLI / BoolQ / ANLI | .894 / .849 / .831 / .481 | .894 / .852 / .834 / .501 | .894 / .859 / .824 / .497 | .904 / .865 / .834 / .507 |
141
+ | MMLU-Pro among-K / Δ_q_sh / false-abstain | .330 / .119 / .006 | .333 / .110 / – | .343 / .127 / .005 | .353 / .122 / .003 |
142
+ | held-out noul / score / style / flip both-correct (full rows) | .699 / .498 / .901 / .565 | .658 / .501 / .876 / .519 | **.715** / **.520** / .859 / **.581** | – |
143
+ | held-out score NLL / typed-decisions NLL (full rows) | 2.03 / 2.06 | 1.56 / 1.71 | **1.01 / 1.25** | – / 1.22 |
144
+ | wh held-out family / grammar / style / level 7 / flip | – | .827 / .881 / .888 / .498 / .710 | **.836 / .898 / .896** / .495 / **.718** | – |
145
+ | u ChaosNLI / real held-out / synthetic (acc, NLL) | – | .537 (1.00) / .623 (.67) / .909 (.67) | .547 (1.00) / **.649** (.66) / .916 (.67) | – |
146
+ | external: PagerDuty (floor .792) / jevlogs (.697) / Mind2Web (.427) / tree-choice / typed-decisions acc | .779 / .522 / .427 / .539 / .457 | .651† / .500† / .346† / – / .421† | **.817 / .710** / .357 / .474 / **.510** | .776 / .697 / .300 / .506 / .487 |
147
+ | JevBench std / easy / hard (Brier std / hard; ECE std) | .750 / 1.00 / .387 (.40 / .88; .15) | .750 / 1.00 / .441 (.42 / .79) | .694 / 1.00 / .432 (.40 / .79; **.11**) | .694 / 1.00 / .378 (.47 / .74) |
148
+ | Noul reversed-label |ΔP(yes)| max | up to .55 | – | **0** (10/11 sets; .009 on one) | – |
149
+
150
+ † `r1_dmv2`'s external numbers are a train-time 1,500-row/256-token pass, not re-run at full length.
151
+
152
+ **Read (REPORT.md §3ae).** Against the frozen preview, `ts1b` keeps E and K inside budget (CLINC −4 vs
153
+ the untrained base is the one edge; TREC/20NG land above base), raises held-out noul/score/flip,
154
+ halves the soft-target NLLs (score 2.03 → 1.01, typed-decisions 2.06 → 1.25), beats or ties `r1_dmv2`
155
+ on every curriculum and uncertainty set, is exactly order-invariant on Noul, and is the first 1.7B
156
+ model above the constant-prediction floor on PagerDuty (.817) and jevlogs (.710, marginal). It gives
157
+ back held-out styles (−4), Mind2Web (below floor again), tree-choice (−6), and JevBench standard
158
+ (.750 → .694, ~1 SE at n_eff = 36; extraction 12 → 9 of 12) while JevBench hard rises (.387 → .432,
159
+ Brier .88 → .79). **Decision: freeze `ts1b` as `typical-small` (Release 1 at 1.7B)**, with the preview
160
+ kept as the reference.
161
+
162
+ ### JevBench disclosures (REPORT.md §3q — apply to the numbers above)
163
+
164
+ This is a **public-subset run, not a ranked entry** (JevBench's own leaderboard requires ≥95%
165
+ coverage including the 146 non-public judge items; we ran the 231 public ids only). Reported
166
+ probabilities are the head's softmax **conditioned on non-∅** (P(∅) is dropped and the rest
167
+ renormalized; mean p_null ≈ .03, native). Checkpoints trained at up to 1,024-token states / 64-token
168
+ queries are run at up to 4096 / 256 tokens (out-of-training-length but not truncated). Latency is
169
+ in-process on one H100, one decision at a time, model load excluded. Cost is null (no tariff — this
170
+ is not a hosted-API run). Option order is the harness's own label order; a reversed-order control on
171
+ a related checkpoint swung standard accuracy ±4 points, larger than the differences between our own
172
+ model variants, so no fine ranking among our own checkpoints is supported by these numbers alone.
173
+ Majority/chance baselines on this split are .311 standard / .284 easy / .336 hard (n=72 standard,
174
+ SE ≈ .058; n_eff = 36) — `ts1b`'s hard tier (.432) is within 1 SE of chance, standard (.694) is ~1 SE
175
+ below the preview.
176
+
177
+ ## Latency (REPORT.md §3ab, apples-to-apples ladder, one pod, one torch build, L_s = 256)
178
+
179
+ | 1.7B | single decision, K = 2 / 32 / 256 (ms) | marginal per query, M = 32, K = 2 / 32 / 256 (ms) | peak memory, K = 2 → 256 (GB) |
180
+ |---|---|---|---|
181
+ | `ts1b`-shape recipe | 45 / 46 / 106 | 2.7 / 3.7 / 28.0 | 6.4 → 9.3 |
182
+
183
+ A single decision costs 45 ms regardless of most of K (the KV-cached state and a short suffix
184
+ dominate); batched marginal cost is 2.7 ms/query at K = 2, rising to 28 ms/query at K = 256. Capability
185
+ per millisecond (JevBench standard / single-decision ms) puts the 4B at the knee (.833/56 vs 1.7B's
186
+ .750/45 vs 14B's .875/60) — this checkpoint is the cheapest point on that ladder, not the highest-
187
+ capability one.
188
+
189
+ ## Known limitations
190
+
191
+ - JevBench standard is ~1 SE below the preview (.694 vs .750, n_eff = 36) — a real trade, not noise
192
+ in the other direction: `ts1b` buys held-out noul/score/flip and both external floors (PagerDuty,
193
+ jevlogs) at that cost (§3ae).
194
+ - Mind2Web stays below its constant-prediction floor (.357 vs floor .427) — general web-agent
195
+ candidate sets remain outside what this training distribution covers.
196
+ - Level-7 composition (temporal / unit / expected-value / trade-off reasoning, `wh_level7`) sits at
197
+ ~.50 — chance — same as every other checkpoint on the ladder (§3ad); nothing trained so far moves
198
+ it, because the generator never produces this composition.
199
+ - Hard tier is .432 with Brier .79 — within 1 SE of the .336 chance baseline; probability quality on
200
+ soft/ordinal gold is still the clearest open defect at this size (§3aa, §3ae).
201
+ - Held-out rubric styles regress −4 vs the preview (.859 vs .901).
202
+ - Very large candidate sets (K in the hundreds to thousands) still need the energy → top-r → native
203
+ path, not this native head directly (unchanged from the preview).
204
+
205
+ ## Licensing of training data
206
+
207
+ - Backbone (`Qwen/Qwen3-1.7B-Base`): Apache-2.0. This release's weights are a LoRA adapter plus
208
+ small Choice/Score/Noul heads over that backbone; the combined artifact is released under
209
+ Apache-2.0.
210
+ - `stanfordnlp/snli`, `nyu-mll/multi_nli`, `facebook/anli`, `google/boolq`, `cais/mmlu`,
211
+ `allenai/ai2_arc`, `allenai/openbookqa`, `tau/commonsense_qa`, `allenai/sciq`, `allenai/qasc`,
212
+ `jeggers/logiqa2_formatted`, `deepmind/aqua_rat`, `openlifescienceai/medmcqa`: per-source licenses
213
+ as triaged in PLAN6.md's dataset-survey table (standard open NLI/QA/knowledge-MCQ benchmark
214
+ licenses, permissive for research and redistribution of derived model weights).
215
+ - `cua-ai/cua-s1-forms`: MIT. `dwidlee/systemone-lite-general`: MIT. `MagaBitmex/jev-4b-distill-data`:
216
+ Apache-2.0.
217
+ - New in this release (PLAN7 Track D, `data_wh` / `data_u`): `data_wh` is a programmatic rule-engine
218
+ corpus generated in-repo (`scripts/decisionmix_v2.py`), no external license constraints.
219
+ `Zhengping/UNLI`'s validation split (trained on) is MIT. **`metaeval/ambient`** (trained on, its
220
+ ambiguous rows) and **`metaeval/chaos-mnli-ambiguity`** (eval-only, never trained on) **do not
221
+ declare a license on their HF cards** — flagged here, not asserted; treat both as unlicensed for
222
+ redistribution/derivative-rights purposes even though `ambient` contributed rows to training.
223
+ - **Evaluation-only, never trained on:** JevBench (`fstandhartinger/jevbench`, public subset),
224
+ `LocalLLaMA/typed-decisions`, `reachjalil/jev-luna-pagerduty-trigger` (PagerDuty), `reachjalil/
225
+ jev-tree-choice-cap` (tree-choice), `reachjalil/jevlogs-log-triage-benchmark` (jevlogs — Loghub
226
+ research license, caveated, block-level labels, never used in commercial claims),
227
+ `AndeyTait/JevForge-Mind2Web` (Mind2Web), and `metaeval/chaos-mnli-ambiguity` (ChaosNLI, MNLI-only
228
+ mirror).
229
+ - JevBench numbers in this card are a **public-subset run** against `fstandhartinger/jevbench` v1.2.1
230
+ (72 standard / 48 easy / 111 hard public ids) — not a submitted or ranked leaderboard entry.
231
+
232
+ ## How to run
233
+
234
+ Inference currently requires this project's code
235
+ (`pcdm_jev.decider.PCDMDecider`, `native.py`), which is **not yet public — code release to follow.**
236
+ The snippet below documents the intended call shape, not something runnable from this repo alone:
237
+
238
+ ```python
239
+ from pcdm_jev.decider import PCDMDecider
240
+
241
+ decider = PCDMDecider("runs/ts1b", mode="native")
242
+ question = {"instructions": "...", "type": "choice", "criteria": {"a": "option a", "b": "option b"}}
243
+ probs, runtime = decider.decide(state="...", question=question, labels=["a", "b"])
244
+ # probs: dict label -> probability; runtime["p_null"] is the abstention mass
245
+ ```
246
+
247
+ (`PCDMDecider.__init__(run_dir=None, device="auto", mode="energy", energy_run=None, max_state=4096,
248
+ max_query=256, backbone=None, tap_layer=0, shots=0)` — pass `mode="native"` and the checkpoint dir.)
249
+
250
+ Held-out workflow / external eval reproduction (post-hoc, full state length; also requires the
251
+ project code):
252
+
253
+ ```bash
254
+ uv run --no-sync python scripts/eval_wf.py --run runs/ts1b --mode native \
255
+ --files data_wf/eval/*.jsonl data_wf_hf/eval/*.jsonl data_wh/eval/*.jsonl data_u/eval/*.jsonl \
256
+ --limit 0 --out runs/ts1b/eval_wf_full.json
257
+ ```
258
+
259
+ JevBench: `scripts/jevbench_run.py` against the checkpoint in `mode="native"` (see `pcdm_jev/` for the
260
+ harness adapter and REPORT.md §3q for the exact protocol and disclosures reproduced above).
261
+
262
+ ## Files in this repository
263
+
264
+ - `best.pt` — this checkpoint (`ts1b`).
265
+ - `results.json` — training args + full eval suite (train-time pass).
266
+ - `eval_wf_full.json` — post-hoc held-out workflow / external eval (full state length, `scripts/eval_wf.py`).
267
+ - `probe_results.json` — Δ_q probe results.
268
+ - `jevbench_summary.json` — JevBench public-subset run summary.
269
+ - `MANIFEST.json` — checksum + provenance record for this release.
270
+
271
+ ## Public release
272
+
273
+ - https://huggingface.co/OzLabs/typical-small
274
+ - Parent release: https://huggingface.co/OzLabs/typical-small-preview
best.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f62d6e083ea80a08716d901f8409510b6a97cfe00bc65db25a37691ed555c46e
3
+ size 33945265
eval_wf_full.json ADDED
@@ -0,0 +1,3292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "runs/ts1b",
3
+ "mode": "native",
4
+ "max_state": 4096,
5
+ "limit": 0,
6
+ "eval": {
7
+ "u_chaosnli": {
8
+ "n": 1594,
9
+ "n_total": 1594,
10
+ "suffix_overflow": 0,
11
+ "raw": {
12
+ "nll": 0.9986934642201571,
13
+ "brier": 0.15597729602040825,
14
+ "jsd": 0.06276268334293099,
15
+ "acc": 0.5470514429109159,
16
+ "acc_k": 0.5470514429109159,
17
+ "ece": 0.05734760788184035,
18
+ "sel_acc80": 0.5626959247648903,
19
+ "coverage": 1.0,
20
+ "false_abstain": 0.0,
21
+ "null_recall": "n/a"
22
+ },
23
+ "by_family": {
24
+ "": {
25
+ "nll": 0.9986934642201571,
26
+ "brier": 0.15597729602040825,
27
+ "jsd": 0.06276268334293099,
28
+ "acc": 0.5470514429109159,
29
+ "acc_k": 0.5470514429109159,
30
+ "ece": 0.05734760788184035,
31
+ "sel_acc80": 0.5626959247648903,
32
+ "coverage": 1.0,
33
+ "false_abstain": 0.0,
34
+ "null_recall": "n/a",
35
+ "n": 1594
36
+ }
37
+ },
38
+ "by_qtype": {},
39
+ "by_label_class": {
40
+ "answer": {
41
+ "nll": 0.9986934642201571,
42
+ "brier": 0.15597729602040825,
43
+ "jsd": 0.06276268334293099,
44
+ "acc": 0.5470514429109159,
45
+ "acc_k": 0.5470514429109159,
46
+ "ece": 0.05734760788184035,
47
+ "sel_acc80": 0.5626959247648903,
48
+ "coverage": 1.0,
49
+ "false_abstain": 0.0,
50
+ "null_recall": "n/a",
51
+ "n": 1594
52
+ }
53
+ },
54
+ "baselines": {
55
+ "majority_class": 0,
56
+ "majority_acc": 0.46361355081555833
57
+ }
58
+ },
59
+ "u_real_heldout": {
60
+ "n": 562,
61
+ "n_total": 562,
62
+ "suffix_overflow": 0,
63
+ "raw": {
64
+ "nll": 0.6608096977843964,
65
+ "brier": 0.13465855536230573,
66
+ "jsd": 0.05983032012095985,
67
+ "acc": 0.6494661921708185,
68
+ "acc_k": 0.6494661921708185,
69
+ "ece": 0.061910581334205725,
70
+ "sel_acc80": 0.6911111111111111,
71
+ "coverage": 1.0,
72
+ "false_abstain": 0.0,
73
+ "null_recall": "n/a"
74
+ },
75
+ "by_family": {
76
+ "": {
77
+ "nll": 0.6608096977843964,
78
+ "brier": 0.13465855536230573,
79
+ "jsd": 0.05983032012095985,
80
+ "acc": 0.6494661921708185,
81
+ "acc_k": 0.6494661921708185,
82
+ "ece": 0.061910581334205725,
83
+ "sel_acc80": 0.6911111111111111,
84
+ "coverage": 1.0,
85
+ "false_abstain": 0.0,
86
+ "null_recall": "n/a",
87
+ "n": 562
88
+ }
89
+ },
90
+ "by_qtype": {},
91
+ "by_label_class": {
92
+ "answer": {
93
+ "nll": 0.6608096977843964,
94
+ "brier": 0.13465855536230573,
95
+ "jsd": 0.05983032012095985,
96
+ "acc": 0.6494661921708185,
97
+ "acc_k": 0.6494661921708185,
98
+ "ece": 0.061910581334205725,
99
+ "sel_acc80": 0.6911111111111111,
100
+ "coverage": 1.0,
101
+ "false_abstain": 0.0,
102
+ "null_recall": "n/a",
103
+ "n": 562
104
+ }
105
+ },
106
+ "baselines": {
107
+ "majority_class": 0,
108
+ "majority_acc": 0.5071174377224199
109
+ }
110
+ },
111
+ "u_synthetic_heldout": {
112
+ "n": 475,
113
+ "n_total": 475,
114
+ "suffix_overflow": 0,
115
+ "raw": {
116
+ "nll": 0.6666821294215001,
117
+ "brier": 0.031944849397233334,
118
+ "jsd": 0.01431239917904498,
119
+ "acc": 0.9157894736842105,
120
+ "acc_k": 0.9157894736842105,
121
+ "ece": 0.16440801394613166,
122
+ "sel_acc80": 0.968421052631579,
123
+ "coverage": 1.0,
124
+ "false_abstain": 0.0,
125
+ "null_recall": "n/a"
126
+ },
127
+ "by_family": {
128
+ "": {
129
+ "nll": 0.6666821294215001,
130
+ "brier": 0.031944849397233334,
131
+ "jsd": 0.01431239917904498,
132
+ "acc": 0.9157894736842105,
133
+ "acc_k": 0.9157894736842105,
134
+ "ece": 0.16440801394613166,
135
+ "sel_acc80": 0.968421052631579,
136
+ "coverage": 1.0,
137
+ "false_abstain": 0.0,
138
+ "null_recall": "n/a",
139
+ "n": 475
140
+ }
141
+ },
142
+ "by_qtype": {},
143
+ "by_label_class": {
144
+ "answer": {
145
+ "nll": 0.6666821294215001,
146
+ "brier": 0.031944849397233334,
147
+ "jsd": 0.01431239917904498,
148
+ "acc": 0.9157894736842105,
149
+ "acc_k": 0.9157894736842105,
150
+ "ece": 0.16440801394613166,
151
+ "sel_acc80": 0.968421052631579,
152
+ "coverage": 1.0,
153
+ "false_abstain": 0.0,
154
+ "null_recall": "n/a",
155
+ "n": 475
156
+ }
157
+ },
158
+ "baselines": {
159
+ "majority_class": 1,
160
+ "majority_acc": 0.5747368421052632
161
+ }
162
+ },
163
+ "wf_heldout_choice": {
164
+ "n": 1500,
165
+ "n_total": 1500,
166
+ "suffix_overflow": 0,
167
+ "raw": {
168
+ "nll": 0.02220144267887235,
169
+ "brier": 0.009909953890114582,
170
+ "jsd": 0.0069410462861019965,
171
+ "acc": 0.994,
172
+ "acc_k": 0.9950460014154282,
173
+ "ece": 0.010105259974797547,
174
+ "sel_acc80": 1.0,
175
+ "conf_wrong": 0.0,
176
+ "coverage": 0.9433333333333334,
177
+ "false_abstain": 0.0,
178
+ "null_recall": 0.9770114942528736,
179
+ "auroc_null": 1.0,
180
+ "auroc_conf": 0.999487517387803
181
+ },
182
+ "by_family": {
183
+ "tool_select": {
184
+ "nll": 0.02220144267887235,
185
+ "brier": 0.009909953890114582,
186
+ "jsd": 0.0069410462861019965,
187
+ "acc": 0.994,
188
+ "acc_k": 0.9950460014154282,
189
+ "ece": 0.010105259974797547,
190
+ "sel_acc80": 1.0,
191
+ "conf_wrong": 0.0,
192
+ "coverage": 0.9433333333333334,
193
+ "false_abstain": 0.0,
194
+ "null_recall": 0.9770114942528736,
195
+ "auroc_null": 1.0,
196
+ "auroc_conf": 0.999487517387803,
197
+ "n": 1500
198
+ }
199
+ },
200
+ "by_qtype": {
201
+ "choice": {
202
+ "nll": 0.02220144267887235,
203
+ "brier": 0.009909953890114582,
204
+ "jsd": 0.0069410462861019965,
205
+ "acc": 0.994,
206
+ "acc_k": 0.9950460014154282,
207
+ "ece": 0.010105259974797547,
208
+ "sel_acc80": 1.0,
209
+ "conf_wrong": 0.0,
210
+ "coverage": 0.9433333333333334,
211
+ "false_abstain": 0.0,
212
+ "null_recall": 0.9770114942528736,
213
+ "auroc_null": 1.0,
214
+ "auroc_conf": 0.999487517387803,
215
+ "n": 1500
216
+ }
217
+ },
218
+ "by_label_class": {
219
+ "answer": {
220
+ "nll": 0.020595754550765025,
221
+ "brier": 0.008593300755277252,
222
+ "jsd": 0.006623562877006804,
223
+ "acc": 0.9950460014154282,
224
+ "acc_k": 0.9950460014154282,
225
+ "ece": 0.011593746025344638,
226
+ "sel_acc80": 1.0,
227
+ "coverage": 1.0,
228
+ "false_abstain": 0.0,
229
+ "null_recall": "n/a",
230
+ "n": 1413
231
+ },
232
+ "null": {
233
+ "nll": 0.04828003262158096,
234
+ "brier": 0.03129421687316223,
235
+ "jsd": 0.012097414757958358,
236
+ "acc": 0.9770114942528736,
237
+ "ece": 0.018105437015664985,
238
+ "sel_acc80": 1.0,
239
+ "conf_wrong": 0.0,
240
+ "coverage": 0.02298850574712641,
241
+ "false_abstain": "n/a",
242
+ "null_recall": 0.9770114942528736,
243
+ "n": 87
244
+ }
245
+ },
246
+ "baselines": {
247
+ "majority_class": 2,
248
+ "majority_acc": 0.21333333333333335
249
+ }
250
+ },
251
+ "wf_heldout_noul": {
252
+ "n": 1500,
253
+ "n_total": 1500,
254
+ "suffix_overflow": 0,
255
+ "raw": {
256
+ "nll": 0.6509143929801903,
257
+ "brier": 0.4111122025864996,
258
+ "jsd": 0.16130160222798604,
259
+ "acc": 0.7146666666666667,
260
+ "acc_k": 0.7146666666666667,
261
+ "ece": 0.11673003478844962,
262
+ "sel_acc80": 0.7566666666666667,
263
+ "coverage": 1.0,
264
+ "false_abstain": 0.0,
265
+ "null_recall": "n/a"
266
+ },
267
+ "by_family": {
268
+ "eligibility": {
269
+ "nll": 0.6509143929801903,
270
+ "brier": 0.4111122025864996,
271
+ "jsd": 0.16130160222798604,
272
+ "acc": 0.7146666666666667,
273
+ "acc_k": 0.7146666666666667,
274
+ "ece": 0.11673003478844962,
275
+ "sel_acc80": 0.7566666666666667,
276
+ "coverage": 1.0,
277
+ "false_abstain": 0.0,
278
+ "null_recall": "n/a",
279
+ "n": 1500
280
+ }
281
+ },
282
+ "by_qtype": {
283
+ "noul": {
284
+ "nll": 0.6509143929801903,
285
+ "brier": 0.4111122025864996,
286
+ "jsd": 0.16130160222798604,
287
+ "acc": 0.7146666666666667,
288
+ "acc_k": 0.7146666666666667,
289
+ "ece": 0.11673003478844962,
290
+ "sel_acc80": 0.7566666666666667,
291
+ "coverage": 1.0,
292
+ "false_abstain": 0.0,
293
+ "null_recall": "n/a",
294
+ "n": 1500
295
+ }
296
+ },
297
+ "by_label_class": {
298
+ "answer": {
299
+ "nll": 0.6509143929801903,
300
+ "brier": 0.4111122025864996,
301
+ "jsd": 0.16130160222798604,
302
+ "acc": 0.7146666666666667,
303
+ "acc_k": 0.7146666666666667,
304
+ "ece": 0.11673003478844962,
305
+ "sel_acc80": 0.7566666666666667,
306
+ "coverage": 1.0,
307
+ "false_abstain": 0.0,
308
+ "null_recall": "n/a",
309
+ "n": 1500
310
+ }
311
+ },
312
+ "baselines": {
313
+ "majority_class": 0,
314
+ "majority_acc": 0.5833333333333334,
315
+ "always_positive_acc": 0.4166666666666667
316
+ }
317
+ },
318
+ "wf_heldout_score": {
319
+ "n": 1502,
320
+ "n_total": 1502,
321
+ "suffix_overflow": 0,
322
+ "raw": {
323
+ "nll": 1.007726984380891,
324
+ "brier": 0.5732102322665574,
325
+ "jsd": 0.2761964760754948,
326
+ "acc": 0.5199733688415447,
327
+ "acc_k": 0.5206391478029294,
328
+ "ece": 0.04527540038651061,
329
+ "sel_acc80": 0.5740432612312812,
330
+ "coverage": 0.9993342210386151,
331
+ "false_abstain": 0.0006657789613848203,
332
+ "null_recall": "n/a"
333
+ },
334
+ "by_family": {
335
+ "urgency": {
336
+ "nll": 1.007726984380891,
337
+ "brier": 0.5732102322665574,
338
+ "jsd": 0.2761964760754948,
339
+ "acc": 0.5199733688415447,
340
+ "acc_k": 0.5206391478029294,
341
+ "ece": 0.04527540038651061,
342
+ "sel_acc80": 0.5740432612312812,
343
+ "coverage": 0.9993342210386151,
344
+ "false_abstain": 0.0006657789613848203,
345
+ "null_recall": "n/a",
346
+ "n": 1502
347
+ }
348
+ },
349
+ "by_qtype": {
350
+ "score": {
351
+ "nll": 1.007726984380891,
352
+ "brier": 0.5732102322665574,
353
+ "jsd": 0.2761964760754948,
354
+ "acc": 0.5199733688415447,
355
+ "acc_k": 0.5206391478029294,
356
+ "ece": 0.04527540038651061,
357
+ "sel_acc80": 0.5740432612312812,
358
+ "coverage": 0.9993342210386151,
359
+ "false_abstain": 0.0006657789613848203,
360
+ "null_recall": "n/a",
361
+ "n": 1502
362
+ }
363
+ },
364
+ "by_label_class": {
365
+ "answer": {
366
+ "nll": 1.007726984380891,
367
+ "brier": 0.5732102322665574,
368
+ "jsd": 0.2761964760754948,
369
+ "acc": 0.5199733688415447,
370
+ "acc_k": 0.5206391478029294,
371
+ "ece": 0.04527540038651061,
372
+ "sel_acc80": 0.5740432612312812,
373
+ "coverage": 0.9993342210386151,
374
+ "false_abstain": 0.0006657789613848203,
375
+ "null_recall": "n/a",
376
+ "n": 1502
377
+ }
378
+ },
379
+ "baselines": {
380
+ "majority_class": 2,
381
+ "majority_acc": 0.3548601864181092
382
+ }
383
+ },
384
+ "wf_heldout_style": {
385
+ "n": 1627,
386
+ "n_total": 1627,
387
+ "suffix_overflow": 0,
388
+ "raw": {
389
+ "nll": 0.3835512182973631,
390
+ "brier": 0.21472157101975364,
391
+ "jsd": 0.1094748689100086,
392
+ "acc": 0.8586355255070682,
393
+ "acc_k": 0.8564294631710362,
394
+ "ece": 0.0744189031664843,
395
+ "sel_acc80": 0.9362519201228878,
396
+ "conf_wrong": 0.0,
397
+ "coverage": 0.9846342962507683,
398
+ "false_abstain": 0.0,
399
+ "null_recall": 1.0,
400
+ "auroc_null": 1.0,
401
+ "auroc_conf": 1.0
402
+ },
403
+ "by_family": {
404
+ "relevance": {
405
+ "nll": 0.649669278943508,
406
+ "brier": 0.37928978253549256,
407
+ "jsd": 0.20030994327991516,
408
+ "acc": 0.7733333333333333,
409
+ "acc_k": 0.7733333333333333,
410
+ "ece": 0.21542641083399458,
411
+ "sel_acc80": 0.8416666666666667,
412
+ "coverage": 1.0,
413
+ "false_abstain": 0.0,
414
+ "null_recall": "n/a",
415
+ "n": 150
416
+ },
417
+ "support": {
418
+ "nll": 0.19559202627065286,
419
+ "brier": 0.1084407534057103,
420
+ "jsd": 0.05433300069657259,
421
+ "acc": 0.9328859060402684,
422
+ "acc_k": 0.9328859060402684,
423
+ "ece": 0.05307854301977477,
424
+ "sel_acc80": 0.9666666666666667,
425
+ "coverage": 1.0,
426
+ "false_abstain": 0.0,
427
+ "null_recall": "n/a",
428
+ "n": 149
429
+ },
430
+ "adequacy": {
431
+ "nll": 0.31774727529596786,
432
+ "brier": 0.18597336232659226,
433
+ "jsd": 0.07811762183724544,
434
+ "acc": 0.88,
435
+ "acc_k": 0.88,
436
+ "ece": 0.06820252060890195,
437
+ "sel_acc80": 0.9416666666666667,
438
+ "coverage": 1.0,
439
+ "false_abstain": 0.0,
440
+ "null_recall": "n/a",
441
+ "n": 150
442
+ },
443
+ "action_select": {
444
+ "nll": 0.33021187045608336,
445
+ "brier": 0.1823344595149311,
446
+ "jsd": 0.0826433025603459,
447
+ "acc": 0.8733333333333333,
448
+ "acc_k": 0.8633093525179856,
449
+ "ece": 0.04021375119686121,
450
+ "sel_acc80": 0.9416666666666667,
451
+ "conf_wrong": 0.0,
452
+ "coverage": 0.9266666666666666,
453
+ "false_abstain": 0.0,
454
+ "null_recall": 1.0,
455
+ "auroc_null": 1.0,
456
+ "auroc_conf": 1.0,
457
+ "n": 150
458
+ },
459
+ "completeness": {
460
+ "nll": 0.6991468263927095,
461
+ "brier": 0.39821441371968347,
462
+ "jsd": 0.21188899644050688,
463
+ "acc": 0.7449664429530202,
464
+ "acc_k": 0.7449664429530202,
465
+ "ece": 0.18803769630073702,
466
+ "sel_acc80": 0.8083333333333333,
467
+ "coverage": 1.0,
468
+ "false_abstain": 0.0,
469
+ "null_recall": "n/a",
470
+ "n": 149
471
+ },
472
+ "quality": {
473
+ "nll": 0.8470839023804404,
474
+ "brier": 0.4714291525682993,
475
+ "jsd": 0.24669086204119084,
476
+ "acc": 0.6030534351145038,
477
+ "acc_k": 0.6030534351145038,
478
+ "ece": 0.14125171849745824,
479
+ "sel_acc80": 0.6952380952380952,
480
+ "coverage": 1.0,
481
+ "false_abstain": 0.0,
482
+ "null_recall": "n/a",
483
+ "n": 131
484
+ },
485
+ "enum_extract": {
486
+ "nll": 0.04643218567658875,
487
+ "brier": 0.01990835242030093,
488
+ "jsd": 0.01334930443633335,
489
+ "acc": 0.9933333333333333,
490
+ "acc_k": 0.993103448275862,
491
+ "ece": 0.01974968155225116,
492
+ "sel_acc80": 1.0,
493
+ "conf_wrong": 0.0,
494
+ "coverage": 0.9666666666666667,
495
+ "false_abstain": 0.0,
496
+ "null_recall": 1.0,
497
+ "auroc_null": 1.0,
498
+ "auroc_conf": 1.0,
499
+ "n": 150
500
+ },
501
+ "categorical": {
502
+ "nll": 0.2844007555346479,
503
+ "brier": 0.15958706208754614,
504
+ "jsd": 0.08558952069896848,
505
+ "acc": 0.8976377952755905,
506
+ "acc_k": 0.8951612903225806,
507
+ "ece": 0.07876657641778778,
508
+ "sel_acc80": 0.9705882352941176,
509
+ "conf_wrong": 0.0,
510
+ "coverage": 0.9763779527559056,
511
+ "false_abstain": 0.0,
512
+ "null_recall": 1.0,
513
+ "auroc_null": 1.0,
514
+ "auroc_conf": 1.0,
515
+ "n": 127
516
+ },
517
+ "routing": {
518
+ "nll": 0.2783524443014919,
519
+ "brier": 0.14330485597109158,
520
+ "jsd": 0.06294130823518836,
521
+ "acc": 0.8939393939393939,
522
+ "acc_k": 0.8888888888888888,
523
+ "ece": 0.08102671508536198,
524
+ "sel_acc80": 0.9716981132075472,
525
+ "conf_wrong": 0.0,
526
+ "coverage": 0.9545454545454546,
527
+ "false_abstain": 0.0,
528
+ "null_recall": 1.0,
529
+ "auroc_null": 1.0,
530
+ "auroc_conf": 1.0,
531
+ "n": 132
532
+ },
533
+ "severity": {
534
+ "nll": 0.5977246344600028,
535
+ "brier": 0.3145210200360813,
536
+ "jsd": 0.1824036601057107,
537
+ "acc": 0.8734177215189873,
538
+ "acc_k": 0.8734177215189873,
539
+ "ece": 0.2759175228921673,
540
+ "sel_acc80": 0.9375,
541
+ "coverage": 1.0,
542
+ "false_abstain": 0.0,
543
+ "null_recall": "n/a",
544
+ "n": 79
545
+ },
546
+ "policy_permit": {
547
+ "nll": 0.15034539574002462,
548
+ "brier": 0.07961559213419059,
549
+ "jsd": 0.04591845815543799,
550
+ "acc": 0.9463087248322147,
551
+ "acc_k": 0.9463087248322147,
552
+ "ece": 0.05530904523478257,
553
+ "sel_acc80": 0.9833333333333333,
554
+ "coverage": 1.0,
555
+ "false_abstain": 0.0,
556
+ "null_recall": "n/a",
557
+ "n": 149
558
+ },
559
+ "fact": {
560
+ "nll": 0.3212750600631328,
561
+ "brier": 0.189942179795167,
562
+ "jsd": 0.08593823576001701,
563
+ "acc": 0.8828828828828829,
564
+ "acc_k": 0.8828828828828829,
565
+ "ece": 0.08032987729923147,
566
+ "sel_acc80": 0.9213483146067416,
567
+ "coverage": 1.0,
568
+ "false_abstain": 0.0,
569
+ "null_recall": "n/a",
570
+ "n": 111
571
+ }
572
+ },
573
+ "by_qtype": {
574
+ "score": {
575
+ "nll": 0.70689883361143,
576
+ "brier": 0.39849075558714014,
577
+ "jsd": 0.21285725744080541,
578
+ "acc": 0.7367387033398821,
579
+ "acc_k": 0.7367387033398821,
580
+ "ece": 0.19415499583208723,
581
+ "sel_acc80": 0.821078431372549,
582
+ "coverage": 1.0,
583
+ "false_abstain": 0.0,
584
+ "null_recall": "n/a",
585
+ "n": 509
586
+ },
587
+ "noul": {
588
+ "nll": 0.24126708200535577,
589
+ "brier": 0.1377459423822855,
590
+ "jsd": 0.06474821970285918,
591
+ "acc": 0.9123434704830053,
592
+ "acc_k": 0.9123434704830053,
593
+ "ece": 0.03644280175616786,
594
+ "sel_acc80": 0.9553571428571429,
595
+ "coverage": 1.0,
596
+ "false_abstain": 0.0,
597
+ "null_recall": "n/a",
598
+ "n": 559
599
+ },
600
+ "choice": {
601
+ "nll": 0.2314097084447192,
602
+ "brier": 0.12436533034630996,
603
+ "jsd": 0.06006621263920528,
604
+ "acc": 0.9159212880143113,
605
+ "acc_k": 0.9119850187265918,
606
+ "ece": 0.027254086807503464,
607
+ "sel_acc80": 0.9776785714285714,
608
+ "conf_wrong": 0.0,
609
+ "coverage": 0.9552772808586762,
610
+ "false_abstain": 0.0,
611
+ "null_recall": 1.0,
612
+ "auroc_null": 1.0,
613
+ "auroc_conf": 1.0,
614
+ "n": 559
615
+ }
616
+ },
617
+ "by_label_class": {
618
+ "answer": {
619
+ "nll": 0.3895064652358634,
620
+ "brier": 0.21807185170591808,
621
+ "jsd": 0.1111728077141086,
622
+ "acc": 0.8564294631710362,
623
+ "acc_k": 0.8564294631710362,
624
+ "ece": 0.07561033562328037,
625
+ "sel_acc80": 0.9321372854914196,
626
+ "coverage": 1.0,
627
+ "false_abstain": 0.0,
628
+ "null_recall": "n/a",
629
+ "n": 1602
630
+ },
631
+ "null": {
632
+ "nll": 0.0019389944782613345,
633
+ "brier": 3.5584650339346533e-05,
634
+ "jsd": 0.0006709503432798227,
635
+ "acc": 1.0,
636
+ "ece": 0.0019280886650085982,
637
+ "sel_acc80": 1.0,
638
+ "conf_wrong": 0.0,
639
+ "coverage": 0.0,
640
+ "false_abstain": "n/a",
641
+ "null_recall": 1.0,
642
+ "n": 25
643
+ }
644
+ },
645
+ "baselines": {
646
+ "majority_class": 1,
647
+ "majority_acc": 0.36447449293177625
648
+ }
649
+ },
650
+ "wf_rubric_flip": {
651
+ "n": 2062,
652
+ "n_total": 2062,
653
+ "suffix_overflow": 0,
654
+ "raw": {
655
+ "nll": 0.559445285535001,
656
+ "brier": 0.3272260198998571,
657
+ "jsd": 0.14724147310065305,
658
+ "acc": 0.7516973811833172,
659
+ "acc_k": 0.7516973811833172,
660
+ "ece": 0.0431027509250419,
661
+ "sel_acc80": 0.8351515151515152,
662
+ "coverage": 1.0,
663
+ "false_abstain": 0.0,
664
+ "null_recall": "n/a"
665
+ },
666
+ "by_family": {
667
+ "urgency": {
668
+ "nll": 1.00970911915973,
669
+ "brier": 0.5721670188295802,
670
+ "jsd": 0.276942053921465,
671
+ "acc": 0.5357142857142857,
672
+ "acc_k": 0.5357142857142857,
673
+ "ece": 0.04788954292024886,
674
+ "sel_acc80": 0.6,
675
+ "coverage": 1.0,
676
+ "false_abstain": 0.0,
677
+ "null_recall": "n/a",
678
+ "n": 700
679
+ },
680
+ "eligibility": {
681
+ "nll": 0.669296858496893,
682
+ "brier": 0.4143906038771635,
683
+ "jsd": 0.1633805475660813,
684
+ "acc": 0.7149390243902439,
685
+ "acc_k": 0.7149390243902439,
686
+ "ece": 0.10991631802625769,
687
+ "sel_acc80": 0.7561904761904762,
688
+ "coverage": 1.0,
689
+ "false_abstain": 0.0,
690
+ "null_recall": "n/a",
691
+ "n": 656
692
+ },
693
+ "tool_select": {
694
+ "nll": 0.010936340208780766,
695
+ "brier": 0.0033752177186684626,
696
+ "jsd": 0.0036470829818297796,
697
+ "acc": 1.0,
698
+ "acc_k": 1.0,
699
+ "ece": 0.009704953043724294,
700
+ "sel_acc80": 1.0,
701
+ "coverage": 1.0,
702
+ "false_abstain": 0.0,
703
+ "null_recall": "n/a",
704
+ "n": 706
705
+ }
706
+ },
707
+ "by_qtype": {
708
+ "score": {
709
+ "nll": 1.00970911915973,
710
+ "brier": 0.5721670188295802,
711
+ "jsd": 0.276942053921465,
712
+ "acc": 0.5357142857142857,
713
+ "acc_k": 0.5357142857142857,
714
+ "ece": 0.04788954292024886,
715
+ "sel_acc80": 0.6,
716
+ "coverage": 1.0,
717
+ "false_abstain": 0.0,
718
+ "null_recall": "n/a",
719
+ "n": 700
720
+ },
721
+ "noul": {
722
+ "nll": 0.669296858496893,
723
+ "brier": 0.4143906038771635,
724
+ "jsd": 0.1633805475660813,
725
+ "acc": 0.7149390243902439,
726
+ "acc_k": 0.7149390243902439,
727
+ "ece": 0.10991631802625769,
728
+ "sel_acc80": 0.7561904761904762,
729
+ "coverage": 1.0,
730
+ "false_abstain": 0.0,
731
+ "null_recall": "n/a",
732
+ "n": 656
733
+ },
734
+ "choice": {
735
+ "nll": 0.010936340208780766,
736
+ "brier": 0.0033752177186684626,
737
+ "jsd": 0.0036470829818297796,
738
+ "acc": 1.0,
739
+ "acc_k": 1.0,
740
+ "ece": 0.009704953043724294,
741
+ "sel_acc80": 1.0,
742
+ "coverage": 1.0,
743
+ "false_abstain": 0.0,
744
+ "null_recall": "n/a",
745
+ "n": 706
746
+ }
747
+ },
748
+ "by_label_class": {
749
+ "answer": {
750
+ "nll": 0.559445285535001,
751
+ "brier": 0.3272260198998571,
752
+ "jsd": 0.14724147310065305,
753
+ "acc": 0.7516973811833172,
754
+ "acc_k": 0.7516973811833172,
755
+ "ece": 0.0431027509250419,
756
+ "sel_acc80": 0.8351515151515152,
757
+ "coverage": 1.0,
758
+ "false_abstain": 0.0,
759
+ "null_recall": "n/a",
760
+ "n": 2062
761
+ }
762
+ },
763
+ "baselines": {
764
+ "majority_class": 0,
765
+ "majority_acc": 0.32104752667313285
766
+ },
767
+ "flip": {
768
+ "n_pairs": 1031,
769
+ "flip_rate": 0.7487875848690592,
770
+ "both_correct": 0.5809893307468477,
771
+ "acc": 0.7516973811833172
772
+ },
773
+ "flip_by_family": {
774
+ "eligibility": {
775
+ "n_pairs": 328,
776
+ "flip_rate": 0.5091463414634146,
777
+ "both_correct": 0.4695121951219512,
778
+ "acc": 0.7149390243902439
779
+ },
780
+ "tool_select": {
781
+ "n_pairs": 353,
782
+ "flip_rate": 1.0,
783
+ "both_correct": 1.0,
784
+ "acc": 1.0
785
+ },
786
+ "urgency": {
787
+ "n_pairs": 350,
788
+ "flip_rate": 0.72,
789
+ "both_correct": 0.26285714285714284,
790
+ "acc": 0.5357142857142857
791
+ }
792
+ }
793
+ },
794
+ "wf_rubric_shuffled": {
795
+ "n": 3775,
796
+ "n_total": 3775,
797
+ "suffix_overflow": 0,
798
+ "raw": {
799
+ "nll": 2.698697412970932,
800
+ "brier": 0.938708681237318,
801
+ "jsd": 0.3677264317143067,
802
+ "acc": 0.3748344370860927,
803
+ "acc_k": 0.37509933774834436,
804
+ "ece": 0.36605128677080795,
805
+ "sel_acc80": 0.39006622516556294,
806
+ "coverage": 0.9997350993377483,
807
+ "false_abstain": 0.00026490066225165563,
808
+ "null_recall": "n/a"
809
+ },
810
+ "by_family": {
811
+ "tool_select": {
812
+ "nll": 6.7661947710653765,
813
+ "brier": 1.5175450979035083,
814
+ "jsd": 0.5510631818062146,
815
+ "acc": 0.17016574585635358,
816
+ "acc_k": 0.17016574585635358,
817
+ "ece": 0.7154725696335841,
818
+ "sel_acc80": 0.143646408839779,
819
+ "coverage": 1.0,
820
+ "false_abstain": 0.0,
821
+ "null_recall": "n/a",
822
+ "n": 905
823
+ },
824
+ "eligibility": {
825
+ "nll": 1.2318526208591865,
826
+ "brier": 0.7084869470501419,
827
+ "jsd": 0.2544289537720909,
828
+ "acc": 0.5304835318850736,
829
+ "acc_k": 0.5304835318850736,
830
+ "ece": 0.31344534871289176,
831
+ "sel_acc80": 0.5385288966725044,
832
+ "coverage": 1.0,
833
+ "false_abstain": 0.0,
834
+ "null_recall": "n/a",
835
+ "n": 1427
836
+ },
837
+ "urgency": {
838
+ "nll": 1.5982832821795177,
839
+ "brier": 0.80335140999837,
840
+ "jsd": 0.3647851581109563,
841
+ "acc": 0.3492723492723493,
842
+ "acc_k": 0.34996534996534995,
843
+ "ece": 0.2045487121873156,
844
+ "sel_acc80": 0.3575757575757576,
845
+ "coverage": 0.9993069993069993,
846
+ "false_abstain": 0.000693000693000693,
847
+ "null_recall": "n/a",
848
+ "n": 1443
849
+ }
850
+ },
851
+ "by_qtype": {
852
+ "choice": {
853
+ "nll": 6.7661947710653765,
854
+ "brier": 1.5175450979035083,
855
+ "jsd": 0.5510631818062146,
856
+ "acc": 0.17016574585635358,
857
+ "acc_k": 0.17016574585635358,
858
+ "ece": 0.7154725696335841,
859
+ "sel_acc80": 0.143646408839779,
860
+ "coverage": 1.0,
861
+ "false_abstain": 0.0,
862
+ "null_recall": "n/a",
863
+ "n": 905
864
+ },
865
+ "noul": {
866
+ "nll": 1.2318526208591865,
867
+ "brier": 0.7084869470501419,
868
+ "jsd": 0.2544289537720909,
869
+ "acc": 0.5304835318850736,
870
+ "acc_k": 0.5304835318850736,
871
+ "ece": 0.31344534871289176,
872
+ "sel_acc80": 0.5385288966725044,
873
+ "coverage": 1.0,
874
+ "false_abstain": 0.0,
875
+ "null_recall": "n/a",
876
+ "n": 1427
877
+ },
878
+ "score": {
879
+ "nll": 1.5982832821795177,
880
+ "brier": 0.80335140999837,
881
+ "jsd": 0.3647851581109563,
882
+ "acc": 0.3492723492723493,
883
+ "acc_k": 0.34996534996534995,
884
+ "ece": 0.2045487121873156,
885
+ "sel_acc80": 0.3575757575757576,
886
+ "coverage": 0.9993069993069993,
887
+ "false_abstain": 0.000693000693000693,
888
+ "null_recall": "n/a",
889
+ "n": 1443
890
+ }
891
+ },
892
+ "by_label_class": {
893
+ "answer": {
894
+ "nll": 2.698697412970932,
895
+ "brier": 0.938708681237318,
896
+ "jsd": 0.3677264317143067,
897
+ "acc": 0.3748344370860927,
898
+ "acc_k": 0.37509933774834436,
899
+ "ece": 0.36605128677080795,
900
+ "sel_acc80": 0.39006622516556294,
901
+ "coverage": 0.9997350993377483,
902
+ "false_abstain": 0.00026490066225165563,
903
+ "null_recall": "n/a",
904
+ "n": 3775
905
+ }
906
+ },
907
+ "baselines": {
908
+ "majority_class": 0,
909
+ "majority_acc": 0.36
910
+ }
911
+ },
912
+ "cua_s1_forms_demo": {
913
+ "n": 196,
914
+ "n_total": 196,
915
+ "suffix_overflow": 0,
916
+ "raw": {
917
+ "nll": 0.0009298304114405211,
918
+ "brier": 1.7812480371703415e-05,
919
+ "jsd": 0.00032176846444351295,
920
+ "acc": 1.0,
921
+ "acc_k": 1.0,
922
+ "ece": 0.000924828101177555,
923
+ "sel_acc80": 1.0,
924
+ "coverage": 1.0,
925
+ "false_abstain": 0.0,
926
+ "null_recall": "n/a"
927
+ },
928
+ "by_family": {
929
+ "cua_s1_forms": {
930
+ "nll": 0.0009298304114405211,
931
+ "brier": 1.7812480371703415e-05,
932
+ "jsd": 0.00032176846444351295,
933
+ "acc": 1.0,
934
+ "acc_k": 1.0,
935
+ "ece": 0.000924828101177555,
936
+ "sel_acc80": 1.0,
937
+ "coverage": 1.0,
938
+ "false_abstain": 0.0,
939
+ "null_recall": "n/a",
940
+ "n": 196
941
+ }
942
+ },
943
+ "by_qtype": {
944
+ "choice": {
945
+ "nll": 0.0009298304114405211,
946
+ "brier": 1.7812480371703415e-05,
947
+ "jsd": 0.00032176846444351295,
948
+ "acc": 1.0,
949
+ "acc_k": 1.0,
950
+ "ece": 0.000924828101177555,
951
+ "sel_acc80": 1.0,
952
+ "coverage": 1.0,
953
+ "false_abstain": 0.0,
954
+ "null_recall": "n/a",
955
+ "n": 196
956
+ }
957
+ },
958
+ "by_label_class": {
959
+ "answer": {
960
+ "nll": 0.0009298304114405211,
961
+ "brier": 1.7812480371703415e-05,
962
+ "jsd": 0.00032176846444351295,
963
+ "acc": 1.0,
964
+ "acc_k": 1.0,
965
+ "ece": 0.000924828101177555,
966
+ "sel_acc80": 1.0,
967
+ "coverage": 1.0,
968
+ "false_abstain": 0.0,
969
+ "null_recall": "n/a",
970
+ "n": 196
971
+ }
972
+ },
973
+ "baselines": {
974
+ "majority_class": 18,
975
+ "majority_acc": 0.27040816326530615
976
+ }
977
+ },
978
+ "cua_s1_forms_test": {
979
+ "n": 24370,
980
+ "n_total": 24370,
981
+ "suffix_overflow": 0,
982
+ "raw": {
983
+ "nll": 0.031139153702024643,
984
+ "brier": 0.015968466328798284,
985
+ "jsd": 0.006616339278338574,
986
+ "acc": 0.9897004513746409,
987
+ "acc_k": 0.9897004513746409,
988
+ "ece": 0.0040463498602845566,
989
+ "sel_acc80": 1.0,
990
+ "coverage": 1.0,
991
+ "false_abstain": 0.0,
992
+ "null_recall": "n/a"
993
+ },
994
+ "by_family": {
995
+ "cua_s1_forms": {
996
+ "nll": 0.031139153702024643,
997
+ "brier": 0.015968466328798284,
998
+ "jsd": 0.006616339278338574,
999
+ "acc": 0.9897004513746409,
1000
+ "acc_k": 0.9897004513746409,
1001
+ "ece": 0.0040463498602845566,
1002
+ "sel_acc80": 1.0,
1003
+ "coverage": 1.0,
1004
+ "false_abstain": 0.0,
1005
+ "null_recall": "n/a",
1006
+ "n": 24370
1007
+ }
1008
+ },
1009
+ "by_qtype": {
1010
+ "choice": {
1011
+ "nll": 0.031139153702024643,
1012
+ "brier": 0.015968466328798284,
1013
+ "jsd": 0.006616339278338574,
1014
+ "acc": 0.9897004513746409,
1015
+ "acc_k": 0.9897004513746409,
1016
+ "ece": 0.0040463498602845566,
1017
+ "sel_acc80": 1.0,
1018
+ "coverage": 1.0,
1019
+ "false_abstain": 0.0,
1020
+ "null_recall": "n/a",
1021
+ "n": 24370
1022
+ }
1023
+ },
1024
+ "by_label_class": {
1025
+ "answer": {
1026
+ "nll": 0.031139153702024643,
1027
+ "brier": 0.015968466328798284,
1028
+ "jsd": 0.006616339278338574,
1029
+ "acc": 0.9897004513746409,
1030
+ "acc_k": 0.9897004513746409,
1031
+ "ece": 0.0040463498602845566,
1032
+ "sel_acc80": 1.0,
1033
+ "coverage": 1.0,
1034
+ "false_abstain": 0.0,
1035
+ "null_recall": "n/a",
1036
+ "n": 24370
1037
+ }
1038
+ },
1039
+ "baselines": {
1040
+ "majority_class": 16,
1041
+ "majority_acc": 0.07033237587197373
1042
+ }
1043
+ },
1044
+ "jev4b_adversarial": {
1045
+ "n": 600,
1046
+ "n_total": 600,
1047
+ "suffix_overflow": 0,
1048
+ "raw": {
1049
+ "nll": 0.27162374405442014,
1050
+ "brier": 0.13008828636077177,
1051
+ "jsd": 0.06711174695091494,
1052
+ "acc": 0.935,
1053
+ "acc_k": 0.935,
1054
+ "ece": 0.06803944091002141,
1055
+ "sel_acc80": 0.9645833333333333,
1056
+ "coverage": 1.0,
1057
+ "false_abstain": 0.0,
1058
+ "null_recall": "n/a"
1059
+ },
1060
+ "by_family": {
1061
+ "jev4b_distill": {
1062
+ "nll": 0.27162374405442014,
1063
+ "brier": 0.13008828636077177,
1064
+ "jsd": 0.06711174695091494,
1065
+ "acc": 0.935,
1066
+ "acc_k": 0.935,
1067
+ "ece": 0.06803944091002141,
1068
+ "sel_acc80": 0.9645833333333333,
1069
+ "coverage": 1.0,
1070
+ "false_abstain": 0.0,
1071
+ "null_recall": "n/a",
1072
+ "n": 600
1073
+ }
1074
+ },
1075
+ "by_qtype": {
1076
+ "choice": {
1077
+ "nll": 0.2368204577035528,
1078
+ "brier": 0.09787175781239878,
1079
+ "jsd": 0.045669968907413504,
1080
+ "acc": 0.9555555555555556,
1081
+ "acc_k": 0.9555555555555556,
1082
+ "ece": 0.05371427064140639,
1083
+ "sel_acc80": 0.96875,
1084
+ "coverage": 1.0,
1085
+ "false_abstain": 0.0,
1086
+ "null_recall": "n/a",
1087
+ "n": 360
1088
+ },
1089
+ "score": {
1090
+ "nll": 0.5946715993673614,
1091
+ "brier": 0.3368385842307513,
1092
+ "jsd": 0.1811635778371567,
1093
+ "acc": 0.8166666666666667,
1094
+ "acc_k": 0.8166666666666667,
1095
+ "ece": 0.19491817976037662,
1096
+ "sel_acc80": 0.8854166666666666,
1097
+ "coverage": 1.0,
1098
+ "false_abstain": 0.0,
1099
+ "null_recall": "n/a",
1100
+ "n": 120
1101
+ },
1102
+ "noul": {
1103
+ "nll": 0.05298574779408091,
1104
+ "brier": 0.019987574135911266,
1105
+ "jsd": 0.01738525019517747,
1106
+ "acc": 0.9916666666666667,
1107
+ "acc_k": 0.9916666666666667,
1108
+ "ece": 0.033880017697811134,
1109
+ "sel_acc80": 1.0,
1110
+ "coverage": 1.0,
1111
+ "false_abstain": 0.0,
1112
+ "null_recall": "n/a",
1113
+ "n": 120
1114
+ }
1115
+ },
1116
+ "by_label_class": {
1117
+ "answer": {
1118
+ "nll": 0.27162374405442014,
1119
+ "brier": 0.13008828636077177,
1120
+ "jsd": 0.06711174695091494,
1121
+ "acc": 0.935,
1122
+ "acc_k": 0.935,
1123
+ "ece": 0.06803944091002141,
1124
+ "sel_acc80": 0.9645833333333333,
1125
+ "coverage": 1.0,
1126
+ "false_abstain": 0.0,
1127
+ "null_recall": "n/a",
1128
+ "n": 600
1129
+ }
1130
+ },
1131
+ "baselines": {
1132
+ "majority_class": 0,
1133
+ "majority_acc": 0.42833333333333334
1134
+ }
1135
+ },
1136
+ "jevlogs_triage": {
1137
+ "n": 5080,
1138
+ "n_total": 5080,
1139
+ "suffix_overflow": 0,
1140
+ "raw": {
1141
+ "nll": 0.6941952121477953,
1142
+ "brier": 0.4479620663111127,
1143
+ "jsd": 0.1644331696428951,
1144
+ "acc": 0.7104330708661417,
1145
+ "acc_k": 0.7104330708661417,
1146
+ "ece": 0.20131792304084056,
1147
+ "sel_acc80": 0.8058562992125984,
1148
+ "coverage": 1.0,
1149
+ "false_abstain": 0.0,
1150
+ "null_recall": "n/a"
1151
+ },
1152
+ "by_family": {
1153
+ "log_hdfs": {
1154
+ "nll": 0.8571451724027355,
1155
+ "brier": 0.5156481325333623,
1156
+ "jsd": 0.18672513520111952,
1157
+ "acc": 0.7,
1158
+ "acc_k": 0.7,
1159
+ "ece": 0.21816099820137022,
1160
+ "sel_acc80": 0.716,
1161
+ "coverage": 1.0,
1162
+ "false_abstain": 0.0,
1163
+ "null_recall": "n/a",
1164
+ "n": 2500
1165
+ },
1166
+ "log_bgl": {
1167
+ "nll": 0.519794702051961,
1168
+ "brier": 0.3722576889026077,
1169
+ "jsd": 0.1394862307236892,
1170
+ "acc": 0.7256,
1171
+ "acc_k": 0.7256,
1172
+ "ece": 0.22598801379203798,
1173
+ "sel_acc80": 0.793,
1174
+ "coverage": 1.0,
1175
+ "false_abstain": 0.0,
1176
+ "null_recall": "n/a",
1177
+ "n": 2500
1178
+ },
1179
+ "log_adversarial": {
1180
+ "nll": 1.052024894675745,
1181
+ "brier": 0.6985342908815958,
1182
+ "jsd": 0.24740108717356427,
1183
+ "acc": 0.5625,
1184
+ "acc_k": 0.5625,
1185
+ "ece": 0.3473062872886658,
1186
+ "sel_acc80": 0.53125,
1187
+ "coverage": 1.0,
1188
+ "false_abstain": 0.0,
1189
+ "null_recall": "n/a",
1190
+ "n": 80
1191
+ }
1192
+ },
1193
+ "by_qtype": {
1194
+ "noul": {
1195
+ "nll": 0.6941952121477953,
1196
+ "brier": 0.4479620663111127,
1197
+ "jsd": 0.1644331696428951,
1198
+ "acc": 0.7104330708661417,
1199
+ "acc_k": 0.7104330708661417,
1200
+ "ece": 0.20131792304084056,
1201
+ "sel_acc80": 0.8058562992125984,
1202
+ "coverage": 1.0,
1203
+ "false_abstain": 0.0,
1204
+ "null_recall": "n/a",
1205
+ "n": 5080
1206
+ }
1207
+ },
1208
+ "by_label_class": {
1209
+ "answer": {
1210
+ "nll": 0.6941952121477953,
1211
+ "brier": 0.4479620663111127,
1212
+ "jsd": 0.1644331696428951,
1213
+ "acc": 0.7104330708661417,
1214
+ "acc_k": 0.7104330708661417,
1215
+ "ece": 0.20131792304084056,
1216
+ "sel_acc80": 0.8058562992125984,
1217
+ "coverage": 1.0,
1218
+ "false_abstain": 0.0,
1219
+ "null_recall": "n/a",
1220
+ "n": 5080
1221
+ }
1222
+ },
1223
+ "baselines": {
1224
+ "majority_class": 0,
1225
+ "majority_acc": 0.6968503937007874,
1226
+ "always_positive_acc": 0.3031496062992126
1227
+ }
1228
+ },
1229
+ "mind2web_choice": {
1230
+ "n": 1600,
1231
+ "n_total": 1600,
1232
+ "suffix_overflow": 0,
1233
+ "raw": {
1234
+ "nll": 1.5136479569939831,
1235
+ "brier": 0.7428329076657673,
1236
+ "jsd": 0.34671858092870367,
1237
+ "acc": 0.3575,
1238
+ "acc_k": 0.36,
1239
+ "ece": 0.20080654334276918,
1240
+ "sel_acc80": 0.38828125,
1241
+ "coverage": 0.995625,
1242
+ "false_abstain": 0.004375,
1243
+ "null_recall": "n/a"
1244
+ },
1245
+ "by_family": {
1246
+ "mind2web": {
1247
+ "nll": 1.5136479569939831,
1248
+ "brier": 0.7428329076657673,
1249
+ "jsd": 0.34671858092870367,
1250
+ "acc": 0.3575,
1251
+ "acc_k": 0.36,
1252
+ "ece": 0.20080654334276918,
1253
+ "sel_acc80": 0.38828125,
1254
+ "coverage": 0.995625,
1255
+ "false_abstain": 0.004375,
1256
+ "null_recall": "n/a",
1257
+ "n": 1600
1258
+ }
1259
+ },
1260
+ "by_qtype": {
1261
+ "choice": {
1262
+ "nll": 2.0931226888545518,
1263
+ "brier": 0.7784709109053702,
1264
+ "jsd": 0.42772685578683883,
1265
+ "acc": 0.37,
1266
+ "acc_k": 0.375,
1267
+ "ece": 0.06646796382963659,
1268
+ "sel_acc80": 0.4109375,
1269
+ "coverage": 0.99125,
1270
+ "false_abstain": 0.00875,
1271
+ "null_recall": "n/a",
1272
+ "n": 800
1273
+ },
1274
+ "noul": {
1275
+ "nll": 0.9341732251334145,
1276
+ "brier": 0.7071949044261643,
1277
+ "jsd": 0.2657103060705684,
1278
+ "acc": 0.345,
1279
+ "acc_k": 0.345,
1280
+ "ece": 0.33514512285590164,
1281
+ "sel_acc80": 0.3234375,
1282
+ "coverage": 1.0,
1283
+ "false_abstain": 0.0,
1284
+ "null_recall": "n/a",
1285
+ "n": 800
1286
+ }
1287
+ },
1288
+ "by_label_class": {
1289
+ "answer": {
1290
+ "nll": 1.5136479569939831,
1291
+ "brier": 0.7428329076657673,
1292
+ "jsd": 0.34671858092870367,
1293
+ "acc": 0.3575,
1294
+ "acc_k": 0.36,
1295
+ "ece": 0.20080654334276918,
1296
+ "sel_acc80": 0.38828125,
1297
+ "coverage": 0.995625,
1298
+ "false_abstain": 0.004375,
1299
+ "null_recall": "n/a",
1300
+ "n": 1600
1301
+ }
1302
+ },
1303
+ "baselines": {
1304
+ "majority_class": 0,
1305
+ "majority_acc": 0.426875
1306
+ }
1307
+ },
1308
+ "pagerduty_trigger": {
1309
+ "n": 6000,
1310
+ "n_total": 6000,
1311
+ "suffix_overflow": 0,
1312
+ "raw": {
1313
+ "nll": 0.45057378800764875,
1314
+ "brier": 0.2645510776299787,
1315
+ "jsd": 0.12898428946204524,
1316
+ "acc": 0.8171666666666667,
1317
+ "acc_k": 0.8171666666666667,
1318
+ "ece": 0.08175959949195385,
1319
+ "sel_acc80": 0.8864583333333333,
1320
+ "coverage": 1.0,
1321
+ "false_abstain": 0.0,
1322
+ "null_recall": "n/a"
1323
+ },
1324
+ "by_family": {
1325
+ "replica_lag_data_risk": {
1326
+ "nll": 0.6898488965338108,
1327
+ "brier": 0.4642770455099506,
1328
+ "jsd": 0.21330152188587068,
1329
+ "acc": 0.5789473684210527,
1330
+ "acc_k": 0.5789473684210527,
1331
+ "ece": 0.1760190186793344,
1332
+ "sel_acc80": 0.5869565217391305,
1333
+ "coverage": 1.0,
1334
+ "false_abstain": 0.0,
1335
+ "null_recall": "n/a",
1336
+ "n": 114
1337
+ },
1338
+ "tls_hours_left_serving": {
1339
+ "nll": 0.9503486029544026,
1340
+ "brier": 0.705578046531571,
1341
+ "jsd": 0.24360104904451949,
1342
+ "acc": 0.5,
1343
+ "acc_k": 0.5,
1344
+ "ece": 0.4599647521972656,
1345
+ "sel_acc80": 0.620253164556962,
1346
+ "coverage": 1.0,
1347
+ "false_abstain": 0.0,
1348
+ "null_recall": "n/a",
1349
+ "n": 98
1350
+ },
1351
+ "disk_full_writes_failing": {
1352
+ "nll": 0.9617892499408918,
1353
+ "brier": 0.7153070589979267,
1354
+ "jsd": 0.24323647291422343,
1355
+ "acc": 0.5,
1356
+ "acc_k": 0.5,
1357
+ "ece": 0.4525816380977631,
1358
+ "sel_acc80": 0.625,
1359
+ "coverage": 1.0,
1360
+ "false_abstain": 0.0,
1361
+ "null_recall": "n/a",
1362
+ "n": 110
1363
+ },
1364
+ "revoked_key_used": {
1365
+ "nll": 1.6670805035219922,
1366
+ "brier": 1.0476412905344457,
1367
+ "jsd": 0.38638081928023327,
1368
+ "acc": 0.5,
1369
+ "acc_k": 0.5,
1370
+ "ece": 0.7080944776535034,
1371
+ "sel_acc80": 0.3780487804878049,
1372
+ "coverage": 1.0,
1373
+ "false_abstain": 0.0,
1374
+ "null_recall": "n/a",
1375
+ "n": 102
1376
+ },
1377
+ "checksum_corruption": {
1378
+ "nll": 0.5221658499392701,
1379
+ "brier": 0.3021209402517869,
1380
+ "jsd": 0.1668244358824877,
1381
+ "acc": 1.0,
1382
+ "acc_k": 1.0,
1383
+ "ece": 0.4045400937398274,
1384
+ "sel_acc80": 1.0,
1385
+ "coverage": 1.0,
1386
+ "false_abstain": 0.0,
1387
+ "null_recall": "n/a",
1388
+ "n": 90
1389
+ },
1390
+ "checkout_slo_burn": {
1391
+ "nll": 0.6211973900784526,
1392
+ "brier": 0.47606719857859514,
1393
+ "jsd": 0.1802507371509747,
1394
+ "acc": 0.5,
1395
+ "acc_k": 0.5,
1396
+ "ece": 0.38283807039260864,
1397
+ "sel_acc80": 0.625,
1398
+ "coverage": 1.0,
1399
+ "false_abstain": 0.0,
1400
+ "null_recall": "n/a",
1401
+ "n": 100
1402
+ },
1403
+ "payments_processor_down": {
1404
+ "nll": 0.9761891304656726,
1405
+ "brier": 0.7124004465279202,
1406
+ "jsd": 0.25978390316724703,
1407
+ "acc": 0.5,
1408
+ "acc_k": 0.5,
1409
+ "ece": 0.3435124742223862,
1410
+ "sel_acc80": 0.3815789473684211,
1411
+ "coverage": 1.0,
1412
+ "false_abstain": 0.0,
1413
+ "null_recall": "n/a",
1414
+ "n": 94
1415
+ },
1416
+ "crash_loop_user_impact": {
1417
+ "nll": 0.9124509739053913,
1418
+ "brier": 0.6743519260962172,
1419
+ "jsd": 0.2428845282176394,
1420
+ "acc": 0.5,
1421
+ "acc_k": 0.5,
1422
+ "ece": 0.3444297431657712,
1423
+ "sel_acc80": 0.6233766233766234,
1424
+ "coverage": 1.0,
1425
+ "false_abstain": 0.0,
1426
+ "null_recall": "n/a",
1427
+ "n": 96
1428
+ },
1429
+ "credential_stuffing": {
1430
+ "nll": 0.9519934639477086,
1431
+ "brier": 0.6977005836612166,
1432
+ "jsd": 0.2499338675952664,
1433
+ "acc": 0.5,
1434
+ "acc_k": 0.5,
1435
+ "ece": 0.33941899364193284,
1436
+ "sel_acc80": 0.6233766233766234,
1437
+ "coverage": 1.0,
1438
+ "false_abstain": 0.0,
1439
+ "null_recall": "n/a",
1440
+ "n": 96
1441
+ },
1442
+ "queue_backup_checkout": {
1443
+ "nll": 0.9013995609276422,
1444
+ "brier": 0.6699464517470315,
1445
+ "jsd": 0.24000688623489375,
1446
+ "acc": 0.5,
1447
+ "acc_k": 0.5,
1448
+ "ece": 0.47097641229629517,
1449
+ "sel_acc80": 0.625,
1450
+ "coverage": 1.0,
1451
+ "false_abstain": 0.0,
1452
+ "null_recall": "n/a",
1453
+ "n": 100
1454
+ },
1455
+ "queue_slow_growth": {
1456
+ "nll": 1.3707216103698314,
1457
+ "brier": 0.658629060210223,
1458
+ "jsd": 0.301554799426009,
1459
+ "acc": 0.5,
1460
+ "acc_k": 0.5,
1461
+ "ece": 0.2951798737049103,
1462
+ "sel_acc80": 0.6222222222222222,
1463
+ "coverage": 1.0,
1464
+ "false_abstain": 0.0,
1465
+ "null_recall": "n/a",
1466
+ "n": 56
1467
+ },
1468
+ "cert_30d": {
1469
+ "nll": 0.1727142536714221,
1470
+ "brier": 0.059320063795585645,
1471
+ "jsd": 0.05738444318715433,
1472
+ "acc": 1.0,
1473
+ "acc_k": 1.0,
1474
+ "ece": 0.1501917839050293,
1475
+ "sel_acc80": 1.0,
1476
+ "coverage": 1.0,
1477
+ "false_abstain": 0.0,
1478
+ "null_recall": "n/a",
1479
+ "n": 58
1480
+ },
1481
+ "disk_80pct_48h": {
1482
+ "nll": 0.9509672822978512,
1483
+ "brier": 0.5419380620634318,
1484
+ "jsd": 0.24201387380805145,
1485
+ "acc": 0.5,
1486
+ "acc_k": 0.5,
1487
+ "ece": 0.28837817162275314,
1488
+ "sel_acc80": 0.6153846153846154,
1489
+ "coverage": 1.0,
1490
+ "false_abstain": 0.0,
1491
+ "null_recall": "n/a",
1492
+ "n": 64
1493
+ },
1494
+ "cpu_high_no_impact": {
1495
+ "nll": 0.8010921147001497,
1496
+ "brier": 0.46771347158423976,
1497
+ "jsd": 0.21917217540903014,
1498
+ "acc": 0.5,
1499
+ "acc_k": 0.5,
1500
+ "ece": 0.28503292202949526,
1501
+ "sel_acc80": 0.625,
1502
+ "coverage": 1.0,
1503
+ "false_abstain": 0.0,
1504
+ "null_recall": "n/a",
1505
+ "n": 50
1506
+ },
1507
+ "single_pod_oom": {
1508
+ "nll": 1.314429901386395,
1509
+ "brier": 0.7101412733644461,
1510
+ "jsd": 0.28754181367618376,
1511
+ "acc": 0.5,
1512
+ "acc_k": 0.5,
1513
+ "ece": 0.38835878555591286,
1514
+ "sel_acc80": 0.6190476190476191,
1515
+ "coverage": 1.0,
1516
+ "false_abstain": 0.0,
1517
+ "null_recall": "n/a",
1518
+ "n": 52
1519
+ },
1520
+ "deprecation": {
1521
+ "nll": 1.162277843746267,
1522
+ "brier": 0.7164361718413621,
1523
+ "jsd": 0.2679758230576417,
1524
+ "acc": 0.5,
1525
+ "acc_k": 0.5,
1526
+ "ece": 0.4070015847682953,
1527
+ "sel_acc80": 0.625,
1528
+ "coverage": 1.0,
1529
+ "false_abstain": 0.0,
1530
+ "null_recall": "n/a",
1531
+ "n": 60
1532
+ },
1533
+ "canary_within_budget": {
1534
+ "nll": 0.997527949570193,
1535
+ "brier": 0.6500690785390699,
1536
+ "jsd": 0.24742592715354586,
1537
+ "acc": 0.5,
1538
+ "acc_k": 0.5,
1539
+ "ece": 0.39401811361312866,
1540
+ "sel_acc80": 0.6153846153846154,
1541
+ "coverage": 1.0,
1542
+ "false_abstain": 0.0,
1543
+ "null_recall": "n/a",
1544
+ "n": 48
1545
+ },
1546
+ "replica_lag_seconds": {
1547
+ "nll": 0.9901761862463576,
1548
+ "brier": 0.5515541514998862,
1549
+ "jsd": 0.2506408336180253,
1550
+ "acc": 0.5,
1551
+ "acc_k": 0.5,
1552
+ "ece": 0.2882567197084427,
1553
+ "sel_acc80": 0.625,
1554
+ "coverage": 1.0,
1555
+ "false_abstain": 0.0,
1556
+ "null_recall": "n/a",
1557
+ "n": 50
1558
+ },
1559
+ "tls_expires_7d": {
1560
+ "nll": 1.1567732954045722,
1561
+ "brier": 0.7294461207505658,
1562
+ "jsd": 0.2665059357272688,
1563
+ "acc": 0.5,
1564
+ "acc_k": 0.5,
1565
+ "ece": 0.41363009810447693,
1566
+ "sel_acc80": 0.6190476190476191,
1567
+ "coverage": 1.0,
1568
+ "false_abstain": 0.0,
1569
+ "null_recall": "n/a",
1570
+ "n": 26
1571
+ },
1572
+ "slow_slo_burn": {
1573
+ "nll": 0.6522867336147589,
1574
+ "brier": 0.5114718599657583,
1575
+ "jsd": 0.18362021573543594,
1576
+ "acc": 0.5,
1577
+ "acc_k": 0.5,
1578
+ "ece": 0.36724624037742615,
1579
+ "sel_acc80": 0.6206896551724138,
1580
+ "coverage": 1.0,
1581
+ "false_abstain": 0.0,
1582
+ "null_recall": "n/a",
1583
+ "n": 36
1584
+ },
1585
+ "injection_noise": {
1586
+ "nll": 0.5037106019010942,
1587
+ "brier": 0.37437867309218403,
1588
+ "jsd": 0.151060977826625,
1589
+ "acc": 0.5,
1590
+ "acc_k": 0.5,
1591
+ "ece": 0.32909492561894077,
1592
+ "sel_acc80": 0.625,
1593
+ "coverage": 1.0,
1594
+ "false_abstain": 0.0,
1595
+ "null_recall": "n/a",
1596
+ "n": 310
1597
+ },
1598
+ "health_ok": {
1599
+ "nll": 0.18152995080281234,
1600
+ "brier": 0.047734969981545416,
1601
+ "jsd": 0.061126679924072204,
1602
+ "acc": 1.0,
1603
+ "acc_k": 1.0,
1604
+ "ece": 0.16462426345332776,
1605
+ "sel_acc80": 1.0,
1606
+ "coverage": 1.0,
1607
+ "false_abstain": 0.0,
1608
+ "null_recall": "n/a",
1609
+ "n": 306
1610
+ },
1611
+ "kafka_caught_up": {
1612
+ "nll": 0.5039227087644147,
1613
+ "brier": 0.3316987967829466,
1614
+ "jsd": 0.151070396937696,
1615
+ "acc": 0.6654676258992805,
1616
+ "acc_k": 0.6654676258992805,
1617
+ "ece": 0.21027986587380335,
1618
+ "sel_acc80": 0.672645739910314,
1619
+ "coverage": 1.0,
1620
+ "false_abstain": 0.0,
1621
+ "null_recall": "n/a",
1622
+ "n": 278
1623
+ },
1624
+ "invalid_coupon_error": {
1625
+ "nll": 0.25784577121054475,
1626
+ "brier": 0.1019467076656268,
1627
+ "jsd": 0.0844192917311665,
1628
+ "acc": 1.0,
1629
+ "acc_k": 1.0,
1630
+ "ece": 0.21481417610038142,
1631
+ "sel_acc80": 1.0,
1632
+ "coverage": 1.0,
1633
+ "false_abstain": 0.0,
1634
+ "null_recall": "n/a",
1635
+ "n": 322
1636
+ },
1637
+ "stack_trace_handled": {
1638
+ "nll": 0.30060535025925805,
1639
+ "brier": 0.13121770374213282,
1640
+ "jsd": 0.09815941120194528,
1641
+ "acc": 1.0,
1642
+ "acc_k": 1.0,
1643
+ "ece": 0.24862703573272885,
1644
+ "sel_acc80": 1.0,
1645
+ "coverage": 1.0,
1646
+ "false_abstain": 0.0,
1647
+ "null_recall": "n/a",
1648
+ "n": 332
1649
+ },
1650
+ "retry_then_ok": {
1651
+ "nll": 0.2279076528120555,
1652
+ "brier": 0.08093679155813602,
1653
+ "jsd": 0.0753719006565446,
1654
+ "acc": 1.0,
1655
+ "acc_k": 1.0,
1656
+ "ece": 0.1956012144055165,
1657
+ "sel_acc80": 1.0,
1658
+ "coverage": 1.0,
1659
+ "false_abstain": 0.0,
1660
+ "null_recall": "n/a",
1661
+ "n": 284
1662
+ },
1663
+ "feature_flag": {
1664
+ "nll": 0.13855516551397323,
1665
+ "brier": 0.02981594098678688,
1666
+ "jsd": 0.04691808660224193,
1667
+ "acc": 1.0,
1668
+ "acc_k": 1.0,
1669
+ "ece": 0.12797431425845368,
1670
+ "sel_acc80": 1.0,
1671
+ "coverage": 1.0,
1672
+ "false_abstain": 0.0,
1673
+ "null_recall": "n/a",
1674
+ "n": 282
1675
+ },
1676
+ "cache_hit": {
1677
+ "nll": 0.16067786504109507,
1678
+ "brier": 0.05175275250111869,
1679
+ "jsd": 0.05360337986685591,
1680
+ "acc": 1.0,
1681
+ "acc_k": 1.0,
1682
+ "ece": 0.14153279624618847,
1683
+ "sel_acc80": 1.0,
1684
+ "coverage": 1.0,
1685
+ "false_abstain": 0.0,
1686
+ "null_recall": "n/a",
1687
+ "n": 286
1688
+ },
1689
+ "metrics_scrape": {
1690
+ "nll": 0.5509275726093205,
1691
+ "brier": 0.37247962744689045,
1692
+ "jsd": 0.1658098683017907,
1693
+ "acc": 0.5,
1694
+ "acc_k": 0.5,
1695
+ "ece": 0.3461012159724062,
1696
+ "sel_acc80": 0.623574144486692,
1697
+ "coverage": 1.0,
1698
+ "false_abstain": 0.0,
1699
+ "null_recall": "n/a",
1700
+ "n": 328
1701
+ },
1702
+ "cron_ok": {
1703
+ "nll": 0.25100095482382306,
1704
+ "brier": 0.11987021871014797,
1705
+ "jsd": 0.08118105449703346,
1706
+ "acc": 1.0,
1707
+ "acc_k": 1.0,
1708
+ "ece": 0.20225984857928364,
1709
+ "sel_acc80": 1.0,
1710
+ "coverage": 1.0,
1711
+ "false_abstain": 0.0,
1712
+ "null_recall": "n/a",
1713
+ "n": 310
1714
+ },
1715
+ "rate_limit_bot": {
1716
+ "nll": 0.3944067554874347,
1717
+ "brier": 0.18381001382027407,
1718
+ "jsd": 0.12796267513555565,
1719
+ "acc": 1.0,
1720
+ "acc_k": 1.0,
1721
+ "ece": 0.31977638020180166,
1722
+ "sel_acc80": 1.0,
1723
+ "coverage": 1.0,
1724
+ "false_abstain": 0.0,
1725
+ "null_recall": "n/a",
1726
+ "n": 256
1727
+ },
1728
+ "http_404": {
1729
+ "nll": 0.11865384131473723,
1730
+ "brier": 0.023577944218449064,
1731
+ "jsd": 0.04024055657795258,
1732
+ "acc": 1.0,
1733
+ "acc_k": 1.0,
1734
+ "ece": 0.11015801009891918,
1735
+ "sel_acc80": 1.0,
1736
+ "coverage": 1.0,
1737
+ "false_abstain": 0.0,
1738
+ "null_recall": "n/a",
1739
+ "n": 318
1740
+ },
1741
+ "deploy_success": {
1742
+ "nll": 0.0665660530378659,
1743
+ "brier": 0.00769735271439915,
1744
+ "jsd": 0.022807886905452414,
1745
+ "acc": 1.0,
1746
+ "acc_k": 1.0,
1747
+ "ece": 0.06395390971762233,
1748
+ "sel_acc80": 1.0,
1749
+ "coverage": 1.0,
1750
+ "false_abstain": 0.0,
1751
+ "null_recall": "n/a",
1752
+ "n": 244
1753
+ },
1754
+ "user_not_found": {
1755
+ "nll": 0.4304388910037287,
1756
+ "brier": 0.24248037606659678,
1757
+ "jsd": 0.13315740549304775,
1758
+ "acc": 0.9387755102040817,
1759
+ "acc_k": 0.9387755102040817,
1760
+ "ece": 0.24541204691338703,
1761
+ "sel_acc80": 0.9915254237288136,
1762
+ "coverage": 1.0,
1763
+ "false_abstain": 0.0,
1764
+ "null_recall": "n/a",
1765
+ "n": 294
1766
+ },
1767
+ "request_ok": {
1768
+ "nll": 0.24205577862251662,
1769
+ "brier": 0.09208789387581466,
1770
+ "jsd": 0.08013562084859333,
1771
+ "acc": 1.0,
1772
+ "acc_k": 1.0,
1773
+ "ece": 0.20847708940505985,
1774
+ "sel_acc80": 1.0,
1775
+ "coverage": 1.0,
1776
+ "false_abstain": 0.0,
1777
+ "null_recall": "n/a",
1778
+ "n": 350
1779
+ }
1780
+ },
1781
+ "by_qtype": {
1782
+ "choice": {
1783
+ "nll": 0.5648469629367534,
1784
+ "brier": 0.3173471457305583,
1785
+ "jsd": 0.16576450460353345,
1786
+ "acc": 0.7796666666666666,
1787
+ "acc_k": 0.7796666666666666,
1788
+ "ece": 0.08924249602357544,
1789
+ "sel_acc80": 0.8258333333333333,
1790
+ "coverage": 1.0,
1791
+ "false_abstain": 0.0,
1792
+ "null_recall": "n/a",
1793
+ "n": 3000
1794
+ },
1795
+ "noul": {
1796
+ "nll": 0.336300613078544,
1797
+ "brier": 0.21175500952939919,
1798
+ "jsd": 0.092204074320557,
1799
+ "acc": 0.8546666666666667,
1800
+ "acc_k": 0.8546666666666667,
1801
+ "ece": 0.1531165913939476,
1802
+ "sel_acc80": 0.9783333333333334,
1803
+ "coverage": 1.0,
1804
+ "false_abstain": 0.0,
1805
+ "null_recall": "n/a",
1806
+ "n": 3000
1807
+ }
1808
+ },
1809
+ "by_label_class": {
1810
+ "answer": {
1811
+ "nll": 0.45057378800764875,
1812
+ "brier": 0.2645510776299787,
1813
+ "jsd": 0.12898428946204524,
1814
+ "acc": 0.8171666666666667,
1815
+ "acc_k": 0.8171666666666667,
1816
+ "ece": 0.08175959949195385,
1817
+ "sel_acc80": 0.8864583333333333,
1818
+ "coverage": 1.0,
1819
+ "false_abstain": 0.0,
1820
+ "null_recall": "n/a",
1821
+ "n": 6000
1822
+ }
1823
+ },
1824
+ "baselines": {
1825
+ "majority_class": 0,
1826
+ "majority_acc": 0.7916666666666666
1827
+ }
1828
+ },
1829
+ "systemone_lite_hard": {
1830
+ "n": 5400,
1831
+ "n_total": 5400,
1832
+ "suffix_overflow": 0,
1833
+ "raw": {
1834
+ "nll": 0.47733330766959625,
1835
+ "brier": 0.293311500019393,
1836
+ "jsd": 0.1385319090950344,
1837
+ "acc": 0.7833333333333333,
1838
+ "acc_k": 0.7837037037037037,
1839
+ "ece": 0.03959470673843665,
1840
+ "sel_acc80": 0.8453703703703703,
1841
+ "coverage": 0.9996296296296296,
1842
+ "false_abstain": 0.00037037037037037035,
1843
+ "null_recall": "n/a"
1844
+ },
1845
+ "by_family": {
1846
+ "systemone_lite": {
1847
+ "nll": 0.47733330766959625,
1848
+ "brier": 0.293311500019393,
1849
+ "jsd": 0.1385319090950344,
1850
+ "acc": 0.7833333333333333,
1851
+ "acc_k": 0.7837037037037037,
1852
+ "ece": 0.03959470673843665,
1853
+ "sel_acc80": 0.8453703703703703,
1854
+ "coverage": 0.9996296296296296,
1855
+ "false_abstain": 0.00037037037037037035,
1856
+ "null_recall": "n/a",
1857
+ "n": 5400
1858
+ }
1859
+ },
1860
+ "by_qtype": {
1861
+ "choice": {
1862
+ "nll": 0.24729003141422154,
1863
+ "brier": 0.1407608173373836,
1864
+ "jsd": 0.060118104806292254,
1865
+ "acc": 0.8994444444444445,
1866
+ "acc_k": 0.9005555555555556,
1867
+ "ece": 0.02308347546392016,
1868
+ "sel_acc80": 0.96875,
1869
+ "coverage": 0.9988888888888889,
1870
+ "false_abstain": 0.0011111111111111111,
1871
+ "null_recall": "n/a",
1872
+ "n": 1800
1873
+ },
1874
+ "noul": {
1875
+ "nll": 0.4667849208576605,
1876
+ "brier": 0.3108541755307743,
1877
+ "jsd": 0.1404008352752232,
1878
+ "acc": 0.7583333333333333,
1879
+ "acc_k": 0.7583333333333333,
1880
+ "ece": 0.026432335111829974,
1881
+ "sel_acc80": 0.8138888888888889,
1882
+ "coverage": 1.0,
1883
+ "false_abstain": 0.0,
1884
+ "null_recall": "n/a",
1885
+ "n": 1800
1886
+ },
1887
+ "score": {
1888
+ "nll": 0.7179249707369069,
1889
+ "brier": 0.42831950719002104,
1890
+ "jsd": 0.21507678720358783,
1891
+ "acc": 0.6922222222222222,
1892
+ "acc_k": 0.6922222222222222,
1893
+ "ece": 0.12505331529511346,
1894
+ "sel_acc80": 0.7430555555555556,
1895
+ "coverage": 1.0,
1896
+ "false_abstain": 0.0,
1897
+ "null_recall": "n/a",
1898
+ "n": 1800
1899
+ }
1900
+ },
1901
+ "by_label_class": {
1902
+ "answer": {
1903
+ "nll": 0.47733330766959625,
1904
+ "brier": 0.293311500019393,
1905
+ "jsd": 0.1385319090950344,
1906
+ "acc": 0.7833333333333333,
1907
+ "acc_k": 0.7837037037037037,
1908
+ "ece": 0.03959470673843665,
1909
+ "sel_acc80": 0.8453703703703703,
1910
+ "coverage": 0.9996296296296296,
1911
+ "false_abstain": 0.00037037037037037035,
1912
+ "null_recall": "n/a",
1913
+ "n": 5400
1914
+ }
1915
+ },
1916
+ "baselines": {
1917
+ "majority_class": 1,
1918
+ "majority_acc": 0.44166666666666665
1919
+ }
1920
+ },
1921
+ "tree_choice_cap": {
1922
+ "n": 720,
1923
+ "n_total": 720,
1924
+ "suffix_overflow": 0,
1925
+ "raw": {
1926
+ "nll": 1.9904785368215312,
1927
+ "brier": 0.776743891433536,
1928
+ "jsd": 0.35171423237051735,
1929
+ "acc": 0.4736111111111111,
1930
+ "acc_k": 0.7069444444444445,
1931
+ "ece": 0.2539656482378228,
1932
+ "sel_acc80": 0.5295138888888888,
1933
+ "coverage": 0.6055555555555556,
1934
+ "false_abstain": 0.39444444444444443,
1935
+ "null_recall": "n/a"
1936
+ },
1937
+ "by_family": {
1938
+ "incident_catalog": {
1939
+ "nll": 1.9904785368215312,
1940
+ "brier": 0.776743891433536,
1941
+ "jsd": 0.35171423237051735,
1942
+ "acc": 0.4736111111111111,
1943
+ "acc_k": 0.7069444444444445,
1944
+ "ece": 0.2539656482378228,
1945
+ "sel_acc80": 0.5295138888888888,
1946
+ "coverage": 0.6055555555555556,
1947
+ "false_abstain": 0.39444444444444443,
1948
+ "null_recall": "n/a",
1949
+ "n": 720
1950
+ }
1951
+ },
1952
+ "by_qtype": {
1953
+ "choice": {
1954
+ "nll": 1.9904785368215312,
1955
+ "brier": 0.776743891433536,
1956
+ "jsd": 0.35171423237051735,
1957
+ "acc": 0.4736111111111111,
1958
+ "acc_k": 0.7069444444444445,
1959
+ "ece": 0.2539656482378228,
1960
+ "sel_acc80": 0.5295138888888888,
1961
+ "coverage": 0.6055555555555556,
1962
+ "false_abstain": 0.39444444444444443,
1963
+ "null_recall": "n/a",
1964
+ "n": 720
1965
+ }
1966
+ },
1967
+ "by_label_class": {
1968
+ "answer": {
1969
+ "nll": 1.9904785368215312,
1970
+ "brier": 0.776743891433536,
1971
+ "jsd": 0.35171423237051735,
1972
+ "acc": 0.4736111111111111,
1973
+ "acc_k": 0.7069444444444445,
1974
+ "ece": 0.2539656482378228,
1975
+ "sel_acc80": 0.5295138888888888,
1976
+ "coverage": 0.6055555555555556,
1977
+ "false_abstain": 0.39444444444444443,
1978
+ "null_recall": "n/a",
1979
+ "n": 720
1980
+ }
1981
+ },
1982
+ "baselines": {
1983
+ "majority_class": 3,
1984
+ "majority_acc": 0.11388888888888889
1985
+ }
1986
+ },
1987
+ "typed_decisions_test": {
1988
+ "n": 2000,
1989
+ "n_total": 2000,
1990
+ "suffix_overflow": 0,
1991
+ "raw": {
1992
+ "nll": 1.2466883275443574,
1993
+ "brier": 0.227664179143571,
1994
+ "jsd": 0.10287304051205891,
1995
+ "acc": 0.5105,
1996
+ "acc_k": 0.5105,
1997
+ "ece": 0.08129057371616365,
1998
+ "sel_acc80": 0.538125,
1999
+ "coverage": 1.0,
2000
+ "false_abstain": 0.0,
2001
+ "null_recall": "n/a"
2002
+ },
2003
+ "by_family": {
2004
+ "agent_trace_observability": {
2005
+ "nll": 1.3732859886964037,
2006
+ "brier": 0.24645547871103435,
2007
+ "jsd": 0.11540308138807402,
2008
+ "acc": 0.328,
2009
+ "acc_k": 0.328,
2010
+ "ece": 0.17124855417013168,
2011
+ "sel_acc80": 0.3525,
2012
+ "coverage": 1.0,
2013
+ "false_abstain": 0.0,
2014
+ "null_recall": "n/a",
2015
+ "n": 500
2016
+ },
2017
+ "customer_service": {
2018
+ "nll": 1.3569411167874115,
2019
+ "brier": 0.23403111645670874,
2020
+ "jsd": 0.10776170713367438,
2021
+ "acc": 0.588,
2022
+ "acc_k": 0.588,
2023
+ "ece": 0.14043137019872665,
2024
+ "sel_acc80": 0.6125,
2025
+ "coverage": 1.0,
2026
+ "false_abstain": 0.0,
2027
+ "null_recall": "n/a",
2028
+ "n": 500
2029
+ },
2030
+ "invoice_processing": {
2031
+ "nll": 1.1945578194974424,
2032
+ "brier": 0.32410382732931886,
2033
+ "jsd": 0.14279780746854975,
2034
+ "acc": 0.52,
2035
+ "acc_k": 0.52,
2036
+ "ece": 0.18908735823631287,
2037
+ "sel_acc80": 0.525,
2038
+ "coverage": 1.0,
2039
+ "false_abstain": 0.0,
2040
+ "null_recall": "n/a",
2041
+ "n": 500
2042
+ },
2043
+ "security_incidents": {
2044
+ "nll": 1.0619683851961725,
2045
+ "brier": 0.10606629407722211,
2046
+ "jsd": 0.04552956605793744,
2047
+ "acc": 0.606,
2048
+ "acc_k": 0.606,
2049
+ "ece": 0.12156618690490725,
2050
+ "sel_acc80": 0.615,
2051
+ "coverage": 1.0,
2052
+ "false_abstain": 0.0,
2053
+ "null_recall": "n/a",
2054
+ "n": 500
2055
+ }
2056
+ },
2057
+ "by_qtype": {
2058
+ "choice": {
2059
+ "nll": 1.6514770692249574,
2060
+ "brier": 0.3027703648728876,
2061
+ "jsd": 0.13991168303102539,
2062
+ "acc": 0.4716666666666667,
2063
+ "acc_k": 0.4716666666666667,
2064
+ "ece": 0.20745025321841243,
2065
+ "sel_acc80": 0.45416666666666666,
2066
+ "coverage": 1.0,
2067
+ "false_abstain": 0.0,
2068
+ "null_recall": "n/a",
2069
+ "n": 600
2070
+ },
2071
+ "noul": {
2072
+ "nll": 0.7184778931000595,
2073
+ "brier": 0.1950290479092196,
2074
+ "jsd": 0.0656428058548165,
2075
+ "acc": 0.6616666666666666,
2076
+ "acc_k": 0.6616666666666666,
2077
+ "ece": 0.09978362967570625,
2078
+ "sel_acc80": 0.6979166666666666,
2079
+ "coverage": 1.0,
2080
+ "false_abstain": 0.0,
2081
+ "null_recall": "n/a",
2082
+ "n": 600
2083
+ },
2084
+ "score": {
2085
+ "nll": 1.3392545971171312,
2086
+ "brier": 0.19581088827234716,
2087
+ "jsd": 0.10301673461576585,
2088
+ "acc": 0.42625,
2089
+ "acc_k": 0.42625,
2090
+ "ece": 0.04972978692501786,
2091
+ "sel_acc80": 0.43125,
2092
+ "coverage": 1.0,
2093
+ "false_abstain": 0.0,
2094
+ "null_recall": "n/a",
2095
+ "n": 800
2096
+ }
2097
+ },
2098
+ "by_label_class": {
2099
+ "answer": {
2100
+ "nll": 1.2466883275443574,
2101
+ "brier": 0.227664179143571,
2102
+ "jsd": 0.10287304051205891,
2103
+ "acc": 0.5105,
2104
+ "acc_k": 0.5105,
2105
+ "ece": 0.08129057371616365,
2106
+ "sel_acc80": 0.538125,
2107
+ "coverage": 1.0,
2108
+ "false_abstain": 0.0,
2109
+ "null_recall": "n/a",
2110
+ "n": 2000
2111
+ }
2112
+ },
2113
+ "baselines": {
2114
+ "majority_class": 1,
2115
+ "majority_acc": 0.2905
2116
+ }
2117
+ },
2118
+ "typed_decisions_train": {
2119
+ "n": 6000,
2120
+ "n_total": 6000,
2121
+ "suffix_overflow": 0,
2122
+ "raw": {
2123
+ "nll": 1.2344590285798438,
2124
+ "brier": 0.23647349698787676,
2125
+ "jsd": 0.10473287517672107,
2126
+ "acc": 0.49433333333333335,
2127
+ "acc_k": 0.4945,
2128
+ "ece": 0.09849566911160947,
2129
+ "sel_acc80": 0.515625,
2130
+ "coverage": 0.9995,
2131
+ "false_abstain": 0.0005,
2132
+ "null_recall": "n/a"
2133
+ },
2134
+ "by_family": {
2135
+ "agent_trace_observability": {
2136
+ "nll": 1.3073141496670826,
2137
+ "brier": 0.21427903025345926,
2138
+ "jsd": 0.09931973903182195,
2139
+ "acc": 0.35933333333333334,
2140
+ "acc_k": 0.35933333333333334,
2141
+ "ece": 0.14304505523045857,
2142
+ "sel_acc80": 0.37666666666666665,
2143
+ "coverage": 0.9993333333333333,
2144
+ "false_abstain": 0.0006666666666666666,
2145
+ "null_recall": "n/a",
2146
+ "n": 1500
2147
+ },
2148
+ "customer_service": {
2149
+ "nll": 1.3559591669399014,
2150
+ "brier": 0.25086907210466536,
2151
+ "jsd": 0.11360149087906343,
2152
+ "acc": 0.564,
2153
+ "acc_k": 0.5646666666666667,
2154
+ "ece": 0.14432381959756216,
2155
+ "sel_acc80": 0.5766666666666667,
2156
+ "coverage": 0.9986666666666667,
2157
+ "false_abstain": 0.0013333333333333333,
2158
+ "null_recall": "n/a",
2159
+ "n": 1500
2160
+ },
2161
+ "invoice_processing": {
2162
+ "nll": 1.1943725988328637,
2163
+ "brier": 0.33419169017658246,
2164
+ "jsd": 0.14631432442839343,
2165
+ "acc": 0.5133333333333333,
2166
+ "acc_k": 0.5133333333333333,
2167
+ "ece": 0.16757222191492716,
2168
+ "sel_acc80": 0.5283333333333333,
2169
+ "coverage": 1.0,
2170
+ "false_abstain": 0.0,
2171
+ "null_recall": "n/a",
2172
+ "n": 1500
2173
+ },
2174
+ "security_incidents": {
2175
+ "nll": 1.080190198879528,
2176
+ "brier": 0.1465541954168001,
2177
+ "jsd": 0.05969594636760551,
2178
+ "acc": 0.5406666666666666,
2179
+ "acc_k": 0.5406666666666666,
2180
+ "ece": 0.08952901397148767,
2181
+ "sel_acc80": 0.5558333333333333,
2182
+ "coverage": 1.0,
2183
+ "false_abstain": 0.0,
2184
+ "null_recall": "n/a",
2185
+ "n": 1500
2186
+ }
2187
+ },
2188
+ "by_qtype": {
2189
+ "choice": {
2190
+ "nll": 1.617389415082502,
2191
+ "brier": 0.3021717268631206,
2192
+ "jsd": 0.1377373031403558,
2193
+ "acc": 0.4777777777777778,
2194
+ "acc_k": 0.4777777777777778,
2195
+ "ece": 0.2272327849931187,
2196
+ "sel_acc80": 0.4625,
2197
+ "coverage": 1.0,
2198
+ "false_abstain": 0.0,
2199
+ "null_recall": "n/a",
2200
+ "n": 1800
2201
+ },
2202
+ "noul": {
2203
+ "nll": 0.7349958482352102,
2204
+ "brier": 0.21425469853856416,
2205
+ "jsd": 0.07184645864360849,
2206
+ "acc": 0.6216666666666667,
2207
+ "acc_k": 0.6216666666666667,
2208
+ "ece": 0.09909297843774163,
2209
+ "sel_acc80": 0.6618055555555555,
2210
+ "coverage": 1.0,
2211
+ "false_abstain": 0.0,
2212
+ "null_recall": "n/a",
2213
+ "n": 1800
2214
+ },
2215
+ "score": {
2216
+ "nll": 1.321858623961326,
2217
+ "brier": 0.20386392341842843,
2218
+ "jsd": 0.10464436660382949,
2219
+ "acc": 0.41125,
2220
+ "acc_k": 0.4116666666666667,
2221
+ "ece": 0.05669290943692127,
2222
+ "sel_acc80": 0.4119791666666667,
2223
+ "coverage": 0.99875,
2224
+ "false_abstain": 0.00125,
2225
+ "null_recall": "n/a",
2226
+ "n": 2400
2227
+ }
2228
+ },
2229
+ "by_label_class": {
2230
+ "answer": {
2231
+ "nll": 1.2344590285798438,
2232
+ "brier": 0.23647349698787676,
2233
+ "jsd": 0.10473287517672107,
2234
+ "acc": 0.49433333333333335,
2235
+ "acc_k": 0.4945,
2236
+ "ece": 0.09849566911160947,
2237
+ "sel_acc80": 0.515625,
2238
+ "coverage": 0.9995,
2239
+ "false_abstain": 0.0005,
2240
+ "null_recall": "n/a",
2241
+ "n": 6000
2242
+ }
2243
+ },
2244
+ "baselines": {
2245
+ "majority_class": 1,
2246
+ "majority_acc": 0.3105
2247
+ }
2248
+ },
2249
+ "wh_heldout_family": {
2250
+ "n": 422,
2251
+ "n_total": 422,
2252
+ "suffix_overflow": 0,
2253
+ "raw": {
2254
+ "nll": 0.5457370583615766,
2255
+ "brier": 0.251593830017657,
2256
+ "jsd": 0.10264112252818929,
2257
+ "acc": 0.8364928909952607,
2258
+ "acc_k": 0.8647214854111406,
2259
+ "ece": 0.07603205083670775,
2260
+ "sel_acc80": 0.9112426035502958,
2261
+ "conf_wrong": 0.0,
2262
+ "coverage": 0.8767772511848342,
2263
+ "false_abstain": 0.034482758620689655,
2264
+ "null_recall": 0.8666666666666667,
2265
+ "auroc_null": 0.9863837312113173,
2266
+ "auroc_conf": 0.9761862658414382
2267
+ },
2268
+ "by_family": {
2269
+ "procurement_request": {
2270
+ "nll": 0.7397600080900862,
2271
+ "brier": 0.32257210974221484,
2272
+ "jsd": 0.11885485091987728,
2273
+ "acc": 0.8028169014084507,
2274
+ "acc_k": 0.828125,
2275
+ "ece": 0.16723090347269892,
2276
+ "sel_acc80": 0.8421052631578947,
2277
+ "conf_wrong": 0.0,
2278
+ "coverage": 0.9014084507042254,
2279
+ "false_abstain": 0.0234375,
2280
+ "null_recall": 0.7857142857142857,
2281
+ "auroc_null": 0.9827008928571429,
2282
+ "auroc_conf": 0.9754464285714286,
2283
+ "n": 142
2284
+ },
2285
+ "research_ethics_review": {
2286
+ "nll": 0.3750520999482153,
2287
+ "brier": 0.19797667690325785,
2288
+ "jsd": 0.0921636871204694,
2289
+ "acc": 0.8549618320610687,
2290
+ "acc_k": 0.8717948717948718,
2291
+ "ece": 0.048931630285641604,
2292
+ "sel_acc80": 0.9523809523809523,
2293
+ "conf_wrong": 0.0,
2294
+ "coverage": 0.8625954198473282,
2295
+ "false_abstain": 0.03418803418803419,
2296
+ "null_recall": 1.0,
2297
+ "auroc_null": 0.9981684981684983,
2298
+ "auroc_conf": 0.996947496947497,
2299
+ "n": 131
2300
+ },
2301
+ "subscription_cancellation": {
2302
+ "nll": 0.5108945797756836,
2303
+ "brier": 0.2310900134881206,
2304
+ "jsd": 0.09640081787578399,
2305
+ "acc": 0.8523489932885906,
2306
+ "acc_k": 0.8939393939393939,
2307
+ "ece": 0.06457566934944,
2308
+ "sel_acc80": 0.9083333333333333,
2309
+ "conf_wrong": 0.0,
2310
+ "coverage": 0.8657718120805369,
2311
+ "false_abstain": 0.045454545454545456,
2312
+ "null_recall": 0.8235294117647058,
2313
+ "auroc_null": 0.9803921568627451,
2314
+ "auroc_conf": 0.9670231729055259,
2315
+ "n": 149
2316
+ }
2317
+ },
2318
+ "by_qtype": {},
2319
+ "by_label_class": {
2320
+ "answer": {
2321
+ "nll": 0.5753806494950396,
2322
+ "brier": 0.26114281214808394,
2323
+ "jsd": 0.10430497921088148,
2324
+ "acc": 0.8328912466843501,
2325
+ "acc_k": 0.8647214854111406,
2326
+ "ece": 0.08095392489939215,
2327
+ "sel_acc80": 0.9039735099337748,
2328
+ "coverage": 0.9655172413793104,
2329
+ "false_abstain": 0.034482758620689655,
2330
+ "null_recall": "n/a",
2331
+ "n": 377
2332
+ },
2333
+ "null": {
2334
+ "nll": 0.29738963931011947,
2335
+ "brier": 0.17159457972496925,
2336
+ "jsd": 0.08870170098652361,
2337
+ "acc": 0.8666666666666667,
2338
+ "ece": 0.07395036551687453,
2339
+ "sel_acc80": 0.9444444444444444,
2340
+ "conf_wrong": 0.0,
2341
+ "coverage": 0.1333333333333333,
2342
+ "false_abstain": "n/a",
2343
+ "null_recall": 0.8666666666666667,
2344
+ "n": 45
2345
+ }
2346
+ },
2347
+ "baselines": {
2348
+ "majority_class": 0,
2349
+ "majority_acc": 0.38388625592417064
2350
+ }
2351
+ },
2352
+ "wh_heldout_grammar": {
2353
+ "n": 816,
2354
+ "n_total": 816,
2355
+ "suffix_overflow": 0,
2356
+ "raw": {
2357
+ "nll": 0.22640730243759352,
2358
+ "brier": 0.1324652194635447,
2359
+ "jsd": 0.06170653080121566,
2360
+ "acc": 0.8982843137254902,
2361
+ "acc_k": 0.9350828729281768,
2362
+ "ece": 0.02578789211225277,
2363
+ "sel_acc80": 0.9816232771822359,
2364
+ "conf_wrong": 0.0,
2365
+ "coverage": 0.8651960784313726,
2366
+ "false_abstain": 0.04281767955801105,
2367
+ "null_recall": 0.8586956521739131,
2368
+ "auroc_null": 0.9852720393946673,
2369
+ "auroc_conf": 0.9800024021138602
2370
+ },
2371
+ "by_family": {
2372
+ "visa_extension": {
2373
+ "nll": 0.3076947705638739,
2374
+ "brier": 0.171614329447727,
2375
+ "jsd": 0.08494472212807742,
2376
+ "acc": 0.8794326241134752,
2377
+ "acc_k": 0.9217391304347826,
2378
+ "ece": 0.08212125449315875,
2379
+ "sel_acc80": 0.9646017699115044,
2380
+ "conf_wrong": 0.0,
2381
+ "coverage": 0.7872340425531915,
2382
+ "false_abstain": 0.06086956521739131,
2383
+ "null_recall": 0.8846153846153846,
2384
+ "auroc_null": 0.977257525083612,
2385
+ "auroc_conf": 0.9782608695652173,
2386
+ "n": 141
2387
+ },
2388
+ "loan_approval": {
2389
+ "nll": 0.0009530163086001497,
2390
+ "brier": 1.5960591192212375e-05,
2391
+ "jsd": 0.0003298857214853352,
2392
+ "acc": 1.0,
2393
+ "acc_k": 1.0,
2394
+ "ece": 0.0009488833211633851,
2395
+ "sel_acc80": 1.0,
2396
+ "conf_wrong": 0.0,
2397
+ "coverage": 0.9708029197080292,
2398
+ "false_abstain": 0.0,
2399
+ "null_recall": 1.0,
2400
+ "auroc_null": 1.0,
2401
+ "auroc_conf": 1.0,
2402
+ "n": 137
2403
+ },
2404
+ "content_moderation": {
2405
+ "nll": 0.6091939270687743,
2406
+ "brier": 0.35397883806023056,
2407
+ "jsd": 0.16220081909887452,
2408
+ "acc": 0.7109375,
2409
+ "acc_k": 0.8144329896907216,
2410
+ "ece": 0.07935253647156058,
2411
+ "sel_acc80": 0.7961165048543689,
2412
+ "conf_wrong": 0.0,
2413
+ "coverage": 0.6796875,
2414
+ "false_abstain": 0.17525773195876287,
2415
+ "null_recall": 0.7741935483870968,
2416
+ "auroc_null": 0.9165281010974393,
2417
+ "auroc_conf": 0.8979048885932823,
2418
+ "n": 128
2419
+ },
2420
+ "warranty_claim": {
2421
+ "nll": 0.024673367126092534,
2422
+ "brier": 0.015632475701378046,
2423
+ "jsd": 0.007645685972270448,
2424
+ "acc": 0.9788732394366197,
2425
+ "acc_k": 0.9782608695652174,
2426
+ "ece": 0.01797439724626676,
2427
+ "sel_acc80": 1.0,
2428
+ "conf_wrong": 0.0,
2429
+ "coverage": 0.971830985915493,
2430
+ "false_abstain": 0.0,
2431
+ "null_recall": 1.0,
2432
+ "auroc_null": 1.0,
2433
+ "auroc_conf": 0.9999999999999999,
2434
+ "n": 142
2435
+ },
2436
+ "shift_swap": {
2437
+ "nll": 0.24592318700294757,
2438
+ "brier": 0.1523044670340519,
2439
+ "jsd": 0.06410268473755716,
2440
+ "acc": 0.9,
2441
+ "acc_k": 0.9327731092436975,
2442
+ "ece": 0.07901871639948625,
2443
+ "sel_acc80": 0.9711538461538461,
2444
+ "conf_wrong": 0.0,
2445
+ "coverage": 0.9153846153846154,
2446
+ "false_abstain": 0.025210084033613446,
2447
+ "null_recall": 0.7272727272727273,
2448
+ "auroc_null": 0.988540870893812,
2449
+ "auroc_conf": 0.9549274255156609,
2450
+ "n": 130
2451
+ },
2452
+ "vendor_onboarding": {
2453
+ "nll": 0.2013215782478294,
2454
+ "brier": 0.12002267024201238,
2455
+ "jsd": 0.05905353406731011,
2456
+ "acc": 0.9057971014492754,
2457
+ "acc_k": 0.9262295081967213,
2458
+ "ece": 0.04172966316126395,
2459
+ "sel_acc80": 0.9819819819819819,
2460
+ "conf_wrong": 0.0,
2461
+ "coverage": 0.855072463768116,
2462
+ "false_abstain": 0.03278688524590164,
2463
+ "null_recall": 1.0,
2464
+ "auroc_null": 1.0,
2465
+ "auroc_conf": 0.9994877049180327,
2466
+ "n": 138
2467
+ }
2468
+ },
2469
+ "by_qtype": {},
2470
+ "by_label_class": {
2471
+ "null": {
2472
+ "nll": 0.31717554619495864,
2473
+ "brier": 0.17949680628344114,
2474
+ "jsd": 0.09801949919342964,
2475
+ "acc": 0.8586956521739131,
2476
+ "ece": 0.0898055029304131,
2477
+ "sel_acc80": 0.972972972972973,
2478
+ "conf_wrong": 0.0,
2479
+ "coverage": 0.14130434782608692,
2480
+ "false_abstain": "n/a",
2481
+ "null_recall": 0.8586956521739131,
2482
+ "n": 92
2483
+ },
2484
+ "answer": {
2485
+ "nll": 0.21487321621428196,
2486
+ "brier": 0.12648882997814348,
2487
+ "jsd": 0.05709217570165255,
2488
+ "acc": 0.9033149171270718,
2489
+ "acc_k": 0.9350828729281768,
2490
+ "ece": 0.03330249582206351,
2491
+ "sel_acc80": 0.9896551724137931,
2492
+ "coverage": 0.9571823204419889,
2493
+ "false_abstain": 0.04281767955801105,
2494
+ "null_recall": "n/a",
2495
+ "n": 724
2496
+ }
2497
+ },
2498
+ "baselines": {
2499
+ "majority_class": 0,
2500
+ "majority_acc": 0.38480392156862747
2501
+ }
2502
+ },
2503
+ "wh_heldout_style": {
2504
+ "n": 403,
2505
+ "n_total": 403,
2506
+ "suffix_overflow": 0,
2507
+ "raw": {
2508
+ "nll": 0.2542938206897888,
2509
+ "brier": 0.1392317138092667,
2510
+ "jsd": 0.06525751858240728,
2511
+ "acc": 0.8957816377171216,
2512
+ "acc_k": 0.8957816377171216,
2513
+ "ece": 0.034956357689116715,
2514
+ "sel_acc80": 0.9845201238390093,
2515
+ "coverage": 1.0,
2516
+ "false_abstain": 0.0,
2517
+ "null_recall": "n/a"
2518
+ },
2519
+ "by_family": {
2520
+ "feature_flag_rollout": {
2521
+ "nll": 0.23742768955897636,
2522
+ "brier": 0.14074202037136146,
2523
+ "jsd": 0.06479365092611951,
2524
+ "acc": 0.8909090909090909,
2525
+ "acc_k": 0.8909090909090909,
2526
+ "ece": 0.05244000716642898,
2527
+ "sel_acc80": 1.0,
2528
+ "coverage": 1.0,
2529
+ "false_abstain": 0.0,
2530
+ "null_recall": "n/a",
2531
+ "n": 55
2532
+ },
2533
+ "shift_swap": {
2534
+ "nll": 0.1414648972583102,
2535
+ "brier": 0.08477401343554897,
2536
+ "jsd": 0.04047649253786089,
2537
+ "acc": 0.9393939393939394,
2538
+ "acc_k": 0.9393939393939394,
2539
+ "ece": 0.04961613859191087,
2540
+ "sel_acc80": 1.0,
2541
+ "coverage": 1.0,
2542
+ "false_abstain": 0.0,
2543
+ "null_recall": "n/a",
2544
+ "n": 66
2545
+ },
2546
+ "content_moderation": {
2547
+ "nll": 0.48871335155887385,
2548
+ "brier": 0.25520722330845164,
2549
+ "jsd": 0.12672497899891835,
2550
+ "acc": 0.813953488372093,
2551
+ "acc_k": 0.813953488372093,
2552
+ "ece": 0.09241127967834474,
2553
+ "sel_acc80": 0.8857142857142857,
2554
+ "coverage": 1.0,
2555
+ "false_abstain": 0.0,
2556
+ "null_recall": "n/a",
2557
+ "n": 43
2558
+ },
2559
+ "insurance_claim": {
2560
+ "nll": 0.2962156035545307,
2561
+ "brier": 0.15384219869919274,
2562
+ "jsd": 0.06801950475356011,
2563
+ "acc": 0.8529411764705882,
2564
+ "acc_k": 0.8529411764705882,
2565
+ "ece": 0.10579277136746572,
2566
+ "sel_acc80": 1.0,
2567
+ "coverage": 1.0,
2568
+ "false_abstain": 0.0,
2569
+ "null_recall": "n/a",
2570
+ "n": 34
2571
+ },
2572
+ "visa_extension": {
2573
+ "nll": 0.0750532288436404,
2574
+ "brier": 0.03625713740971259,
2575
+ "jsd": 0.021720251237289188,
2576
+ "acc": 0.972972972972973,
2577
+ "acc_k": 0.972972972972973,
2578
+ "ece": 0.04014645557145816,
2579
+ "sel_acc80": 1.0,
2580
+ "coverage": 1.0,
2581
+ "false_abstain": 0.0,
2582
+ "null_recall": "n/a",
2583
+ "n": 37
2584
+ },
2585
+ "warranty_claim": {
2586
+ "nll": 0.35407836222932965,
2587
+ "brier": 0.1812433311456853,
2588
+ "jsd": 0.08922631164563992,
2589
+ "acc": 0.864406779661017,
2590
+ "acc_k": 0.864406779661017,
2591
+ "ece": 0.07289651036262515,
2592
+ "sel_acc80": 0.9583333333333334,
2593
+ "coverage": 1.0,
2594
+ "false_abstain": 0.0,
2595
+ "null_recall": "n/a",
2596
+ "n": 59
2597
+ },
2598
+ "loan_approval": {
2599
+ "nll": 0.23993441668876092,
2600
+ "brier": 0.12858445334961865,
2601
+ "jsd": 0.05228019078464824,
2602
+ "acc": 0.9166666666666666,
2603
+ "acc_k": 0.9166666666666666,
2604
+ "ece": 0.08637669434150061,
2605
+ "sel_acc80": 0.9487179487179487,
2606
+ "coverage": 1.0,
2607
+ "false_abstain": 0.0,
2608
+ "null_recall": "n/a",
2609
+ "n": 48
2610
+ },
2611
+ "vendor_onboarding": {
2612
+ "nll": 0.2264713789426459,
2613
+ "brier": 0.13709860520312941,
2614
+ "jsd": 0.06105561608466616,
2615
+ "acc": 0.9016393442622951,
2616
+ "acc_k": 0.9016393442622951,
2617
+ "ece": 0.04174451700976639,
2618
+ "sel_acc80": 0.9795918367346939,
2619
+ "coverage": 1.0,
2620
+ "false_abstain": 0.0,
2621
+ "null_recall": "n/a",
2622
+ "n": 61
2623
+ }
2624
+ },
2625
+ "by_qtype": {},
2626
+ "by_label_class": {
2627
+ "answer": {
2628
+ "nll": 0.2542938206897888,
2629
+ "brier": 0.1392317138092667,
2630
+ "jsd": 0.06525751858240728,
2631
+ "acc": 0.8957816377171216,
2632
+ "acc_k": 0.8957816377171216,
2633
+ "ece": 0.034956357689116715,
2634
+ "sel_acc80": 0.9845201238390093,
2635
+ "coverage": 1.0,
2636
+ "false_abstain": 0.0,
2637
+ "null_recall": "n/a",
2638
+ "n": 403
2639
+ }
2640
+ },
2641
+ "baselines": {
2642
+ "majority_class": 1,
2643
+ "majority_acc": 0.4267990074441687
2644
+ }
2645
+ },
2646
+ "wh_level7": {
2647
+ "n": 880,
2648
+ "n_total": 880,
2649
+ "suffix_overflow": 0,
2650
+ "raw": {
2651
+ "nll": 1.0933667634837307,
2652
+ "brier": 0.6994650652421958,
2653
+ "jsd": 0.27543897835324205,
2654
+ "acc": 0.4954545454545455,
2655
+ "acc_k": 0.4954545454545455,
2656
+ "ece": 0.21106459528885108,
2657
+ "sel_acc80": 0.5,
2658
+ "coverage": 1.0,
2659
+ "false_abstain": 0.0,
2660
+ "null_recall": "n/a"
2661
+ },
2662
+ "by_family": {
2663
+ "temporal": {
2664
+ "nll": 1.3079168514120685,
2665
+ "brier": 0.824014775523723,
2666
+ "jsd": 0.2990870419496964,
2667
+ "acc": 0.4409090909090909,
2668
+ "acc_k": 0.4409090909090909,
2669
+ "ece": 0.33636484200304206,
2670
+ "sel_acc80": 0.4375,
2671
+ "coverage": 1.0,
2672
+ "false_abstain": 0.0,
2673
+ "null_recall": "n/a",
2674
+ "n": 220
2675
+ },
2676
+ "probability": {
2677
+ "nll": 0.9432978854839996,
2678
+ "brier": 0.6099551660833942,
2679
+ "jsd": 0.26288584878103655,
2680
+ "acc": 0.5136363636363637,
2681
+ "acc_k": 0.5136363636363637,
2682
+ "ece": 0.11962269801985136,
2683
+ "sel_acc80": 0.5454545454545454,
2684
+ "coverage": 1.0,
2685
+ "false_abstain": 0.0,
2686
+ "null_recall": "n/a",
2687
+ "n": 220
2688
+ },
2689
+ "numeric": {
2690
+ "nll": 1.1800030350316557,
2691
+ "brier": 0.7649147915563862,
2692
+ "jsd": 0.27636414067609133,
2693
+ "acc": 0.4772727272727273,
2694
+ "acc_k": 0.4772727272727273,
2695
+ "ece": 0.3373937555334785,
2696
+ "sel_acc80": 0.4602272727272727,
2697
+ "coverage": 1.0,
2698
+ "false_abstain": 0.0,
2699
+ "null_recall": "n/a",
2700
+ "n": 220
2701
+ },
2702
+ "tradeoff": {
2703
+ "nll": 0.9422492820071992,
2704
+ "brier": 0.5989755278052799,
2705
+ "jsd": 0.26341888200614394,
2706
+ "acc": 0.55,
2707
+ "acc_k": 0.55,
2708
+ "ece": 0.1418665295297449,
2709
+ "sel_acc80": 0.5909090909090909,
2710
+ "coverage": 1.0,
2711
+ "false_abstain": 0.0,
2712
+ "null_recall": "n/a",
2713
+ "n": 220
2714
+ }
2715
+ },
2716
+ "by_qtype": {},
2717
+ "by_label_class": {
2718
+ "answer": {
2719
+ "nll": 1.0933667634837307,
2720
+ "brier": 0.6994650652421958,
2721
+ "jsd": 0.27543897835324205,
2722
+ "acc": 0.4954545454545455,
2723
+ "acc_k": 0.4954545454545455,
2724
+ "ece": 0.21106459528885108,
2725
+ "sel_acc80": 0.5,
2726
+ "coverage": 1.0,
2727
+ "false_abstain": 0.0,
2728
+ "null_recall": "n/a",
2729
+ "n": 880
2730
+ }
2731
+ },
2732
+ "baselines": {
2733
+ "majority_class": 1,
2734
+ "majority_acc": 0.4602272727272727
2735
+ }
2736
+ },
2737
+ "wh_rubric_flip": {
2738
+ "n": 2206,
2739
+ "n_total": 2206,
2740
+ "suffix_overflow": 0,
2741
+ "raw": {
2742
+ "nll": 0.6494624590463446,
2743
+ "brier": 0.3918518011011721,
2744
+ "jsd": 0.15974617492224252,
2745
+ "acc": 0.7180417044424298,
2746
+ "acc_k": 0.7261847774054572,
2747
+ "ece": 0.10136204418926378,
2748
+ "sel_acc80": 0.7705382436260623,
2749
+ "conf_wrong": 0.0,
2750
+ "coverage": 0.9360834088848595,
2751
+ "false_abstain": 0.019147917663954045,
2752
+ "null_recall": 0.8632478632478633,
2753
+ "auroc_null": 0.993150937143278,
2754
+ "auroc_conf": 0.9631361670614902
2755
+ },
2756
+ "by_family": {
2757
+ "loan_approval": {
2758
+ "nll": 0.08023158175565527,
2759
+ "brier": 0.04282249159650115,
2760
+ "jsd": 0.017602264591185385,
2761
+ "acc": 0.9722222222222222,
2762
+ "acc_k": 0.9716312056737588,
2763
+ "ece": 0.028174331204758714,
2764
+ "sel_acc80": 1.0,
2765
+ "conf_wrong": 0.0,
2766
+ "coverage": 0.9791666666666666,
2767
+ "false_abstain": 0.0,
2768
+ "null_recall": 1.0,
2769
+ "auroc_null": 1.0,
2770
+ "auroc_conf": 1.0,
2771
+ "n": 144
2772
+ },
2773
+ "shift_swap": {
2774
+ "nll": 0.2465878122980097,
2775
+ "brier": 0.15229776876027482,
2776
+ "jsd": 0.06560878599344538,
2777
+ "acc": 0.8975903614457831,
2778
+ "acc_k": 0.9230769230769231,
2779
+ "ece": 0.07610932674752656,
2780
+ "sel_acc80": 0.9699248120300752,
2781
+ "conf_wrong": 0.0,
2782
+ "coverage": 0.9397590361445783,
2783
+ "false_abstain": 0.019230769230769232,
2784
+ "null_recall": 0.7,
2785
+ "auroc_null": 0.9903846153846154,
2786
+ "auroc_conf": 0.948076923076923,
2787
+ "n": 166
2788
+ },
2789
+ "vendor_onboarding": {
2790
+ "nll": 0.2647327007139645,
2791
+ "brier": 0.16004266399217373,
2792
+ "jsd": 0.07542805791158898,
2793
+ "acc": 0.8782051282051282,
2794
+ "acc_k": 0.8943661971830986,
2795
+ "ece": 0.03141318280727436,
2796
+ "sel_acc80": 0.96,
2797
+ "conf_wrong": 0.0,
2798
+ "coverage": 0.8846153846153846,
2799
+ "false_abstain": 0.028169014084507043,
2800
+ "null_recall": 1.0,
2801
+ "auroc_null": 1.0,
2802
+ "auroc_conf": 1.0,
2803
+ "n": 156
2804
+ },
2805
+ "content_moderation": {
2806
+ "nll": 0.582850606871788,
2807
+ "brier": 0.33758376751430147,
2808
+ "jsd": 0.15752963013387192,
2809
+ "acc": 0.722972972972973,
2810
+ "acc_k": 0.8083333333333333,
2811
+ "ece": 0.0678231577212746,
2812
+ "sel_acc80": 0.8067226890756303,
2813
+ "conf_wrong": 0.0,
2814
+ "coverage": 0.7567567567567568,
2815
+ "false_abstain": 0.11666666666666667,
2816
+ "null_recall": 0.7857142857142857,
2817
+ "auroc_null": 0.949702380952381,
2818
+ "auroc_conf": 0.9166666666666667,
2819
+ "n": 148
2820
+ },
2821
+ "warranty_claim": {
2822
+ "nll": 0.1483655041739517,
2823
+ "brier": 0.07844069543217347,
2824
+ "jsd": 0.037962196465545404,
2825
+ "acc": 0.935064935064935,
2826
+ "acc_k": 0.9342105263157895,
2827
+ "ece": 0.037407628901593104,
2828
+ "sel_acc80": 1.0,
2829
+ "conf_wrong": 0.0,
2830
+ "coverage": 0.987012987012987,
2831
+ "false_abstain": 0.0,
2832
+ "null_recall": 1.0,
2833
+ "auroc_null": 1.0,
2834
+ "auroc_conf": 1.0,
2835
+ "n": 154
2836
+ },
2837
+ "visa_extension": {
2838
+ "nll": 0.30290840119057044,
2839
+ "brier": 0.1692011181936417,
2840
+ "jsd": 0.08325356240443149,
2841
+ "acc": 0.8835616438356164,
2842
+ "acc_k": 0.9193548387096774,
2843
+ "ece": 0.0858401405076458,
2844
+ "sel_acc80": 0.9658119658119658,
2845
+ "conf_wrong": 0.0,
2846
+ "coverage": 0.815068493150685,
2847
+ "false_abstain": 0.056451612903225805,
2848
+ "null_recall": 0.9090909090909091,
2849
+ "auroc_null": 0.9798387096774193,
2850
+ "auroc_conf": 0.9791055718475072,
2851
+ "n": 146
2852
+ },
2853
+ "procurement_request": {
2854
+ "nll": 0.8189641494161496,
2855
+ "brier": 0.37043124136320893,
2856
+ "jsd": 0.1353190948728516,
2857
+ "acc": 0.7727272727272727,
2858
+ "acc_k": 0.797979797979798,
2859
+ "ece": 0.19727202355861662,
2860
+ "sel_acc80": 0.8068181818181818,
2861
+ "conf_wrong": 0.0,
2862
+ "coverage": 0.8909090909090909,
2863
+ "false_abstain": 0.030303030303030304,
2864
+ "null_recall": 0.8181818181818182,
2865
+ "auroc_null": 0.9770431588613407,
2866
+ "auroc_conf": 0.9669421487603305,
2867
+ "n": 110
2868
+ },
2869
+ "research_ethics_review": {
2870
+ "nll": 0.42143617293867885,
2871
+ "brier": 0.22013918081838285,
2872
+ "jsd": 0.10185807680534407,
2873
+ "acc": 0.8425925925925926,
2874
+ "acc_k": 0.8526315789473684,
2875
+ "ece": 0.07520498804472113,
2876
+ "sel_acc80": 0.9310344827586207,
2877
+ "conf_wrong": 0.0,
2878
+ "coverage": 0.8518518518518519,
2879
+ "false_abstain": 0.031578947368421054,
2880
+ "null_recall": 1.0,
2881
+ "auroc_null": 1.0,
2882
+ "auroc_conf": 1.0,
2883
+ "n": 108
2884
+ },
2885
+ "subscription_cancellation": {
2886
+ "nll": 0.5600637160483035,
2887
+ "brier": 0.2487595991587257,
2888
+ "jsd": 0.10345149368419938,
2889
+ "acc": 0.847457627118644,
2890
+ "acc_k": 0.9038461538461539,
2891
+ "ece": 0.07515307351694268,
2892
+ "sel_acc80": 0.8842105263157894,
2893
+ "conf_wrong": 0.0,
2894
+ "coverage": 0.8559322033898304,
2895
+ "false_abstain": 0.057692307692307696,
2896
+ "null_recall": 0.7857142857142857,
2897
+ "auroc_null": 0.9732142857142857,
2898
+ "auroc_conf": 0.9546703296703296,
2899
+ "n": 118
2900
+ },
2901
+ "feature_flag_rollout": {
2902
+ "nll": 0.25305771148082157,
2903
+ "brier": 0.14540700448198826,
2904
+ "jsd": 0.0686071941981727,
2905
+ "acc": 0.8913043478260869,
2906
+ "acc_k": 0.8913043478260869,
2907
+ "ece": 0.04402980985848801,
2908
+ "sel_acc80": 1.0,
2909
+ "coverage": 1.0,
2910
+ "false_abstain": 0.0,
2911
+ "null_recall": "n/a",
2912
+ "n": 46
2913
+ },
2914
+ "insurance_claim": {
2915
+ "nll": 0.33136397536790363,
2916
+ "brier": 0.17315097095655338,
2917
+ "jsd": 0.07520947753065718,
2918
+ "acc": 0.8333333333333334,
2919
+ "acc_k": 0.8333333333333334,
2920
+ "ece": 0.11505284309387206,
2921
+ "sel_acc80": 1.0,
2922
+ "coverage": 1.0,
2923
+ "false_abstain": 0.0,
2924
+ "null_recall": "n/a",
2925
+ "n": 30
2926
+ },
2927
+ "temporal": {
2928
+ "nll": 1.307820071546593,
2929
+ "brier": 0.8239995465563967,
2930
+ "jsd": 0.29908146201461855,
2931
+ "acc": 0.4409090909090909,
2932
+ "acc_k": 0.4409090909090909,
2933
+ "ece": 0.336357594891028,
2934
+ "sel_acc80": 0.4375,
2935
+ "coverage": 1.0,
2936
+ "false_abstain": 0.0,
2937
+ "null_recall": "n/a",
2938
+ "n": 220
2939
+ },
2940
+ "numeric": {
2941
+ "nll": 1.1808323022813243,
2942
+ "brier": 0.764822347173263,
2943
+ "jsd": 0.27637131819240957,
2944
+ "acc": 0.4772727272727273,
2945
+ "acc_k": 0.4772727272727273,
2946
+ "ece": 0.3372702945362438,
2947
+ "sel_acc80": 0.4602272727272727,
2948
+ "coverage": 1.0,
2949
+ "false_abstain": 0.0,
2950
+ "null_recall": "n/a",
2951
+ "n": 220
2952
+ },
2953
+ "probability": {
2954
+ "nll": 0.9432978854035944,
2955
+ "brier": 0.6099551661446684,
2956
+ "jsd": 0.26288584877310456,
2957
+ "acc": 0.5136363636363637,
2958
+ "acc_k": 0.5136363636363637,
2959
+ "ece": 0.11962269707159565,
2960
+ "sel_acc80": 0.5454545454545454,
2961
+ "coverage": 1.0,
2962
+ "false_abstain": 0.0,
2963
+ "null_recall": "n/a",
2964
+ "n": 220
2965
+ },
2966
+ "tradeoff": {
2967
+ "nll": 0.9422492821363462,
2968
+ "brier": 0.5989755273230615,
2969
+ "jsd": 0.2634188819148693,
2970
+ "acc": 0.55,
2971
+ "acc_k": 0.55,
2972
+ "ece": 0.14186652952974493,
2973
+ "sel_acc80": 0.5909090909090909,
2974
+ "coverage": 1.0,
2975
+ "false_abstain": 0.0,
2976
+ "null_recall": "n/a",
2977
+ "n": 220
2978
+ }
2979
+ },
2980
+ "by_qtype": {},
2981
+ "by_label_class": {
2982
+ "answer": {
2983
+ "nll": 0.6678934681317353,
2984
+ "brier": 0.40360123283903593,
2985
+ "jsd": 0.16322706971212872,
2986
+ "acc": 0.7099090473910962,
2987
+ "acc_k": 0.7261847774054572,
2988
+ "ece": 0.10932367604962376,
2989
+ "sel_acc80": 0.7607655502392344,
2990
+ "coverage": 0.9808520823360459,
2991
+ "false_abstain": 0.019147917663954045,
2992
+ "null_recall": "n/a",
2993
+ "n": 2089
2994
+ },
2995
+ "null": {
2996
+ "nll": 0.32038230537641876,
2997
+ "brier": 0.18206921220888533,
2998
+ "jsd": 0.09759583974213762,
2999
+ "acc": 0.8632478632478633,
3000
+ "ece": 0.10108689569000506,
3001
+ "sel_acc80": 0.9574468085106383,
3002
+ "conf_wrong": 0.0,
3003
+ "coverage": 0.1367521367521367,
3004
+ "false_abstain": "n/a",
3005
+ "null_recall": 0.8632478632478633,
3006
+ "n": 117
3007
+ }
3008
+ },
3009
+ "baselines": {
3010
+ "majority_class": 0,
3011
+ "majority_acc": 0.39981867633726204
3012
+ }
3013
+ },
3014
+ "wh_rubric_shuffled": {
3015
+ "n": 2437,
3016
+ "n_total": 2437,
3017
+ "suffix_overflow": 0,
3018
+ "raw": {
3019
+ "nll": 2.8769261526719228,
3020
+ "brier": 0.9536641216585189,
3021
+ "jsd": 0.3499893495991487,
3022
+ "acc": 0.4103405826836274,
3023
+ "acc_k": 0.4380704041720991,
3024
+ "ece": 0.4225114438809671,
3025
+ "sel_acc80": 0.41435897435897434,
3026
+ "conf_wrong": 0.1323529411764706,
3027
+ "coverage": 0.9343455067706197,
3028
+ "false_abstain": 0.04389395914819644,
3029
+ "null_recall": 0.4338235294117647,
3030
+ "auroc_null": 0.9573986374210701,
3031
+ "auroc_conf": 0.7991522228187234
3032
+ },
3033
+ "by_family": {
3034
+ "temporal": {
3035
+ "nll": 0.810321112506112,
3036
+ "brier": 0.5886056243893594,
3037
+ "jsd": 0.21902761009641206,
3038
+ "acc": 0.5363636363636364,
3039
+ "acc_k": 0.5363636363636364,
3040
+ "ece": 0.2581337186423215,
3041
+ "sel_acc80": 0.5284090909090909,
3042
+ "coverage": 1.0,
3043
+ "false_abstain": 0.0,
3044
+ "null_recall": "n/a",
3045
+ "n": 220
3046
+ },
3047
+ "shift_swap": {
3048
+ "nll": 4.311645628513916,
3049
+ "brier": 1.1784696578549476,
3050
+ "jsd": 0.4179276671059122,
3051
+ "acc": 0.3403141361256545,
3052
+ "acc_k": 0.35,
3053
+ "ece": 0.5781365601804244,
3054
+ "sel_acc80": 0.33986928104575165,
3055
+ "conf_wrong": 0.2727272727272727,
3056
+ "coverage": 0.9424083769633508,
3057
+ "false_abstain": 0.03888888888888889,
3058
+ "null_recall": 0.36363636363636365,
3059
+ "auroc_null": 0.9570707070707071,
3060
+ "auroc_conf": 0.8712121212121213,
3061
+ "n": 191
3062
+ },
3063
+ "vendor_onboarding": {
3064
+ "nll": 3.8966259577761173,
3065
+ "brier": 1.1348027236364968,
3066
+ "jsd": 0.40071265182720534,
3067
+ "acc": 0.34536082474226804,
3068
+ "acc_k": 0.398876404494382,
3069
+ "ece": 0.5599819146787998,
3070
+ "sel_acc80": 0.358974358974359,
3071
+ "conf_wrong": 0.1875,
3072
+ "coverage": 0.9020618556701031,
3073
+ "false_abstain": 0.07303370786516854,
3074
+ "null_recall": 0.375,
3075
+ "auroc_null": 0.9332865168539325,
3076
+ "auroc_conf": 0.8381320224719101,
3077
+ "n": 194
3078
+ },
3079
+ "loan_approval": {
3080
+ "nll": 5.6109657929658034,
3081
+ "brier": 1.2977122539746735,
3082
+ "jsd": 0.4490296748646031,
3083
+ "acc": 0.34269662921348315,
3084
+ "acc_k": 0.3620689655172414,
3085
+ "ece": 0.6546829131881843,
3086
+ "sel_acc80": 0.35664335664335667,
3087
+ "conf_wrong": 0.5,
3088
+ "coverage": 0.9662921348314607,
3089
+ "false_abstain": 0.022988505747126436,
3090
+ "null_recall": 0.5,
3091
+ "auroc_null": 0.9604885057471264,
3092
+ "auroc_conf": 0.8153735632183908,
3093
+ "n": 178
3094
+ },
3095
+ "probability": {
3096
+ "nll": 1.1624105344814657,
3097
+ "brier": 0.749513864357648,
3098
+ "jsd": 0.30859526077870086,
3099
+ "acc": 0.3409090909090909,
3100
+ "acc_k": 0.3409090909090909,
3101
+ "ece": 0.2819210297682068,
3102
+ "sel_acc80": 0.36363636363636365,
3103
+ "coverage": 1.0,
3104
+ "false_abstain": 0.0,
3105
+ "null_recall": "n/a",
3106
+ "n": 220
3107
+ },
3108
+ "feature_flag_rollout": {
3109
+ "nll": 4.149504215160249,
3110
+ "brier": 1.1719086356322357,
3111
+ "jsd": 0.41258248701508426,
3112
+ "acc": 0.3695652173913043,
3113
+ "acc_k": 0.3695652173913043,
3114
+ "ece": 0.5901150580333627,
3115
+ "sel_acc80": 0.35135135135135137,
3116
+ "coverage": 1.0,
3117
+ "false_abstain": 0.0,
3118
+ "null_recall": "n/a",
3119
+ "n": 46
3120
+ },
3121
+ "research_ethics_review": {
3122
+ "nll": 3.6036260558065356,
3123
+ "brier": 1.1577480248499434,
3124
+ "jsd": 0.41970080669001747,
3125
+ "acc": 0.30952380952380953,
3126
+ "acc_k": 0.375,
3127
+ "ece": 0.5508490079452122,
3128
+ "sel_acc80": 0.3069306930693069,
3129
+ "conf_wrong": 0.07142857142857142,
3130
+ "coverage": 0.8571428571428572,
3131
+ "false_abstain": 0.125,
3132
+ "null_recall": 0.2857142857142857,
3133
+ "auroc_null": 0.8801020408163265,
3134
+ "auroc_conf": 0.7704081632653063,
3135
+ "n": 126
3136
+ },
3137
+ "warranty_claim": {
3138
+ "nll": 4.5780489648436005,
3139
+ "brier": 1.1941381442153995,
3140
+ "jsd": 0.4155502092598703,
3141
+ "acc": 0.36507936507936506,
3142
+ "acc_k": 0.372972972972973,
3143
+ "ece": 0.599066392612205,
3144
+ "sel_acc80": 0.3684210526315789,
3145
+ "conf_wrong": 0.5,
3146
+ "coverage": 0.9788359788359788,
3147
+ "false_abstain": 0.010810810810810811,
3148
+ "null_recall": 0.5,
3149
+ "auroc_null": 0.9864864864864865,
3150
+ "auroc_conf": 0.85,
3151
+ "n": 189
3152
+ },
3153
+ "numeric": {
3154
+ "nll": 0.6629021089805935,
3155
+ "brier": 0.41855250364207175,
3156
+ "jsd": 0.1702423570572618,
3157
+ "acc": 0.7272727272727273,
3158
+ "acc_k": 0.7272727272727273,
3159
+ "ece": 0.1474715121767738,
3160
+ "sel_acc80": 0.7670454545454546,
3161
+ "coverage": 1.0,
3162
+ "false_abstain": 0.0,
3163
+ "null_recall": "n/a",
3164
+ "n": 220
3165
+ },
3166
+ "content_moderation": {
3167
+ "nll": 2.264195490666992,
3168
+ "brier": 0.9375361619969083,
3169
+ "jsd": 0.3576627856018118,
3170
+ "acc": 0.37341772151898733,
3171
+ "acc_k": 0.4251968503937008,
3172
+ "ece": 0.3937665090153489,
3173
+ "sel_acc80": 0.36220472440944884,
3174
+ "conf_wrong": 0.0,
3175
+ "coverage": 0.7341772151898734,
3176
+ "false_abstain": 0.1889763779527559,
3177
+ "null_recall": 0.5806451612903226,
3178
+ "auroc_null": 0.8219456438912878,
3179
+ "auroc_conf": 0.8003556007112014,
3180
+ "n": 158
3181
+ },
3182
+ "visa_extension": {
3183
+ "nll": 3.391274110219176,
3184
+ "brier": 1.221747900683501,
3185
+ "jsd": 0.432846333134556,
3186
+ "acc": 0.27325581395348836,
3187
+ "acc_k": 0.3424657534246575,
3188
+ "ece": 0.6095082028660663,
3189
+ "sel_acc80": 0.2753623188405797,
3190
+ "conf_wrong": 0.038461538461538464,
3191
+ "coverage": 0.8313953488372093,
3192
+ "false_abstain": 0.13013698630136986,
3193
+ "null_recall": 0.38461538461538464,
3194
+ "auroc_null": 0.8746048472075869,
3195
+ "auroc_conf": 0.8253424657534246,
3196
+ "n": 172
3197
+ },
3198
+ "subscription_cancellation": {
3199
+ "nll": 3.274311082023101,
3200
+ "brier": 1.0411595744094404,
3201
+ "jsd": 0.37374148544328023,
3202
+ "acc": 0.3829787234042553,
3203
+ "acc_k": 0.44,
3204
+ "ece": 0.5014726749971403,
3205
+ "sel_acc80": 0.40707964601769914,
3206
+ "conf_wrong": 0.1875,
3207
+ "coverage": 0.8723404255319149,
3208
+ "false_abstain": 0.08,
3209
+ "null_recall": 0.5,
3210
+ "auroc_null": 0.9175,
3211
+ "auroc_conf": 0.8180000000000001,
3212
+ "n": 141
3213
+ },
3214
+ "procurement_request": {
3215
+ "nll": 4.339458520994687,
3216
+ "brier": 1.139405576261341,
3217
+ "jsd": 0.4016875272695481,
3218
+ "acc": 0.37593984962406013,
3219
+ "acc_k": 0.4369747899159664,
3220
+ "ece": 0.557007525424312,
3221
+ "sel_acc80": 0.3925233644859813,
3222
+ "conf_wrong": 0.21428571428571427,
3223
+ "coverage": 0.9022556390977443,
3224
+ "false_abstain": 0.06722689075630252,
3225
+ "null_recall": 0.35714285714285715,
3226
+ "auroc_null": 0.9381752701080432,
3227
+ "auroc_conf": 0.858343337334934,
3228
+ "n": 133
3229
+ },
3230
+ "tradeoff": {
3231
+ "nll": 1.0485382653766773,
3232
+ "brier": 0.6644986709620295,
3233
+ "jsd": 0.2847643375853819,
3234
+ "acc": 0.4863636363636364,
3235
+ "acc_k": 0.4863636363636364,
3236
+ "ece": 0.18097281794656409,
3237
+ "sel_acc80": 0.45454545454545453,
3238
+ "coverage": 1.0,
3239
+ "false_abstain": 0.0,
3240
+ "null_recall": "n/a",
3241
+ "n": 220
3242
+ },
3243
+ "insurance_claim": {
3244
+ "nll": 4.561252266824203,
3245
+ "brier": 1.1413018112706845,
3246
+ "jsd": 0.4059811712854699,
3247
+ "acc": 0.41379310344827586,
3248
+ "acc_k": 0.41379310344827586,
3249
+ "ece": 0.5548728519472582,
3250
+ "sel_acc80": 0.375,
3251
+ "coverage": 1.0,
3252
+ "false_abstain": 0.0,
3253
+ "null_recall": "n/a",
3254
+ "n": 29
3255
+ }
3256
+ },
3257
+ "by_qtype": {},
3258
+ "by_label_class": {
3259
+ "answer": {
3260
+ "nll": 2.961086866040626,
3261
+ "brier": 0.9651978530800092,
3262
+ "jsd": 0.35318397177542504,
3263
+ "acc": 0.40895262929161236,
3264
+ "acc_k": 0.4380704041720991,
3265
+ "ece": 0.43056437424968913,
3266
+ "sel_acc80": 0.41227593699076587,
3267
+ "coverage": 0.9561060408518036,
3268
+ "false_abstain": 0.04389395914819644,
3269
+ "null_recall": "n/a",
3270
+ "n": 2301
3271
+ },
3272
+ "null": {
3273
+ "nll": 1.4530011419264246,
3274
+ "brier": 0.7585235628287442,
3275
+ "jsd": 0.2959391611608268,
3276
+ "acc": 0.4338235294117647,
3277
+ "ece": 0.29192931555649815,
3278
+ "sel_acc80": 0.46788990825688076,
3279
+ "conf_wrong": 0.1323529411764706,
3280
+ "coverage": 0.5661764705882353,
3281
+ "false_abstain": "n/a",
3282
+ "null_recall": 0.4338235294117647,
3283
+ "n": 136
3284
+ }
3285
+ },
3286
+ "baselines": {
3287
+ "majority_class": 0,
3288
+ "majority_acc": 0.4095199015182602
3289
+ }
3290
+ }
3291
+ }
3292
+ }
jevbench_summary.json ADDED
@@ -0,0 +1,2824 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "easy": {
3
+ "accuracy": 1.0,
4
+ "brier_mean": 0.0047407252593300855,
5
+ "calibration_n": 48,
6
+ "complete": true,
7
+ "cost_basis": [
8
+ "local_gpu_no_provider_tariff"
9
+ ],
10
+ "coverage": 1.0,
11
+ "dataset_hash": "3cbd0e951fa8059741955ef2bd3273b0bb60aecae8f4a68b89872f55f7376b3d",
12
+ "ece": {
13
+ "bins": [
14
+ {
15
+ "accuracy": null,
16
+ "hi": 0.1,
17
+ "lo": 0.0,
18
+ "mean_confidence": null,
19
+ "n": 0
20
+ },
21
+ {
22
+ "accuracy": null,
23
+ "hi": 0.2,
24
+ "lo": 0.1,
25
+ "mean_confidence": null,
26
+ "n": 0
27
+ },
28
+ {
29
+ "accuracy": null,
30
+ "hi": 0.3,
31
+ "lo": 0.2,
32
+ "mean_confidence": null,
33
+ "n": 0
34
+ },
35
+ {
36
+ "accuracy": null,
37
+ "hi": 0.4,
38
+ "lo": 0.3,
39
+ "mean_confidence": null,
40
+ "n": 0
41
+ },
42
+ {
43
+ "accuracy": null,
44
+ "hi": 0.5,
45
+ "lo": 0.4,
46
+ "mean_confidence": null,
47
+ "n": 0
48
+ },
49
+ {
50
+ "accuracy": null,
51
+ "hi": 0.6,
52
+ "lo": 0.5,
53
+ "mean_confidence": null,
54
+ "n": 0
55
+ },
56
+ {
57
+ "accuracy": null,
58
+ "hi": 0.7,
59
+ "lo": 0.6,
60
+ "mean_confidence": null,
61
+ "n": 0
62
+ },
63
+ {
64
+ "accuracy": 1.0,
65
+ "hi": 0.8,
66
+ "lo": 0.7,
67
+ "mean_confidence": 0.7135202288627625,
68
+ "n": 1
69
+ },
70
+ {
71
+ "accuracy": 1.0,
72
+ "hi": 0.9,
73
+ "lo": 0.8,
74
+ "mean_confidence": 0.8538632392883301,
75
+ "n": 1
76
+ },
77
+ {
78
+ "accuracy": 1.0,
79
+ "hi": 1.0,
80
+ "lo": 0.9,
81
+ "mean_confidence": 0.9924926835557689,
82
+ "n": 46
83
+ }
84
+ ],
85
+ "ece": 0.016207356005907027,
86
+ "n": 48
87
+ },
88
+ "latency": {
89
+ "n": 48,
90
+ "p50_s": 0.06309161661192775,
91
+ "p95_s": 0.0954449666896835
92
+ },
93
+ "latency_failures": {
94
+ "n": 0,
95
+ "p50_s": null,
96
+ "p95_s": null
97
+ },
98
+ "ledger_charged_usd": null,
99
+ "macro_accuracy": 1.0,
100
+ "model_identities": [
101
+ "pcdm:native:runs/ts1b"
102
+ ],
103
+ "n_attempted": 48,
104
+ "n_correct": 48,
105
+ "n_planned": 48,
106
+ "n_renormalized": 0,
107
+ "n_scorable": 48,
108
+ "n_valid": 48,
109
+ "operational_success": 1.0,
110
+ "ordinal_mae": null,
111
+ "paraphrase_consistency": {
112
+ "agree": 0,
113
+ "agreement": null,
114
+ "both_correct_rate_all_pairs": null,
115
+ "both_valid": 0,
116
+ "pairs": 0
117
+ },
118
+ "per_family": {
119
+ "extraction": {
120
+ "accuracy": 1.0,
121
+ "brier_mean": 0.0006161597573289517,
122
+ "calibration_n": 12,
123
+ "cost_basis": [
124
+ "local_gpu_no_provider_tariff"
125
+ ],
126
+ "coverage": 1.0,
127
+ "ece": {
128
+ "bins": [
129
+ {
130
+ "accuracy": null,
131
+ "hi": 0.1,
132
+ "lo": 0.0,
133
+ "mean_confidence": null,
134
+ "n": 0
135
+ },
136
+ {
137
+ "accuracy": null,
138
+ "hi": 0.2,
139
+ "lo": 0.1,
140
+ "mean_confidence": null,
141
+ "n": 0
142
+ },
143
+ {
144
+ "accuracy": null,
145
+ "hi": 0.3,
146
+ "lo": 0.2,
147
+ "mean_confidence": null,
148
+ "n": 0
149
+ },
150
+ {
151
+ "accuracy": null,
152
+ "hi": 0.4,
153
+ "lo": 0.3,
154
+ "mean_confidence": null,
155
+ "n": 0
156
+ },
157
+ {
158
+ "accuracy": null,
159
+ "hi": 0.5,
160
+ "lo": 0.4,
161
+ "mean_confidence": null,
162
+ "n": 0
163
+ },
164
+ {
165
+ "accuracy": null,
166
+ "hi": 0.6,
167
+ "lo": 0.5,
168
+ "mean_confidence": null,
169
+ "n": 0
170
+ },
171
+ {
172
+ "accuracy": null,
173
+ "hi": 0.7,
174
+ "lo": 0.6,
175
+ "mean_confidence": null,
176
+ "n": 0
177
+ },
178
+ {
179
+ "accuracy": null,
180
+ "hi": 0.8,
181
+ "lo": 0.7,
182
+ "mean_confidence": null,
183
+ "n": 0
184
+ },
185
+ {
186
+ "accuracy": null,
187
+ "hi": 0.9,
188
+ "lo": 0.8,
189
+ "mean_confidence": null,
190
+ "n": 0
191
+ },
192
+ {
193
+ "accuracy": 1.0,
194
+ "hi": 1.0,
195
+ "lo": 0.9,
196
+ "mean_confidence": 0.9888500918944677,
197
+ "n": 12
198
+ }
199
+ ],
200
+ "ece": 0.011149908105532291,
201
+ "n": 12
202
+ },
203
+ "latency": {
204
+ "n": 12,
205
+ "p50_s": 0.06396663328632712,
206
+ "p95_s": 0.08068648739717899
207
+ },
208
+ "latency_failures": {
209
+ "n": 0,
210
+ "p50_s": null,
211
+ "p95_s": null
212
+ },
213
+ "n_attempted": 12,
214
+ "n_correct": 12,
215
+ "n_planned": 12,
216
+ "n_renormalized": 0,
217
+ "n_scorable": 12,
218
+ "n_valid": 12,
219
+ "operational_success": 1.0,
220
+ "ordinal_mae": null,
221
+ "price_per_1000_decisions_usd": null,
222
+ "schema_validity": 1.0,
223
+ "schema_validity_strict": 1.0
224
+ },
225
+ "fact": {
226
+ "accuracy": 1.0,
227
+ "brier_mean": 0.018346439983612378,
228
+ "calibration_n": 12,
229
+ "cost_basis": [
230
+ "local_gpu_no_provider_tariff"
231
+ ],
232
+ "coverage": 1.0,
233
+ "ece": {
234
+ "bins": [
235
+ {
236
+ "accuracy": null,
237
+ "hi": 0.1,
238
+ "lo": 0.0,
239
+ "mean_confidence": null,
240
+ "n": 0
241
+ },
242
+ {
243
+ "accuracy": null,
244
+ "hi": 0.2,
245
+ "lo": 0.1,
246
+ "mean_confidence": null,
247
+ "n": 0
248
+ },
249
+ {
250
+ "accuracy": null,
251
+ "hi": 0.3,
252
+ "lo": 0.2,
253
+ "mean_confidence": null,
254
+ "n": 0
255
+ },
256
+ {
257
+ "accuracy": null,
258
+ "hi": 0.4,
259
+ "lo": 0.3,
260
+ "mean_confidence": null,
261
+ "n": 0
262
+ },
263
+ {
264
+ "accuracy": null,
265
+ "hi": 0.5,
266
+ "lo": 0.4,
267
+ "mean_confidence": null,
268
+ "n": 0
269
+ },
270
+ {
271
+ "accuracy": null,
272
+ "hi": 0.6,
273
+ "lo": 0.5,
274
+ "mean_confidence": null,
275
+ "n": 0
276
+ },
277
+ {
278
+ "accuracy": null,
279
+ "hi": 0.7,
280
+ "lo": 0.6,
281
+ "mean_confidence": null,
282
+ "n": 0
283
+ },
284
+ {
285
+ "accuracy": 1.0,
286
+ "hi": 0.8,
287
+ "lo": 0.7,
288
+ "mean_confidence": 0.7135202288627625,
289
+ "n": 1
290
+ },
291
+ {
292
+ "accuracy": 1.0,
293
+ "hi": 0.9,
294
+ "lo": 0.8,
295
+ "mean_confidence": 0.8538632392883301,
296
+ "n": 1
297
+ },
298
+ {
299
+ "accuracy": 1.0,
300
+ "hi": 1.0,
301
+ "lo": 0.9,
302
+ "mean_confidence": 0.9790186047554016,
303
+ "n": 10
304
+ }
305
+ ],
306
+ "ece": 0.053535873691240965,
307
+ "n": 12
308
+ },
309
+ "latency": {
310
+ "n": 12,
311
+ "p50_s": 0.06284108175896108,
312
+ "p95_s": 0.09521680150646716
313
+ },
314
+ "latency_failures": {
315
+ "n": 0,
316
+ "p50_s": null,
317
+ "p95_s": null
318
+ },
319
+ "n_attempted": 12,
320
+ "n_correct": 12,
321
+ "n_planned": 12,
322
+ "n_renormalized": 0,
323
+ "n_scorable": 12,
324
+ "n_valid": 12,
325
+ "operational_success": 1.0,
326
+ "ordinal_mae": null,
327
+ "price_per_1000_decisions_usd": null,
328
+ "schema_validity": 1.0,
329
+ "schema_validity_strict": 1.0
330
+ },
331
+ "intent": {
332
+ "accuracy": 1.0,
333
+ "brier_mean": 3.0127070021926845e-07,
334
+ "calibration_n": 12,
335
+ "cost_basis": [
336
+ "local_gpu_no_provider_tariff"
337
+ ],
338
+ "coverage": 1.0,
339
+ "ece": {
340
+ "bins": [
341
+ {
342
+ "accuracy": null,
343
+ "hi": 0.1,
344
+ "lo": 0.0,
345
+ "mean_confidence": null,
346
+ "n": 0
347
+ },
348
+ {
349
+ "accuracy": null,
350
+ "hi": 0.2,
351
+ "lo": 0.1,
352
+ "mean_confidence": null,
353
+ "n": 0
354
+ },
355
+ {
356
+ "accuracy": null,
357
+ "hi": 0.3,
358
+ "lo": 0.2,
359
+ "mean_confidence": null,
360
+ "n": 0
361
+ },
362
+ {
363
+ "accuracy": null,
364
+ "hi": 0.4,
365
+ "lo": 0.3,
366
+ "mean_confidence": null,
367
+ "n": 0
368
+ },
369
+ {
370
+ "accuracy": null,
371
+ "hi": 0.5,
372
+ "lo": 0.4,
373
+ "mean_confidence": null,
374
+ "n": 0
375
+ },
376
+ {
377
+ "accuracy": null,
378
+ "hi": 0.6,
379
+ "lo": 0.5,
380
+ "mean_confidence": null,
381
+ "n": 0
382
+ },
383
+ {
384
+ "accuracy": null,
385
+ "hi": 0.7,
386
+ "lo": 0.6,
387
+ "mean_confidence": null,
388
+ "n": 0
389
+ },
390
+ {
391
+ "accuracy": null,
392
+ "hi": 0.8,
393
+ "lo": 0.7,
394
+ "mean_confidence": null,
395
+ "n": 0
396
+ },
397
+ {
398
+ "accuracy": null,
399
+ "hi": 0.9,
400
+ "lo": 0.8,
401
+ "mean_confidence": null,
402
+ "n": 0
403
+ },
404
+ {
405
+ "accuracy": 1.0,
406
+ "hi": 1.0,
407
+ "lo": 0.9,
408
+ "mean_confidence": 0.9998585979143778,
409
+ "n": 12
410
+ }
411
+ ],
412
+ "ece": 0.0001414020856221887,
413
+ "n": 12
414
+ },
415
+ "latency": {
416
+ "n": 12,
417
+ "p50_s": 0.06062154611572623,
418
+ "p95_s": 0.12287021279335017
419
+ },
420
+ "latency_failures": {
421
+ "n": 0,
422
+ "p50_s": null,
423
+ "p95_s": null
424
+ },
425
+ "n_attempted": 12,
426
+ "n_correct": 12,
427
+ "n_planned": 12,
428
+ "n_renormalized": 0,
429
+ "n_scorable": 12,
430
+ "n_valid": 12,
431
+ "operational_success": 1.0,
432
+ "ordinal_mae": null,
433
+ "price_per_1000_decisions_usd": null,
434
+ "schema_validity": 1.0,
435
+ "schema_validity_strict": 1.0
436
+ },
437
+ "tool_selection": {
438
+ "accuracy": 1.0,
439
+ "brier_mean": 2.5678791016948264e-11,
440
+ "calibration_n": 12,
441
+ "cost_basis": [
442
+ "local_gpu_no_provider_tariff"
443
+ ],
444
+ "coverage": 1.0,
445
+ "ece": {
446
+ "bins": [
447
+ {
448
+ "accuracy": null,
449
+ "hi": 0.1,
450
+ "lo": 0.0,
451
+ "mean_confidence": null,
452
+ "n": 0
453
+ },
454
+ {
455
+ "accuracy": null,
456
+ "hi": 0.2,
457
+ "lo": 0.1,
458
+ "mean_confidence": null,
459
+ "n": 0
460
+ },
461
+ {
462
+ "accuracy": null,
463
+ "hi": 0.3,
464
+ "lo": 0.2,
465
+ "mean_confidence": null,
466
+ "n": 0
467
+ },
468
+ {
469
+ "accuracy": null,
470
+ "hi": 0.4,
471
+ "lo": 0.3,
472
+ "mean_confidence": null,
473
+ "n": 0
474
+ },
475
+ {
476
+ "accuracy": null,
477
+ "hi": 0.5,
478
+ "lo": 0.4,
479
+ "mean_confidence": null,
480
+ "n": 0
481
+ },
482
+ {
483
+ "accuracy": null,
484
+ "hi": 0.6,
485
+ "lo": 0.5,
486
+ "mean_confidence": null,
487
+ "n": 0
488
+ },
489
+ {
490
+ "accuracy": null,
491
+ "hi": 0.7,
492
+ "lo": 0.6,
493
+ "mean_confidence": null,
494
+ "n": 0
495
+ },
496
+ {
497
+ "accuracy": null,
498
+ "hi": 0.8,
499
+ "lo": 0.7,
500
+ "mean_confidence": null,
501
+ "n": 0
502
+ },
503
+ {
504
+ "accuracy": null,
505
+ "hi": 0.9,
506
+ "lo": 0.8,
507
+ "mean_confidence": null,
508
+ "n": 0
509
+ },
510
+ {
511
+ "accuracy": 1.0,
512
+ "hi": 1.0,
513
+ "lo": 0.9,
514
+ "mean_confidence": 0.9999977598587672,
515
+ "n": 12
516
+ }
517
+ ],
518
+ "ece": 2.2401412328454384e-06,
519
+ "n": 12
520
+ },
521
+ "latency": {
522
+ "n": 12,
523
+ "p50_s": 0.06312976451590657,
524
+ "p95_s": 0.06712341390084475
525
+ },
526
+ "latency_failures": {
527
+ "n": 0,
528
+ "p50_s": null,
529
+ "p95_s": null
530
+ },
531
+ "n_attempted": 12,
532
+ "n_correct": 12,
533
+ "n_planned": 12,
534
+ "n_renormalized": 0,
535
+ "n_scorable": 12,
536
+ "n_valid": 12,
537
+ "operational_success": 1.0,
538
+ "ordinal_mae": null,
539
+ "price_per_1000_decisions_usd": null,
540
+ "schema_validity": 1.0,
541
+ "schema_validity_strict": 1.0
542
+ }
543
+ },
544
+ "price_per_1000_decisions_usd": null,
545
+ "probability_sources": [
546
+ "native"
547
+ ],
548
+ "schema_validity": 1.0,
549
+ "schema_validity_strict": 1.0,
550
+ "splits": {
551
+ "public": {
552
+ "accuracy": 1.0,
553
+ "brier_mean": 0.0047407252593300855,
554
+ "calibration_n": 48,
555
+ "cost_basis": [
556
+ "local_gpu_no_provider_tariff"
557
+ ],
558
+ "coverage": 1.0,
559
+ "ece": {
560
+ "bins": [
561
+ {
562
+ "accuracy": null,
563
+ "hi": 0.1,
564
+ "lo": 0.0,
565
+ "mean_confidence": null,
566
+ "n": 0
567
+ },
568
+ {
569
+ "accuracy": null,
570
+ "hi": 0.2,
571
+ "lo": 0.1,
572
+ "mean_confidence": null,
573
+ "n": 0
574
+ },
575
+ {
576
+ "accuracy": null,
577
+ "hi": 0.3,
578
+ "lo": 0.2,
579
+ "mean_confidence": null,
580
+ "n": 0
581
+ },
582
+ {
583
+ "accuracy": null,
584
+ "hi": 0.4,
585
+ "lo": 0.3,
586
+ "mean_confidence": null,
587
+ "n": 0
588
+ },
589
+ {
590
+ "accuracy": null,
591
+ "hi": 0.5,
592
+ "lo": 0.4,
593
+ "mean_confidence": null,
594
+ "n": 0
595
+ },
596
+ {
597
+ "accuracy": null,
598
+ "hi": 0.6,
599
+ "lo": 0.5,
600
+ "mean_confidence": null,
601
+ "n": 0
602
+ },
603
+ {
604
+ "accuracy": null,
605
+ "hi": 0.7,
606
+ "lo": 0.6,
607
+ "mean_confidence": null,
608
+ "n": 0
609
+ },
610
+ {
611
+ "accuracy": 1.0,
612
+ "hi": 0.8,
613
+ "lo": 0.7,
614
+ "mean_confidence": 0.7135202288627625,
615
+ "n": 1
616
+ },
617
+ {
618
+ "accuracy": 1.0,
619
+ "hi": 0.9,
620
+ "lo": 0.8,
621
+ "mean_confidence": 0.8538632392883301,
622
+ "n": 1
623
+ },
624
+ {
625
+ "accuracy": 1.0,
626
+ "hi": 1.0,
627
+ "lo": 0.9,
628
+ "mean_confidence": 0.9924926835557689,
629
+ "n": 46
630
+ }
631
+ ],
632
+ "ece": 0.016207356005907027,
633
+ "n": 48
634
+ },
635
+ "latency": {
636
+ "n": 48,
637
+ "p50_s": 0.06309161661192775,
638
+ "p95_s": 0.0954449666896835
639
+ },
640
+ "latency_failures": {
641
+ "n": 0,
642
+ "p50_s": null,
643
+ "p95_s": null
644
+ },
645
+ "n_attempted": 48,
646
+ "n_correct": 48,
647
+ "n_planned": 48,
648
+ "n_renormalized": 0,
649
+ "n_scorable": 48,
650
+ "n_valid": 48,
651
+ "operational_success": 1.0,
652
+ "ordinal_mae": null,
653
+ "price_per_1000_decisions_usd": null,
654
+ "schema_validity": 1.0,
655
+ "schema_validity_strict": 1.0
656
+ }
657
+ },
658
+ "version": "jevbench-v1"
659
+ },
660
+ "hard": {
661
+ "accuracy": 0.43243243243243246,
662
+ "brier_mean": 0.7909445007358973,
663
+ "calibration_n": 111,
664
+ "complete": true,
665
+ "cost_basis": [
666
+ "local_gpu_no_provider_tariff"
667
+ ],
668
+ "coverage": 1.0,
669
+ "dataset_hash": "e816e106f562dec75a8da12e69b75e957a4bbd9f6300f1145fc71837327a3c61",
670
+ "ece": {
671
+ "bins": [
672
+ {
673
+ "accuracy": null,
674
+ "hi": 0.1,
675
+ "lo": 0.0,
676
+ "mean_confidence": null,
677
+ "n": 0
678
+ },
679
+ {
680
+ "accuracy": null,
681
+ "hi": 0.2,
682
+ "lo": 0.1,
683
+ "mean_confidence": null,
684
+ "n": 0
685
+ },
686
+ {
687
+ "accuracy": 0.42857142857142855,
688
+ "hi": 0.3,
689
+ "lo": 0.2,
690
+ "mean_confidence": 0.27341367091451374,
691
+ "n": 7
692
+ },
693
+ {
694
+ "accuracy": 0.3076923076923077,
695
+ "hi": 0.4,
696
+ "lo": 0.3,
697
+ "mean_confidence": 0.36204201441544753,
698
+ "n": 13
699
+ },
700
+ {
701
+ "accuracy": 0.4166666666666667,
702
+ "hi": 0.5,
703
+ "lo": 0.4,
704
+ "mean_confidence": 0.4574062079191208,
705
+ "n": 12
706
+ },
707
+ {
708
+ "accuracy": 0.2857142857142857,
709
+ "hi": 0.6,
710
+ "lo": 0.5,
711
+ "mean_confidence": 0.5474048810345786,
712
+ "n": 14
713
+ },
714
+ {
715
+ "accuracy": 0.5454545454545454,
716
+ "hi": 0.7,
717
+ "lo": 0.6,
718
+ "mean_confidence": 0.6413787711750377,
719
+ "n": 11
720
+ },
721
+ {
722
+ "accuracy": 0.5454545454545454,
723
+ "hi": 0.8,
724
+ "lo": 0.7,
725
+ "mean_confidence": 0.743556255644018,
726
+ "n": 11
727
+ },
728
+ {
729
+ "accuracy": 0.4117647058823529,
730
+ "hi": 0.9,
731
+ "lo": 0.8,
732
+ "mean_confidence": 0.8476672382915721,
733
+ "n": 17
734
+ },
735
+ {
736
+ "accuracy": 0.5,
737
+ "hi": 1.0,
738
+ "lo": 0.9,
739
+ "mean_confidence": 0.9562405920945681,
740
+ "n": 26
741
+ }
742
+ ],
743
+ "ece": 0.25632501856700796,
744
+ "n": 111
745
+ },
746
+ "latency": {
747
+ "n": 111,
748
+ "p50_s": 0.09745004074648023,
749
+ "p95_s": 0.11606183066032827
750
+ },
751
+ "latency_failures": {
752
+ "n": 0,
753
+ "p50_s": null,
754
+ "p95_s": null
755
+ },
756
+ "ledger_charged_usd": null,
757
+ "macro_accuracy": 0.4864605140301735,
758
+ "model_identities": [
759
+ "pcdm:native:runs/ts1b"
760
+ ],
761
+ "n_attempted": 111,
762
+ "n_correct": 48,
763
+ "n_planned": 111,
764
+ "n_renormalized": 0,
765
+ "n_scorable": 111,
766
+ "n_valid": 111,
767
+ "operational_success": 1.0,
768
+ "ordinal_mae": 0.682668519516786,
769
+ "paraphrase_consistency": {
770
+ "agree": 0,
771
+ "agreement": null,
772
+ "both_correct_rate_all_pairs": null,
773
+ "both_valid": 0,
774
+ "pairs": 0
775
+ },
776
+ "per_family": {
777
+ "adversarial": {
778
+ "accuracy": 0.3333333333333333,
779
+ "brier_mean": 0.8822798833915354,
780
+ "calibration_n": 6,
781
+ "cost_basis": [
782
+ "local_gpu_no_provider_tariff"
783
+ ],
784
+ "coverage": 1.0,
785
+ "ece": {
786
+ "bins": [
787
+ {
788
+ "accuracy": null,
789
+ "hi": 0.1,
790
+ "lo": 0.0,
791
+ "mean_confidence": null,
792
+ "n": 0
793
+ },
794
+ {
795
+ "accuracy": null,
796
+ "hi": 0.2,
797
+ "lo": 0.1,
798
+ "mean_confidence": null,
799
+ "n": 0
800
+ },
801
+ {
802
+ "accuracy": null,
803
+ "hi": 0.3,
804
+ "lo": 0.2,
805
+ "mean_confidence": null,
806
+ "n": 0
807
+ },
808
+ {
809
+ "accuracy": 0.0,
810
+ "hi": 0.4,
811
+ "lo": 0.3,
812
+ "mean_confidence": 0.392834335565567,
813
+ "n": 1
814
+ },
815
+ {
816
+ "accuracy": null,
817
+ "hi": 0.5,
818
+ "lo": 0.4,
819
+ "mean_confidence": null,
820
+ "n": 0
821
+ },
822
+ {
823
+ "accuracy": null,
824
+ "hi": 0.6,
825
+ "lo": 0.5,
826
+ "mean_confidence": null,
827
+ "n": 0
828
+ },
829
+ {
830
+ "accuracy": 0.0,
831
+ "hi": 0.7,
832
+ "lo": 0.6,
833
+ "mean_confidence": 0.6013781428337097,
834
+ "n": 1
835
+ },
836
+ {
837
+ "accuracy": null,
838
+ "hi": 0.8,
839
+ "lo": 0.7,
840
+ "mean_confidence": null,
841
+ "n": 0
842
+ },
843
+ {
844
+ "accuracy": 1.0,
845
+ "hi": 0.9,
846
+ "lo": 0.8,
847
+ "mean_confidence": 0.8074021339416504,
848
+ "n": 1
849
+ },
850
+ {
851
+ "accuracy": 0.3333333333333333,
852
+ "hi": 1.0,
853
+ "lo": 0.9,
854
+ "mean_confidence": 0.9713454445203146,
855
+ "n": 3
856
+ }
857
+ ],
858
+ "ece": 0.5168077796697617,
859
+ "n": 6
860
+ },
861
+ "latency": {
862
+ "n": 6,
863
+ "p50_s": 0.09237208357080817,
864
+ "p95_s": 0.0955617759609595
865
+ },
866
+ "latency_failures": {
867
+ "n": 0,
868
+ "p50_s": null,
869
+ "p95_s": null
870
+ },
871
+ "n_attempted": 6,
872
+ "n_correct": 2,
873
+ "n_planned": 6,
874
+ "n_renormalized": 0,
875
+ "n_scorable": 6,
876
+ "n_valid": 6,
877
+ "operational_success": 1.0,
878
+ "ordinal_mae": 0.5170895606279373,
879
+ "price_per_1000_decisions_usd": null,
880
+ "schema_validity": 1.0,
881
+ "schema_validity_strict": 1.0
882
+ },
883
+ "ambiguous": {
884
+ "accuracy": 0.42857142857142855,
885
+ "brier_mean": 0.7734691553069819,
886
+ "calibration_n": 7,
887
+ "cost_basis": [
888
+ "local_gpu_no_provider_tariff"
889
+ ],
890
+ "coverage": 1.0,
891
+ "ece": {
892
+ "bins": [
893
+ {
894
+ "accuracy": null,
895
+ "hi": 0.1,
896
+ "lo": 0.0,
897
+ "mean_confidence": null,
898
+ "n": 0
899
+ },
900
+ {
901
+ "accuracy": null,
902
+ "hi": 0.2,
903
+ "lo": 0.1,
904
+ "mean_confidence": null,
905
+ "n": 0
906
+ },
907
+ {
908
+ "accuracy": null,
909
+ "hi": 0.3,
910
+ "lo": 0.2,
911
+ "mean_confidence": null,
912
+ "n": 0
913
+ },
914
+ {
915
+ "accuracy": null,
916
+ "hi": 0.4,
917
+ "lo": 0.3,
918
+ "mean_confidence": null,
919
+ "n": 0
920
+ },
921
+ {
922
+ "accuracy": 0.0,
923
+ "hi": 0.5,
924
+ "lo": 0.4,
925
+ "mean_confidence": 0.43299877643585205,
926
+ "n": 1
927
+ },
928
+ {
929
+ "accuracy": null,
930
+ "hi": 0.6,
931
+ "lo": 0.5,
932
+ "mean_confidence": null,
933
+ "n": 0
934
+ },
935
+ {
936
+ "accuracy": 1.0,
937
+ "hi": 0.7,
938
+ "lo": 0.6,
939
+ "mean_confidence": 0.6301537156105042,
940
+ "n": 1
941
+ },
942
+ {
943
+ "accuracy": 0.0,
944
+ "hi": 0.8,
945
+ "lo": 0.7,
946
+ "mean_confidence": 0.751984715461731,
947
+ "n": 1
948
+ },
949
+ {
950
+ "accuracy": 0.3333333333333333,
951
+ "hi": 0.9,
952
+ "lo": 0.8,
953
+ "mean_confidence": 0.8551861643791199,
954
+ "n": 3
955
+ },
956
+ {
957
+ "accuracy": 1.0,
958
+ "hi": 1.0,
959
+ "lo": 0.9,
960
+ "mean_confidence": 0.9342254996299744,
961
+ "n": 1
962
+ }
963
+ ],
964
+ "ece": 0.4551661099706378,
965
+ "n": 7
966
+ },
967
+ "latency": {
968
+ "n": 7,
969
+ "p50_s": 0.0957035981118679,
970
+ "p95_s": 0.10261358409188688
971
+ },
972
+ "latency_failures": {
973
+ "n": 0,
974
+ "p50_s": null,
975
+ "p95_s": null
976
+ },
977
+ "n_attempted": 7,
978
+ "n_correct": 3,
979
+ "n_planned": 7,
980
+ "n_renormalized": 0,
981
+ "n_scorable": 7,
982
+ "n_valid": 7,
983
+ "operational_success": 1.0,
984
+ "ordinal_mae": null,
985
+ "price_per_1000_decisions_usd": null,
986
+ "schema_validity": 1.0,
987
+ "schema_validity_strict": 1.0
988
+ },
989
+ "judge_hard": {
990
+ "accuracy": 0.4117647058823529,
991
+ "brier_mean": 0.9550899406207743,
992
+ "calibration_n": 17,
993
+ "cost_basis": [
994
+ "local_gpu_no_provider_tariff"
995
+ ],
996
+ "coverage": 1.0,
997
+ "ece": {
998
+ "bins": [
999
+ {
1000
+ "accuracy": null,
1001
+ "hi": 0.1,
1002
+ "lo": 0.0,
1003
+ "mean_confidence": null,
1004
+ "n": 0
1005
+ },
1006
+ {
1007
+ "accuracy": null,
1008
+ "hi": 0.2,
1009
+ "lo": 0.1,
1010
+ "mean_confidence": null,
1011
+ "n": 0
1012
+ },
1013
+ {
1014
+ "accuracy": null,
1015
+ "hi": 0.3,
1016
+ "lo": 0.2,
1017
+ "mean_confidence": null,
1018
+ "n": 0
1019
+ },
1020
+ {
1021
+ "accuracy": null,
1022
+ "hi": 0.4,
1023
+ "lo": 0.3,
1024
+ "mean_confidence": null,
1025
+ "n": 0
1026
+ },
1027
+ {
1028
+ "accuracy": null,
1029
+ "hi": 0.5,
1030
+ "lo": 0.4,
1031
+ "mean_confidence": null,
1032
+ "n": 0
1033
+ },
1034
+ {
1035
+ "accuracy": 0.0,
1036
+ "hi": 0.6,
1037
+ "lo": 0.5,
1038
+ "mean_confidence": 0.5444756150245667,
1039
+ "n": 1
1040
+ },
1041
+ {
1042
+ "accuracy": null,
1043
+ "hi": 0.7,
1044
+ "lo": 0.6,
1045
+ "mean_confidence": null,
1046
+ "n": 0
1047
+ },
1048
+ {
1049
+ "accuracy": null,
1050
+ "hi": 0.8,
1051
+ "lo": 0.7,
1052
+ "mean_confidence": null,
1053
+ "n": 0
1054
+ },
1055
+ {
1056
+ "accuracy": 0.3333333333333333,
1057
+ "hi": 0.9,
1058
+ "lo": 0.8,
1059
+ "mean_confidence": 0.8754598399003347,
1060
+ "n": 6
1061
+ },
1062
+ {
1063
+ "accuracy": 0.5,
1064
+ "hi": 1.0,
1065
+ "lo": 0.9,
1066
+ "mean_confidence": 0.9572476387023926,
1067
+ "n": 10
1068
+ }
1069
+ ],
1070
+ "ece": 0.49233594361473537,
1071
+ "n": 17
1072
+ },
1073
+ "latency": {
1074
+ "n": 17,
1075
+ "p50_s": 0.09264156315475702,
1076
+ "p95_s": 0.0934916635043919
1077
+ },
1078
+ "latency_failures": {
1079
+ "n": 0,
1080
+ "p50_s": null,
1081
+ "p95_s": null
1082
+ },
1083
+ "n_attempted": 17,
1084
+ "n_correct": 7,
1085
+ "n_planned": 17,
1086
+ "n_renormalized": 0,
1087
+ "n_scorable": 17,
1088
+ "n_valid": 17,
1089
+ "operational_success": 1.0,
1090
+ "ordinal_mae": null,
1091
+ "price_per_1000_decisions_usd": null,
1092
+ "schema_validity": 1.0,
1093
+ "schema_validity_strict": 1.0
1094
+ },
1095
+ "long_policy": {
1096
+ "accuracy": 0.2631578947368421,
1097
+ "brier_mean": 0.8194034386815301,
1098
+ "calibration_n": 19,
1099
+ "cost_basis": [
1100
+ "local_gpu_no_provider_tariff"
1101
+ ],
1102
+ "coverage": 1.0,
1103
+ "ece": {
1104
+ "bins": [
1105
+ {
1106
+ "accuracy": null,
1107
+ "hi": 0.1,
1108
+ "lo": 0.0,
1109
+ "mean_confidence": null,
1110
+ "n": 0
1111
+ },
1112
+ {
1113
+ "accuracy": null,
1114
+ "hi": 0.2,
1115
+ "lo": 0.1,
1116
+ "mean_confidence": null,
1117
+ "n": 0
1118
+ },
1119
+ {
1120
+ "accuracy": 0.3333333333333333,
1121
+ "hi": 0.3,
1122
+ "lo": 0.2,
1123
+ "mean_confidence": 0.2753242254257202,
1124
+ "n": 3
1125
+ },
1126
+ {
1127
+ "accuracy": 0.16666666666666666,
1128
+ "hi": 0.4,
1129
+ "lo": 0.3,
1130
+ "mean_confidence": 0.3495459308226903,
1131
+ "n": 6
1132
+ },
1133
+ {
1134
+ "accuracy": 0.0,
1135
+ "hi": 0.5,
1136
+ "lo": 0.4,
1137
+ "mean_confidence": 0.44854867458343506,
1138
+ "n": 1
1139
+ },
1140
+ {
1141
+ "accuracy": 0.3333333333333333,
1142
+ "hi": 0.6,
1143
+ "lo": 0.5,
1144
+ "mean_confidence": 0.5452876488367716,
1145
+ "n": 3
1146
+ },
1147
+ {
1148
+ "accuracy": 0.0,
1149
+ "hi": 0.7,
1150
+ "lo": 0.6,
1151
+ "mean_confidence": 0.6139329075813293,
1152
+ "n": 1
1153
+ },
1154
+ {
1155
+ "accuracy": 0.5,
1156
+ "hi": 0.8,
1157
+ "lo": 0.7,
1158
+ "mean_confidence": 0.7318354845046997,
1159
+ "n": 2
1160
+ },
1161
+ {
1162
+ "accuracy": null,
1163
+ "hi": 0.9,
1164
+ "lo": 0.8,
1165
+ "mean_confidence": null,
1166
+ "n": 0
1167
+ },
1168
+ {
1169
+ "accuracy": 0.3333333333333333,
1170
+ "hi": 1.0,
1171
+ "lo": 0.9,
1172
+ "mean_confidence": 0.9440843065579733,
1173
+ "n": 3
1174
+ }
1175
+ ],
1176
+ "ece": 0.27713533294828313,
1177
+ "n": 19
1178
+ },
1179
+ "latency": {
1180
+ "n": 19,
1181
+ "p50_s": 0.11336228996515274,
1182
+ "p95_s": 0.12370569789782163
1183
+ },
1184
+ "latency_failures": {
1185
+ "n": 0,
1186
+ "p50_s": null,
1187
+ "p95_s": null
1188
+ },
1189
+ "n_attempted": 19,
1190
+ "n_correct": 5,
1191
+ "n_planned": 19,
1192
+ "n_renormalized": 0,
1193
+ "n_scorable": 19,
1194
+ "n_valid": 19,
1195
+ "operational_success": 1.0,
1196
+ "ordinal_mae": 0.4551819935441017,
1197
+ "price_per_1000_decisions_usd": null,
1198
+ "schema_validity": 1.0,
1199
+ "schema_validity_strict": 1.0
1200
+ },
1201
+ "multi_hop": {
1202
+ "accuracy": 0.4444444444444444,
1203
+ "brier_mean": 0.7247968278780245,
1204
+ "calibration_n": 18,
1205
+ "cost_basis": [
1206
+ "local_gpu_no_provider_tariff"
1207
+ ],
1208
+ "coverage": 1.0,
1209
+ "ece": {
1210
+ "bins": [
1211
+ {
1212
+ "accuracy": null,
1213
+ "hi": 0.1,
1214
+ "lo": 0.0,
1215
+ "mean_confidence": null,
1216
+ "n": 0
1217
+ },
1218
+ {
1219
+ "accuracy": null,
1220
+ "hi": 0.2,
1221
+ "lo": 0.1,
1222
+ "mean_confidence": null,
1223
+ "n": 0
1224
+ },
1225
+ {
1226
+ "accuracy": 0.5,
1227
+ "hi": 0.3,
1228
+ "lo": 0.2,
1229
+ "mean_confidence": 0.2637081891298294,
1230
+ "n": 2
1231
+ },
1232
+ {
1233
+ "accuracy": 0.6666666666666666,
1234
+ "hi": 0.4,
1235
+ "lo": 0.3,
1236
+ "mean_confidence": 0.3658106327056885,
1237
+ "n": 3
1238
+ },
1239
+ {
1240
+ "accuracy": 0.2,
1241
+ "hi": 0.5,
1242
+ "lo": 0.4,
1243
+ "mean_confidence": 0.47336041927337646,
1244
+ "n": 5
1245
+ },
1246
+ {
1247
+ "accuracy": 0.0,
1248
+ "hi": 0.6,
1249
+ "lo": 0.5,
1250
+ "mean_confidence": 0.5131076773007711,
1251
+ "n": 3
1252
+ },
1253
+ {
1254
+ "accuracy": 1.0,
1255
+ "hi": 0.7,
1256
+ "lo": 0.6,
1257
+ "mean_confidence": 0.6799950003623962,
1258
+ "n": 1
1259
+ },
1260
+ {
1261
+ "accuracy": 0.5,
1262
+ "hi": 0.8,
1263
+ "lo": 0.7,
1264
+ "mean_confidence": 0.7172561585903168,
1265
+ "n": 2
1266
+ },
1267
+ {
1268
+ "accuracy": 1.0,
1269
+ "hi": 0.9,
1270
+ "lo": 0.8,
1271
+ "mean_confidence": 0.8304344415664673,
1272
+ "n": 2
1273
+ },
1274
+ {
1275
+ "accuracy": null,
1276
+ "hi": 1.0,
1277
+ "lo": 0.9,
1278
+ "mean_confidence": null,
1279
+ "n": 0
1280
+ }
1281
+ ],
1282
+ "ece": 0.29860696030987627,
1283
+ "n": 18
1284
+ },
1285
+ "latency": {
1286
+ "n": 18,
1287
+ "p50_s": 0.10305563476867974,
1288
+ "p95_s": 0.11012194301001728
1289
+ },
1290
+ "latency_failures": {
1291
+ "n": 0,
1292
+ "p50_s": null,
1293
+ "p95_s": null
1294
+ },
1295
+ "n_attempted": 18,
1296
+ "n_correct": 8,
1297
+ "n_planned": 18,
1298
+ "n_renormalized": 0,
1299
+ "n_scorable": 18,
1300
+ "n_valid": 18,
1301
+ "operational_success": 1.0,
1302
+ "ordinal_mae": 0.16368193924427032,
1303
+ "price_per_1000_decisions_usd": null,
1304
+ "schema_validity": 1.0,
1305
+ "schema_validity_strict": 1.0
1306
+ },
1307
+ "probability": {
1308
+ "accuracy": 0.4,
1309
+ "brier_mean": 0.9394113544796729,
1310
+ "calibration_n": 10,
1311
+ "cost_basis": [
1312
+ "local_gpu_no_provider_tariff"
1313
+ ],
1314
+ "coverage": 1.0,
1315
+ "ece": {
1316
+ "bins": [
1317
+ {
1318
+ "accuracy": null,
1319
+ "hi": 0.1,
1320
+ "lo": 0.0,
1321
+ "mean_confidence": null,
1322
+ "n": 0
1323
+ },
1324
+ {
1325
+ "accuracy": null,
1326
+ "hi": 0.2,
1327
+ "lo": 0.1,
1328
+ "mean_confidence": null,
1329
+ "n": 0
1330
+ },
1331
+ {
1332
+ "accuracy": null,
1333
+ "hi": 0.3,
1334
+ "lo": 0.2,
1335
+ "mean_confidence": null,
1336
+ "n": 0
1337
+ },
1338
+ {
1339
+ "accuracy": null,
1340
+ "hi": 0.4,
1341
+ "lo": 0.3,
1342
+ "mean_confidence": null,
1343
+ "n": 0
1344
+ },
1345
+ {
1346
+ "accuracy": 1.0,
1347
+ "hi": 0.5,
1348
+ "lo": 0.4,
1349
+ "mean_confidence": 0.4608645737171173,
1350
+ "n": 1
1351
+ },
1352
+ {
1353
+ "accuracy": 0.5,
1354
+ "hi": 0.6,
1355
+ "lo": 0.5,
1356
+ "mean_confidence": 0.5587047636508942,
1357
+ "n": 2
1358
+ },
1359
+ {
1360
+ "accuracy": 0.3333333333333333,
1361
+ "hi": 0.7,
1362
+ "lo": 0.6,
1363
+ "mean_confidence": 0.6281190514564514,
1364
+ "n": 3
1365
+ },
1366
+ {
1367
+ "accuracy": 1.0,
1368
+ "hi": 0.8,
1369
+ "lo": 0.7,
1370
+ "mean_confidence": 0.708816409111023,
1371
+ "n": 1
1372
+ },
1373
+ {
1374
+ "accuracy": 0.0,
1375
+ "hi": 0.9,
1376
+ "lo": 0.8,
1377
+ "mean_confidence": 0.8403750658035278,
1378
+ "n": 2
1379
+ },
1380
+ {
1381
+ "accuracy": 0.0,
1382
+ "hi": 1.0,
1383
+ "lo": 0.9,
1384
+ "mean_confidence": 0.9059809446334839,
1385
+ "n": 1
1386
+ }
1387
+ ],
1388
+ "ece": 0.4418816775083542,
1389
+ "n": 10
1390
+ },
1391
+ "latency": {
1392
+ "n": 10,
1393
+ "p50_s": 0.09962555137462914,
1394
+ "p95_s": 0.10702485411893575
1395
+ },
1396
+ "latency_failures": {
1397
+ "n": 0,
1398
+ "p50_s": null,
1399
+ "p95_s": null
1400
+ },
1401
+ "n_attempted": 10,
1402
+ "n_correct": 4,
1403
+ "n_planned": 10,
1404
+ "n_renormalized": 0,
1405
+ "n_scorable": 10,
1406
+ "n_valid": 10,
1407
+ "operational_success": 1.0,
1408
+ "ordinal_mae": null,
1409
+ "price_per_1000_decisions_usd": null,
1410
+ "schema_validity": 1.0,
1411
+ "schema_validity_strict": 1.0
1412
+ },
1413
+ "routing_hard": {
1414
+ "accuracy": 1.0,
1415
+ "brier_mean": 0.13808678499012011,
1416
+ "calibration_n": 5,
1417
+ "cost_basis": [
1418
+ "local_gpu_no_provider_tariff"
1419
+ ],
1420
+ "coverage": 1.0,
1421
+ "ece": {
1422
+ "bins": [
1423
+ {
1424
+ "accuracy": null,
1425
+ "hi": 0.1,
1426
+ "lo": 0.0,
1427
+ "mean_confidence": null,
1428
+ "n": 0
1429
+ },
1430
+ {
1431
+ "accuracy": null,
1432
+ "hi": 0.2,
1433
+ "lo": 0.1,
1434
+ "mean_confidence": null,
1435
+ "n": 0
1436
+ },
1437
+ {
1438
+ "accuracy": null,
1439
+ "hi": 0.3,
1440
+ "lo": 0.2,
1441
+ "mean_confidence": null,
1442
+ "n": 0
1443
+ },
1444
+ {
1445
+ "accuracy": null,
1446
+ "hi": 0.4,
1447
+ "lo": 0.3,
1448
+ "mean_confidence": null,
1449
+ "n": 0
1450
+ },
1451
+ {
1452
+ "accuracy": 1.0,
1453
+ "hi": 0.5,
1454
+ "lo": 0.4,
1455
+ "mean_confidence": 0.43413904309272766,
1456
+ "n": 1
1457
+ },
1458
+ {
1459
+ "accuracy": 1.0,
1460
+ "hi": 0.6,
1461
+ "lo": 0.5,
1462
+ "mean_confidence": 0.5734199285507202,
1463
+ "n": 1
1464
+ },
1465
+ {
1466
+ "accuracy": null,
1467
+ "hi": 0.7,
1468
+ "lo": 0.6,
1469
+ "mean_confidence": null,
1470
+ "n": 0
1471
+ },
1472
+ {
1473
+ "accuracy": null,
1474
+ "hi": 0.8,
1475
+ "lo": 0.7,
1476
+ "mean_confidence": null,
1477
+ "n": 0
1478
+ },
1479
+ {
1480
+ "accuracy": null,
1481
+ "hi": 0.9,
1482
+ "lo": 0.8,
1483
+ "mean_confidence": null,
1484
+ "n": 0
1485
+ },
1486
+ {
1487
+ "accuracy": 1.0,
1488
+ "hi": 1.0,
1489
+ "lo": 0.9,
1490
+ "mean_confidence": 0.9703327616055807,
1491
+ "n": 3
1492
+ }
1493
+ ],
1494
+ "ece": 0.21628854870796202,
1495
+ "n": 5
1496
+ },
1497
+ "latency": {
1498
+ "n": 5,
1499
+ "p50_s": 0.09371192893013358,
1500
+ "p95_s": 0.09542652918025851
1501
+ },
1502
+ "latency_failures": {
1503
+ "n": 0,
1504
+ "p50_s": null,
1505
+ "p95_s": null
1506
+ },
1507
+ "n_attempted": 5,
1508
+ "n_correct": 5,
1509
+ "n_planned": 5,
1510
+ "n_renormalized": 0,
1511
+ "n_scorable": 5,
1512
+ "n_valid": 5,
1513
+ "operational_success": 1.0,
1514
+ "ordinal_mae": null,
1515
+ "price_per_1000_decisions_usd": null,
1516
+ "schema_validity": 1.0,
1517
+ "schema_validity_strict": 1.0
1518
+ },
1519
+ "temporal_numeric": {
1520
+ "accuracy": 0.3333333333333333,
1521
+ "brier_mean": 0.8541664497755332,
1522
+ "calibration_n": 15,
1523
+ "cost_basis": [
1524
+ "local_gpu_no_provider_tariff"
1525
+ ],
1526
+ "coverage": 1.0,
1527
+ "ece": {
1528
+ "bins": [
1529
+ {
1530
+ "accuracy": null,
1531
+ "hi": 0.1,
1532
+ "lo": 0.0,
1533
+ "mean_confidence": null,
1534
+ "n": 0
1535
+ },
1536
+ {
1537
+ "accuracy": null,
1538
+ "hi": 0.2,
1539
+ "lo": 0.1,
1540
+ "mean_confidence": null,
1541
+ "n": 0
1542
+ },
1543
+ {
1544
+ "accuracy": 0.5,
1545
+ "hi": 0.3,
1546
+ "lo": 0.2,
1547
+ "mean_confidence": 0.2802533209323883,
1548
+ "n": 2
1549
+ },
1550
+ {
1551
+ "accuracy": 0.3333333333333333,
1552
+ "hi": 0.4,
1553
+ "lo": 0.3,
1554
+ "mean_confidence": 0.37300145626068115,
1555
+ "n": 3
1556
+ },
1557
+ {
1558
+ "accuracy": 0.5,
1559
+ "hi": 0.5,
1560
+ "lo": 0.4,
1561
+ "mean_confidence": 0.43682731688022614,
1562
+ "n": 2
1563
+ },
1564
+ {
1565
+ "accuracy": 0.0,
1566
+ "hi": 0.6,
1567
+ "lo": 0.5,
1568
+ "mean_confidence": 0.5656636953353882,
1569
+ "n": 3
1570
+ },
1571
+ {
1572
+ "accuracy": 0.5,
1573
+ "hi": 0.7,
1574
+ "lo": 0.6,
1575
+ "mean_confidence": 0.6869688034057617,
1576
+ "n": 2
1577
+ },
1578
+ {
1579
+ "accuracy": 0.0,
1580
+ "hi": 0.8,
1581
+ "lo": 0.7,
1582
+ "mean_confidence": 0.7070299386978149,
1583
+ "n": 1
1584
+ },
1585
+ {
1586
+ "accuracy": 1.0,
1587
+ "hi": 0.9,
1588
+ "lo": 0.8,
1589
+ "mean_confidence": 0.8164684176445007,
1590
+ "n": 1
1591
+ },
1592
+ {
1593
+ "accuracy": 0.0,
1594
+ "hi": 1.0,
1595
+ "lo": 0.9,
1596
+ "mean_confidence": 0.973138689994812,
1597
+ "n": 1
1598
+ }
1599
+ ],
1600
+ "ece": 0.3079647998015086,
1601
+ "n": 15
1602
+ },
1603
+ "latency": {
1604
+ "n": 15,
1605
+ "p50_s": 0.09936860809102654,
1606
+ "p95_s": 0.10479145389981567
1607
+ },
1608
+ "latency_failures": {
1609
+ "n": 0,
1610
+ "p50_s": null,
1611
+ "p95_s": null
1612
+ },
1613
+ "n_attempted": 15,
1614
+ "n_correct": 5,
1615
+ "n_planned": 15,
1616
+ "n_renormalized": 0,
1617
+ "n_scorable": 15,
1618
+ "n_valid": 15,
1619
+ "operational_success": 1.0,
1620
+ "ordinal_mae": 1.607450231909752,
1621
+ "price_per_1000_decisions_usd": null,
1622
+ "schema_validity": 1.0,
1623
+ "schema_validity_strict": 1.0
1624
+ },
1625
+ "tradeoff": {
1626
+ "accuracy": 0.5,
1627
+ "brier_mean": 0.8777460761269166,
1628
+ "calibration_n": 6,
1629
+ "cost_basis": [
1630
+ "local_gpu_no_provider_tariff"
1631
+ ],
1632
+ "coverage": 1.0,
1633
+ "ece": {
1634
+ "bins": [
1635
+ {
1636
+ "accuracy": null,
1637
+ "hi": 0.1,
1638
+ "lo": 0.0,
1639
+ "mean_confidence": null,
1640
+ "n": 0
1641
+ },
1642
+ {
1643
+ "accuracy": null,
1644
+ "hi": 0.2,
1645
+ "lo": 0.1,
1646
+ "mean_confidence": null,
1647
+ "n": 0
1648
+ },
1649
+ {
1650
+ "accuracy": null,
1651
+ "hi": 0.3,
1652
+ "lo": 0.2,
1653
+ "mean_confidence": null,
1654
+ "n": 0
1655
+ },
1656
+ {
1657
+ "accuracy": null,
1658
+ "hi": 0.4,
1659
+ "lo": 0.3,
1660
+ "mean_confidence": null,
1661
+ "n": 0
1662
+ },
1663
+ {
1664
+ "accuracy": null,
1665
+ "hi": 0.5,
1666
+ "lo": 0.4,
1667
+ "mean_confidence": null,
1668
+ "n": 0
1669
+ },
1670
+ {
1671
+ "accuracy": 1.0,
1672
+ "hi": 0.6,
1673
+ "lo": 0.5,
1674
+ "mean_confidence": 0.5561861991882324,
1675
+ "n": 1
1676
+ },
1677
+ {
1678
+ "accuracy": 1.0,
1679
+ "hi": 0.7,
1680
+ "lo": 0.6,
1681
+ "mean_confidence": 0.646073043346405,
1682
+ "n": 1
1683
+ },
1684
+ {
1685
+ "accuracy": 0.5,
1686
+ "hi": 0.8,
1687
+ "lo": 0.7,
1688
+ "mean_confidence": 0.7889719605445862,
1689
+ "n": 2
1690
+ },
1691
+ {
1692
+ "accuracy": 0.0,
1693
+ "hi": 0.9,
1694
+ "lo": 0.8,
1695
+ "mean_confidence": 0.824085533618927,
1696
+ "n": 1
1697
+ },
1698
+ {
1699
+ "accuracy": 0.0,
1700
+ "hi": 1.0,
1701
+ "lo": 0.9,
1702
+ "mean_confidence": 0.9255269169807434,
1703
+ "n": 1
1704
+ }
1705
+ ],
1706
+ "ece": 0.5208828548590341,
1707
+ "n": 6
1708
+ },
1709
+ "latency": {
1710
+ "n": 6,
1711
+ "p50_s": 0.09490847331471741,
1712
+ "p95_s": 0.10340702254325151
1713
+ },
1714
+ "latency_failures": {
1715
+ "n": 0,
1716
+ "p50_s": null,
1717
+ "p95_s": null
1718
+ },
1719
+ "n_attempted": 6,
1720
+ "n_correct": 3,
1721
+ "n_planned": 6,
1722
+ "n_renormalized": 0,
1723
+ "n_scorable": 6,
1724
+ "n_valid": 6,
1725
+ "operational_success": 1.0,
1726
+ "ordinal_mae": null,
1727
+ "price_per_1000_decisions_usd": null,
1728
+ "schema_validity": 1.0,
1729
+ "schema_validity_strict": 1.0
1730
+ },
1731
+ "trap": {
1732
+ "accuracy": 0.75,
1733
+ "brier_mean": 0.5089772867172517,
1734
+ "calibration_n": 8,
1735
+ "cost_basis": [
1736
+ "local_gpu_no_provider_tariff"
1737
+ ],
1738
+ "coverage": 1.0,
1739
+ "ece": {
1740
+ "bins": [
1741
+ {
1742
+ "accuracy": null,
1743
+ "hi": 0.1,
1744
+ "lo": 0.0,
1745
+ "mean_confidence": null,
1746
+ "n": 0
1747
+ },
1748
+ {
1749
+ "accuracy": null,
1750
+ "hi": 0.2,
1751
+ "lo": 0.1,
1752
+ "mean_confidence": null,
1753
+ "n": 0
1754
+ },
1755
+ {
1756
+ "accuracy": null,
1757
+ "hi": 0.3,
1758
+ "lo": 0.2,
1759
+ "mean_confidence": null,
1760
+ "n": 0
1761
+ },
1762
+ {
1763
+ "accuracy": null,
1764
+ "hi": 0.4,
1765
+ "lo": 0.3,
1766
+ "mean_confidence": null,
1767
+ "n": 0
1768
+ },
1769
+ {
1770
+ "accuracy": 1.0,
1771
+ "hi": 0.5,
1772
+ "lo": 0.4,
1773
+ "mean_confidence": 0.4718666970729828,
1774
+ "n": 1
1775
+ },
1776
+ {
1777
+ "accuracy": null,
1778
+ "hi": 0.6,
1779
+ "lo": 0.5,
1780
+ "mean_confidence": null,
1781
+ "n": 0
1782
+ },
1783
+ {
1784
+ "accuracy": 1.0,
1785
+ "hi": 0.7,
1786
+ "lo": 0.6,
1787
+ "mean_confidence": 0.6253389120101929,
1788
+ "n": 1
1789
+ },
1790
+ {
1791
+ "accuracy": 1.0,
1792
+ "hi": 0.8,
1793
+ "lo": 0.7,
1794
+ "mean_confidence": 0.7675802707672119,
1795
+ "n": 2
1796
+ },
1797
+ {
1798
+ "accuracy": 0.0,
1799
+ "hi": 0.9,
1800
+ "lo": 0.8,
1801
+ "mean_confidence": 0.80245041847229,
1802
+ "n": 1
1803
+ },
1804
+ {
1805
+ "accuracy": 0.6666666666666666,
1806
+ "hi": 1.0,
1807
+ "lo": 0.9,
1808
+ "mean_confidence": 0.9645398060480753,
1809
+ "n": 3
1810
+ }
1811
+ ],
1812
+ "ece": 0.3829629607498646,
1813
+ "n": 8
1814
+ },
1815
+ "latency": {
1816
+ "n": 8,
1817
+ "p50_s": 0.09288997249677777,
1818
+ "p95_s": 0.09611767800524831
1819
+ },
1820
+ "latency_failures": {
1821
+ "n": 0,
1822
+ "p50_s": null,
1823
+ "p95_s": null
1824
+ },
1825
+ "n_attempted": 8,
1826
+ "n_correct": 6,
1827
+ "n_planned": 8,
1828
+ "n_renormalized": 0,
1829
+ "n_scorable": 8,
1830
+ "n_valid": 8,
1831
+ "operational_success": 1.0,
1832
+ "ordinal_mae": 0.8974253982305527,
1833
+ "price_per_1000_decisions_usd": null,
1834
+ "schema_validity": 1.0,
1835
+ "schema_validity_strict": 1.0
1836
+ }
1837
+ },
1838
+ "price_per_1000_decisions_usd": null,
1839
+ "probability_sources": [
1840
+ "native"
1841
+ ],
1842
+ "schema_validity": 1.0,
1843
+ "schema_validity_strict": 1.0,
1844
+ "splits": {
1845
+ "public": {
1846
+ "accuracy": 0.43243243243243246,
1847
+ "brier_mean": 0.7909445007358973,
1848
+ "calibration_n": 111,
1849
+ "cost_basis": [
1850
+ "local_gpu_no_provider_tariff"
1851
+ ],
1852
+ "coverage": 1.0,
1853
+ "ece": {
1854
+ "bins": [
1855
+ {
1856
+ "accuracy": null,
1857
+ "hi": 0.1,
1858
+ "lo": 0.0,
1859
+ "mean_confidence": null,
1860
+ "n": 0
1861
+ },
1862
+ {
1863
+ "accuracy": null,
1864
+ "hi": 0.2,
1865
+ "lo": 0.1,
1866
+ "mean_confidence": null,
1867
+ "n": 0
1868
+ },
1869
+ {
1870
+ "accuracy": 0.42857142857142855,
1871
+ "hi": 0.3,
1872
+ "lo": 0.2,
1873
+ "mean_confidence": 0.27341367091451374,
1874
+ "n": 7
1875
+ },
1876
+ {
1877
+ "accuracy": 0.3076923076923077,
1878
+ "hi": 0.4,
1879
+ "lo": 0.3,
1880
+ "mean_confidence": 0.36204201441544753,
1881
+ "n": 13
1882
+ },
1883
+ {
1884
+ "accuracy": 0.4166666666666667,
1885
+ "hi": 0.5,
1886
+ "lo": 0.4,
1887
+ "mean_confidence": 0.4574062079191208,
1888
+ "n": 12
1889
+ },
1890
+ {
1891
+ "accuracy": 0.2857142857142857,
1892
+ "hi": 0.6,
1893
+ "lo": 0.5,
1894
+ "mean_confidence": 0.5474048810345786,
1895
+ "n": 14
1896
+ },
1897
+ {
1898
+ "accuracy": 0.5454545454545454,
1899
+ "hi": 0.7,
1900
+ "lo": 0.6,
1901
+ "mean_confidence": 0.6413787711750377,
1902
+ "n": 11
1903
+ },
1904
+ {
1905
+ "accuracy": 0.5454545454545454,
1906
+ "hi": 0.8,
1907
+ "lo": 0.7,
1908
+ "mean_confidence": 0.743556255644018,
1909
+ "n": 11
1910
+ },
1911
+ {
1912
+ "accuracy": 0.4117647058823529,
1913
+ "hi": 0.9,
1914
+ "lo": 0.8,
1915
+ "mean_confidence": 0.8476672382915721,
1916
+ "n": 17
1917
+ },
1918
+ {
1919
+ "accuracy": 0.5,
1920
+ "hi": 1.0,
1921
+ "lo": 0.9,
1922
+ "mean_confidence": 0.9562405920945681,
1923
+ "n": 26
1924
+ }
1925
+ ],
1926
+ "ece": 0.25632501856700796,
1927
+ "n": 111
1928
+ },
1929
+ "latency": {
1930
+ "n": 111,
1931
+ "p50_s": 0.09745004074648023,
1932
+ "p95_s": 0.11606183066032827
1933
+ },
1934
+ "latency_failures": {
1935
+ "n": 0,
1936
+ "p50_s": null,
1937
+ "p95_s": null
1938
+ },
1939
+ "n_attempted": 111,
1940
+ "n_correct": 48,
1941
+ "n_planned": 111,
1942
+ "n_renormalized": 0,
1943
+ "n_scorable": 111,
1944
+ "n_valid": 111,
1945
+ "operational_success": 1.0,
1946
+ "ordinal_mae": 0.682668519516786,
1947
+ "price_per_1000_decisions_usd": null,
1948
+ "schema_validity": 1.0,
1949
+ "schema_validity_strict": 1.0
1950
+ }
1951
+ },
1952
+ "version": "jevbench-v1"
1953
+ },
1954
+ "original": {
1955
+ "accuracy": 0.6944444444444444,
1956
+ "brier_mean": 0.39750463081241993,
1957
+ "calibration_n": 72,
1958
+ "complete": true,
1959
+ "cost_basis": [
1960
+ "local_gpu_no_provider_tariff"
1961
+ ],
1962
+ "coverage": 1.0,
1963
+ "dataset_hash": "914276022fa5f0ea70e4231f2af98e68c67b274b61bea8c440b33890401cd2f9",
1964
+ "ece": {
1965
+ "bins": [
1966
+ {
1967
+ "accuracy": null,
1968
+ "hi": 0.1,
1969
+ "lo": 0.0,
1970
+ "mean_confidence": null,
1971
+ "n": 0
1972
+ },
1973
+ {
1974
+ "accuracy": null,
1975
+ "hi": 0.2,
1976
+ "lo": 0.1,
1977
+ "mean_confidence": null,
1978
+ "n": 0
1979
+ },
1980
+ {
1981
+ "accuracy": null,
1982
+ "hi": 0.3,
1983
+ "lo": 0.2,
1984
+ "mean_confidence": null,
1985
+ "n": 0
1986
+ },
1987
+ {
1988
+ "accuracy": null,
1989
+ "hi": 0.4,
1990
+ "lo": 0.3,
1991
+ "mean_confidence": null,
1992
+ "n": 0
1993
+ },
1994
+ {
1995
+ "accuracy": 0.4,
1996
+ "hi": 0.5,
1997
+ "lo": 0.4,
1998
+ "mean_confidence": 0.45405682921409607,
1999
+ "n": 5
2000
+ },
2001
+ {
2002
+ "accuracy": 0.25,
2003
+ "hi": 0.6,
2004
+ "lo": 0.5,
2005
+ "mean_confidence": 0.538318544626236,
2006
+ "n": 8
2007
+ },
2008
+ {
2009
+ "accuracy": 0.4,
2010
+ "hi": 0.7,
2011
+ "lo": 0.6,
2012
+ "mean_confidence": 0.6290562272071838,
2013
+ "n": 5
2014
+ },
2015
+ {
2016
+ "accuracy": 0.7272727272727273,
2017
+ "hi": 0.8,
2018
+ "lo": 0.7,
2019
+ "mean_confidence": 0.7492249011993408,
2020
+ "n": 11
2021
+ },
2022
+ {
2023
+ "accuracy": 0.7,
2024
+ "hi": 0.9,
2025
+ "lo": 0.8,
2026
+ "mean_confidence": 0.8450712203979492,
2027
+ "n": 10
2028
+ },
2029
+ {
2030
+ "accuracy": 0.8787878787878788,
2031
+ "hi": 1.0,
2032
+ "lo": 0.9,
2033
+ "mean_confidence": 0.9637182950973511,
2034
+ "n": 33
2035
+ }
2036
+ ],
2037
+ "ece": 0.11412504853473769,
2038
+ "n": 72
2039
+ },
2040
+ "latency": {
2041
+ "n": 72,
2042
+ "p50_s": 0.06066564843058586,
2043
+ "p95_s": 0.12478509345091895
2044
+ },
2045
+ "latency_failures": {
2046
+ "n": 0,
2047
+ "p50_s": null,
2048
+ "p95_s": null
2049
+ },
2050
+ "ledger_charged_usd": null,
2051
+ "macro_accuracy": 0.6944444444444445,
2052
+ "model_identities": [
2053
+ "pcdm:native:runs/ts1b"
2054
+ ],
2055
+ "n_attempted": 72,
2056
+ "n_correct": 50,
2057
+ "n_planned": 72,
2058
+ "n_renormalized": 0,
2059
+ "n_scorable": 72,
2060
+ "n_valid": 72,
2061
+ "operational_success": 1.0,
2062
+ "ordinal_mae": 0.556312729138881,
2063
+ "paraphrase_consistency": {
2064
+ "agree": 26,
2065
+ "agreement": 0.7222222222222222,
2066
+ "both_correct_rate_all_pairs": 0.5555555555555556,
2067
+ "both_valid": 36,
2068
+ "pairs": 36
2069
+ },
2070
+ "per_family": {
2071
+ "adequacy": {
2072
+ "accuracy": 0.5833333333333334,
2073
+ "brier_mean": 0.5449021935508932,
2074
+ "calibration_n": 12,
2075
+ "cost_basis": [
2076
+ "local_gpu_no_provider_tariff"
2077
+ ],
2078
+ "coverage": 1.0,
2079
+ "ece": {
2080
+ "bins": [
2081
+ {
2082
+ "accuracy": null,
2083
+ "hi": 0.1,
2084
+ "lo": 0.0,
2085
+ "mean_confidence": null,
2086
+ "n": 0
2087
+ },
2088
+ {
2089
+ "accuracy": null,
2090
+ "hi": 0.2,
2091
+ "lo": 0.1,
2092
+ "mean_confidence": null,
2093
+ "n": 0
2094
+ },
2095
+ {
2096
+ "accuracy": null,
2097
+ "hi": 0.3,
2098
+ "lo": 0.2,
2099
+ "mean_confidence": null,
2100
+ "n": 0
2101
+ },
2102
+ {
2103
+ "accuracy": null,
2104
+ "hi": 0.4,
2105
+ "lo": 0.3,
2106
+ "mean_confidence": null,
2107
+ "n": 0
2108
+ },
2109
+ {
2110
+ "accuracy": null,
2111
+ "hi": 0.5,
2112
+ "lo": 0.4,
2113
+ "mean_confidence": null,
2114
+ "n": 0
2115
+ },
2116
+ {
2117
+ "accuracy": null,
2118
+ "hi": 0.6,
2119
+ "lo": 0.5,
2120
+ "mean_confidence": null,
2121
+ "n": 0
2122
+ },
2123
+ {
2124
+ "accuracy": 0.0,
2125
+ "hi": 0.7,
2126
+ "lo": 0.6,
2127
+ "mean_confidence": 0.6186434030532837,
2128
+ "n": 1
2129
+ },
2130
+ {
2131
+ "accuracy": 0.6666666666666666,
2132
+ "hi": 0.8,
2133
+ "lo": 0.7,
2134
+ "mean_confidence": 0.749783436457316,
2135
+ "n": 3
2136
+ },
2137
+ {
2138
+ "accuracy": 0.25,
2139
+ "hi": 0.9,
2140
+ "lo": 0.8,
2141
+ "mean_confidence": 0.8545016646385193,
2142
+ "n": 4
2143
+ },
2144
+ {
2145
+ "accuracy": 1.0,
2146
+ "hi": 1.0,
2147
+ "lo": 0.9,
2148
+ "mean_confidence": 0.9632230699062347,
2149
+ "n": 4
2150
+ }
2151
+ ],
2152
+ "ece": 0.2860923409461975,
2153
+ "n": 12
2154
+ },
2155
+ "latency": {
2156
+ "n": 12,
2157
+ "p50_s": 0.06046207994222641,
2158
+ "p95_s": 0.07607227428816257
2159
+ },
2160
+ "latency_failures": {
2161
+ "n": 0,
2162
+ "p50_s": null,
2163
+ "p95_s": null
2164
+ },
2165
+ "n_attempted": 12,
2166
+ "n_correct": 7,
2167
+ "n_planned": 12,
2168
+ "n_renormalized": 0,
2169
+ "n_scorable": 12,
2170
+ "n_valid": 12,
2171
+ "operational_success": 1.0,
2172
+ "ordinal_mae": null,
2173
+ "price_per_1000_decisions_usd": null,
2174
+ "schema_validity": 1.0,
2175
+ "schema_validity_strict": 1.0
2176
+ },
2177
+ "extraction": {
2178
+ "accuracy": 0.75,
2179
+ "brier_mean": 0.22976573909523312,
2180
+ "calibration_n": 12,
2181
+ "cost_basis": [
2182
+ "local_gpu_no_provider_tariff"
2183
+ ],
2184
+ "coverage": 1.0,
2185
+ "ece": {
2186
+ "bins": [
2187
+ {
2188
+ "accuracy": null,
2189
+ "hi": 0.1,
2190
+ "lo": 0.0,
2191
+ "mean_confidence": null,
2192
+ "n": 0
2193
+ },
2194
+ {
2195
+ "accuracy": null,
2196
+ "hi": 0.2,
2197
+ "lo": 0.1,
2198
+ "mean_confidence": null,
2199
+ "n": 0
2200
+ },
2201
+ {
2202
+ "accuracy": null,
2203
+ "hi": 0.3,
2204
+ "lo": 0.2,
2205
+ "mean_confidence": null,
2206
+ "n": 0
2207
+ },
2208
+ {
2209
+ "accuracy": null,
2210
+ "hi": 0.4,
2211
+ "lo": 0.3,
2212
+ "mean_confidence": null,
2213
+ "n": 0
2214
+ },
2215
+ {
2216
+ "accuracy": 1.0,
2217
+ "hi": 0.5,
2218
+ "lo": 0.4,
2219
+ "mean_confidence": 0.46592795848846436,
2220
+ "n": 1
2221
+ },
2222
+ {
2223
+ "accuracy": 0.0,
2224
+ "hi": 0.6,
2225
+ "lo": 0.5,
2226
+ "mean_confidence": 0.5530417263507843,
2227
+ "n": 2
2228
+ },
2229
+ {
2230
+ "accuracy": 0.0,
2231
+ "hi": 0.7,
2232
+ "lo": 0.6,
2233
+ "mean_confidence": 0.6272711157798767,
2234
+ "n": 1
2235
+ },
2236
+ {
2237
+ "accuracy": 1.0,
2238
+ "hi": 0.8,
2239
+ "lo": 0.7,
2240
+ "mean_confidence": 0.7457770109176636,
2241
+ "n": 1
2242
+ },
2243
+ {
2244
+ "accuracy": 1.0,
2245
+ "hi": 0.9,
2246
+ "lo": 0.8,
2247
+ "mean_confidence": 0.8463155627250671,
2248
+ "n": 1
2249
+ },
2250
+ {
2251
+ "accuracy": 1.0,
2252
+ "hi": 1.0,
2253
+ "lo": 0.9,
2254
+ "mean_confidence": 0.94781161348025,
2255
+ "n": 6
2256
+ }
2257
+ ],
2258
+ "ece": 0.24903869628906247,
2259
+ "n": 12
2260
+ },
2261
+ "latency": {
2262
+ "n": 12,
2263
+ "p50_s": 0.045599582605063915,
2264
+ "p95_s": 0.0653196045430377
2265
+ },
2266
+ "latency_failures": {
2267
+ "n": 0,
2268
+ "p50_s": null,
2269
+ "p95_s": null
2270
+ },
2271
+ "n_attempted": 12,
2272
+ "n_correct": 9,
2273
+ "n_planned": 12,
2274
+ "n_renormalized": 0,
2275
+ "n_scorable": 12,
2276
+ "n_valid": 12,
2277
+ "operational_success": 1.0,
2278
+ "ordinal_mae": null,
2279
+ "price_per_1000_decisions_usd": null,
2280
+ "schema_validity": 1.0,
2281
+ "schema_validity_strict": 1.0
2282
+ },
2283
+ "intent": {
2284
+ "accuracy": 0.75,
2285
+ "brier_mean": 0.37706546705662053,
2286
+ "calibration_n": 12,
2287
+ "cost_basis": [
2288
+ "local_gpu_no_provider_tariff"
2289
+ ],
2290
+ "coverage": 1.0,
2291
+ "ece": {
2292
+ "bins": [
2293
+ {
2294
+ "accuracy": null,
2295
+ "hi": 0.1,
2296
+ "lo": 0.0,
2297
+ "mean_confidence": null,
2298
+ "n": 0
2299
+ },
2300
+ {
2301
+ "accuracy": null,
2302
+ "hi": 0.2,
2303
+ "lo": 0.1,
2304
+ "mean_confidence": null,
2305
+ "n": 0
2306
+ },
2307
+ {
2308
+ "accuracy": null,
2309
+ "hi": 0.3,
2310
+ "lo": 0.2,
2311
+ "mean_confidence": null,
2312
+ "n": 0
2313
+ },
2314
+ {
2315
+ "accuracy": null,
2316
+ "hi": 0.4,
2317
+ "lo": 0.3,
2318
+ "mean_confidence": null,
2319
+ "n": 0
2320
+ },
2321
+ {
2322
+ "accuracy": null,
2323
+ "hi": 0.5,
2324
+ "lo": 0.4,
2325
+ "mean_confidence": null,
2326
+ "n": 0
2327
+ },
2328
+ {
2329
+ "accuracy": null,
2330
+ "hi": 0.6,
2331
+ "lo": 0.5,
2332
+ "mean_confidence": null,
2333
+ "n": 0
2334
+ },
2335
+ {
2336
+ "accuracy": 0.0,
2337
+ "hi": 0.7,
2338
+ "lo": 0.6,
2339
+ "mean_confidence": 0.6453130841255188,
2340
+ "n": 1
2341
+ },
2342
+ {
2343
+ "accuracy": null,
2344
+ "hi": 0.8,
2345
+ "lo": 0.7,
2346
+ "mean_confidence": null,
2347
+ "n": 0
2348
+ },
2349
+ {
2350
+ "accuracy": null,
2351
+ "hi": 0.9,
2352
+ "lo": 0.8,
2353
+ "mean_confidence": null,
2354
+ "n": 0
2355
+ },
2356
+ {
2357
+ "accuracy": 0.8181818181818182,
2358
+ "hi": 1.0,
2359
+ "lo": 0.9,
2360
+ "mean_confidence": 0.9673931002616882,
2361
+ "n": 11
2362
+ }
2363
+ ],
2364
+ "ece": 0.1905530989170074,
2365
+ "n": 12
2366
+ },
2367
+ "latency": {
2368
+ "n": 12,
2369
+ "p50_s": 0.09129977598786354,
2370
+ "p95_s": 0.16712851964402942
2371
+ },
2372
+ "latency_failures": {
2373
+ "n": 0,
2374
+ "p50_s": null,
2375
+ "p95_s": null
2376
+ },
2377
+ "n_attempted": 12,
2378
+ "n_correct": 9,
2379
+ "n_planned": 12,
2380
+ "n_renormalized": 0,
2381
+ "n_scorable": 12,
2382
+ "n_valid": 12,
2383
+ "operational_success": 1.0,
2384
+ "ordinal_mae": null,
2385
+ "price_per_1000_decisions_usd": null,
2386
+ "schema_validity": 1.0,
2387
+ "schema_validity_strict": 1.0
2388
+ },
2389
+ "ordinal": {
2390
+ "accuracy": 0.5833333333333334,
2391
+ "brier_mean": 0.4297174396617977,
2392
+ "calibration_n": 12,
2393
+ "cost_basis": [
2394
+ "local_gpu_no_provider_tariff"
2395
+ ],
2396
+ "coverage": 1.0,
2397
+ "ece": {
2398
+ "bins": [
2399
+ {
2400
+ "accuracy": null,
2401
+ "hi": 0.1,
2402
+ "lo": 0.0,
2403
+ "mean_confidence": null,
2404
+ "n": 0
2405
+ },
2406
+ {
2407
+ "accuracy": null,
2408
+ "hi": 0.2,
2409
+ "lo": 0.1,
2410
+ "mean_confidence": null,
2411
+ "n": 0
2412
+ },
2413
+ {
2414
+ "accuracy": null,
2415
+ "hi": 0.3,
2416
+ "lo": 0.2,
2417
+ "mean_confidence": null,
2418
+ "n": 0
2419
+ },
2420
+ {
2421
+ "accuracy": null,
2422
+ "hi": 0.4,
2423
+ "lo": 0.3,
2424
+ "mean_confidence": null,
2425
+ "n": 0
2426
+ },
2427
+ {
2428
+ "accuracy": 0.3333333333333333,
2429
+ "hi": 0.5,
2430
+ "lo": 0.4,
2431
+ "mean_confidence": 0.4347889920075734,
2432
+ "n": 3
2433
+ },
2434
+ {
2435
+ "accuracy": 0.0,
2436
+ "hi": 0.6,
2437
+ "lo": 0.5,
2438
+ "mean_confidence": 0.5180816451708475,
2439
+ "n": 3
2440
+ },
2441
+ {
2442
+ "accuracy": null,
2443
+ "hi": 0.7,
2444
+ "lo": 0.6,
2445
+ "mean_confidence": null,
2446
+ "n": 0
2447
+ },
2448
+ {
2449
+ "accuracy": 1.0,
2450
+ "hi": 0.8,
2451
+ "lo": 0.7,
2452
+ "mean_confidence": 0.7547558744748434,
2453
+ "n": 3
2454
+ },
2455
+ {
2456
+ "accuracy": 1.0,
2457
+ "hi": 0.9,
2458
+ "lo": 0.8,
2459
+ "mean_confidence": 0.8488078713417053,
2460
+ "n": 3
2461
+ },
2462
+ {
2463
+ "accuracy": null,
2464
+ "hi": 1.0,
2465
+ "lo": 0.9,
2466
+ "mean_confidence": null,
2467
+ "n": 0
2468
+ }
2469
+ ],
2470
+ "ece": 0.25399338950713474,
2471
+ "n": 12
2472
+ },
2473
+ "latency": {
2474
+ "n": 12,
2475
+ "p50_s": 0.06068861088715494,
2476
+ "p95_s": 0.09393381557893007
2477
+ },
2478
+ "latency_failures": {
2479
+ "n": 0,
2480
+ "p50_s": null,
2481
+ "p95_s": null
2482
+ },
2483
+ "n_attempted": 12,
2484
+ "n_correct": 7,
2485
+ "n_planned": 12,
2486
+ "n_renormalized": 0,
2487
+ "n_scorable": 12,
2488
+ "n_valid": 12,
2489
+ "operational_success": 1.0,
2490
+ "ordinal_mae": 0.556312729138881,
2491
+ "price_per_1000_decisions_usd": null,
2492
+ "schema_validity": 1.0,
2493
+ "schema_validity_strict": 1.0
2494
+ },
2495
+ "policy": {
2496
+ "accuracy": 0.6666666666666666,
2497
+ "brier_mean": 0.5632518159517107,
2498
+ "calibration_n": 12,
2499
+ "cost_basis": [
2500
+ "local_gpu_no_provider_tariff"
2501
+ ],
2502
+ "coverage": 1.0,
2503
+ "ece": {
2504
+ "bins": [
2505
+ {
2506
+ "accuracy": null,
2507
+ "hi": 0.1,
2508
+ "lo": 0.0,
2509
+ "mean_confidence": null,
2510
+ "n": 0
2511
+ },
2512
+ {
2513
+ "accuracy": null,
2514
+ "hi": 0.2,
2515
+ "lo": 0.1,
2516
+ "mean_confidence": null,
2517
+ "n": 0
2518
+ },
2519
+ {
2520
+ "accuracy": null,
2521
+ "hi": 0.3,
2522
+ "lo": 0.2,
2523
+ "mean_confidence": null,
2524
+ "n": 0
2525
+ },
2526
+ {
2527
+ "accuracy": null,
2528
+ "hi": 0.4,
2529
+ "lo": 0.3,
2530
+ "mean_confidence": null,
2531
+ "n": 0
2532
+ },
2533
+ {
2534
+ "accuracy": null,
2535
+ "hi": 0.5,
2536
+ "lo": 0.4,
2537
+ "mean_confidence": null,
2538
+ "n": 0
2539
+ },
2540
+ {
2541
+ "accuracy": 1.0,
2542
+ "hi": 0.6,
2543
+ "lo": 0.5,
2544
+ "mean_confidence": 0.5495584607124329,
2545
+ "n": 1
2546
+ },
2547
+ {
2548
+ "accuracy": 1.0,
2549
+ "hi": 0.7,
2550
+ "lo": 0.6,
2551
+ "mean_confidence": 0.616141676902771,
2552
+ "n": 1
2553
+ },
2554
+ {
2555
+ "accuracy": 0.0,
2556
+ "hi": 0.8,
2557
+ "lo": 0.7,
2558
+ "mean_confidence": 0.7557246387004852,
2559
+ "n": 2
2560
+ },
2561
+ {
2562
+ "accuracy": null,
2563
+ "hi": 0.9,
2564
+ "lo": 0.8,
2565
+ "mean_confidence": null,
2566
+ "n": 0
2567
+ },
2568
+ {
2569
+ "accuracy": 0.75,
2570
+ "hi": 1.0,
2571
+ "lo": 0.9,
2572
+ "mean_confidence": 0.9881341382861137,
2573
+ "n": 8
2574
+ }
2575
+ ],
2576
+ "ece": 0.3542351871728897,
2577
+ "n": 12
2578
+ },
2579
+ "latency": {
2580
+ "n": 12,
2581
+ "p50_s": 0.09179099393077195,
2582
+ "p95_s": 0.17634947728365657
2583
+ },
2584
+ "latency_failures": {
2585
+ "n": 0,
2586
+ "p50_s": null,
2587
+ "p95_s": null
2588
+ },
2589
+ "n_attempted": 12,
2590
+ "n_correct": 8,
2591
+ "n_planned": 12,
2592
+ "n_renormalized": 0,
2593
+ "n_scorable": 12,
2594
+ "n_valid": 12,
2595
+ "operational_success": 1.0,
2596
+ "ordinal_mae": null,
2597
+ "price_per_1000_decisions_usd": null,
2598
+ "schema_validity": 1.0,
2599
+ "schema_validity_strict": 1.0
2600
+ },
2601
+ "routing": {
2602
+ "accuracy": 0.8333333333333334,
2603
+ "brier_mean": 0.24032512955826416,
2604
+ "calibration_n": 12,
2605
+ "cost_basis": [
2606
+ "local_gpu_no_provider_tariff"
2607
+ ],
2608
+ "coverage": 1.0,
2609
+ "ece": {
2610
+ "bins": [
2611
+ {
2612
+ "accuracy": null,
2613
+ "hi": 0.1,
2614
+ "lo": 0.0,
2615
+ "mean_confidence": null,
2616
+ "n": 0
2617
+ },
2618
+ {
2619
+ "accuracy": null,
2620
+ "hi": 0.2,
2621
+ "lo": 0.1,
2622
+ "mean_confidence": null,
2623
+ "n": 0
2624
+ },
2625
+ {
2626
+ "accuracy": null,
2627
+ "hi": 0.3,
2628
+ "lo": 0.2,
2629
+ "mean_confidence": null,
2630
+ "n": 0
2631
+ },
2632
+ {
2633
+ "accuracy": null,
2634
+ "hi": 0.4,
2635
+ "lo": 0.3,
2636
+ "mean_confidence": null,
2637
+ "n": 0
2638
+ },
2639
+ {
2640
+ "accuracy": 0.0,
2641
+ "hi": 0.5,
2642
+ "lo": 0.4,
2643
+ "mean_confidence": 0.49998921155929565,
2644
+ "n": 1
2645
+ },
2646
+ {
2647
+ "accuracy": 0.5,
2648
+ "hi": 0.6,
2649
+ "lo": 0.5,
2650
+ "mean_confidence": 0.5483307540416718,
2651
+ "n": 2
2652
+ },
2653
+ {
2654
+ "accuracy": 1.0,
2655
+ "hi": 0.7,
2656
+ "lo": 0.6,
2657
+ "mean_confidence": 0.637911856174469,
2658
+ "n": 1
2659
+ },
2660
+ {
2661
+ "accuracy": 1.0,
2662
+ "hi": 0.8,
2663
+ "lo": 0.7,
2664
+ "mean_confidence": 0.7353148460388184,
2665
+ "n": 2
2666
+ },
2667
+ {
2668
+ "accuracy": 1.0,
2669
+ "hi": 0.9,
2670
+ "lo": 0.8,
2671
+ "mean_confidence": 0.819983184337616,
2672
+ "n": 2
2673
+ },
2674
+ {
2675
+ "accuracy": 1.0,
2676
+ "hi": 1.0,
2677
+ "lo": 0.9,
2678
+ "mean_confidence": 0.9291361421346664,
2679
+ "n": 4
2680
+ }
2681
+ ],
2682
+ "ece": 0.17763318618138627,
2683
+ "n": 12
2684
+ },
2685
+ "latency": {
2686
+ "n": 12,
2687
+ "p50_s": 0.04560648952610791,
2688
+ "p95_s": 0.08044822961091994
2689
+ },
2690
+ "latency_failures": {
2691
+ "n": 0,
2692
+ "p50_s": null,
2693
+ "p95_s": null
2694
+ },
2695
+ "n_attempted": 12,
2696
+ "n_correct": 10,
2697
+ "n_planned": 12,
2698
+ "n_renormalized": 0,
2699
+ "n_scorable": 12,
2700
+ "n_valid": 12,
2701
+ "operational_success": 1.0,
2702
+ "ordinal_mae": null,
2703
+ "price_per_1000_decisions_usd": null,
2704
+ "schema_validity": 1.0,
2705
+ "schema_validity_strict": 1.0
2706
+ }
2707
+ },
2708
+ "price_per_1000_decisions_usd": null,
2709
+ "probability_sources": [
2710
+ "native"
2711
+ ],
2712
+ "schema_validity": 1.0,
2713
+ "schema_validity_strict": 1.0,
2714
+ "splits": {
2715
+ "public": {
2716
+ "accuracy": 0.6944444444444444,
2717
+ "brier_mean": 0.39750463081241993,
2718
+ "calibration_n": 72,
2719
+ "cost_basis": [
2720
+ "local_gpu_no_provider_tariff"
2721
+ ],
2722
+ "coverage": 1.0,
2723
+ "ece": {
2724
+ "bins": [
2725
+ {
2726
+ "accuracy": null,
2727
+ "hi": 0.1,
2728
+ "lo": 0.0,
2729
+ "mean_confidence": null,
2730
+ "n": 0
2731
+ },
2732
+ {
2733
+ "accuracy": null,
2734
+ "hi": 0.2,
2735
+ "lo": 0.1,
2736
+ "mean_confidence": null,
2737
+ "n": 0
2738
+ },
2739
+ {
2740
+ "accuracy": null,
2741
+ "hi": 0.3,
2742
+ "lo": 0.2,
2743
+ "mean_confidence": null,
2744
+ "n": 0
2745
+ },
2746
+ {
2747
+ "accuracy": null,
2748
+ "hi": 0.4,
2749
+ "lo": 0.3,
2750
+ "mean_confidence": null,
2751
+ "n": 0
2752
+ },
2753
+ {
2754
+ "accuracy": 0.4,
2755
+ "hi": 0.5,
2756
+ "lo": 0.4,
2757
+ "mean_confidence": 0.45405682921409607,
2758
+ "n": 5
2759
+ },
2760
+ {
2761
+ "accuracy": 0.25,
2762
+ "hi": 0.6,
2763
+ "lo": 0.5,
2764
+ "mean_confidence": 0.538318544626236,
2765
+ "n": 8
2766
+ },
2767
+ {
2768
+ "accuracy": 0.4,
2769
+ "hi": 0.7,
2770
+ "lo": 0.6,
2771
+ "mean_confidence": 0.6290562272071838,
2772
+ "n": 5
2773
+ },
2774
+ {
2775
+ "accuracy": 0.7272727272727273,
2776
+ "hi": 0.8,
2777
+ "lo": 0.7,
2778
+ "mean_confidence": 0.7492249011993408,
2779
+ "n": 11
2780
+ },
2781
+ {
2782
+ "accuracy": 0.7,
2783
+ "hi": 0.9,
2784
+ "lo": 0.8,
2785
+ "mean_confidence": 0.8450712203979492,
2786
+ "n": 10
2787
+ },
2788
+ {
2789
+ "accuracy": 0.8787878787878788,
2790
+ "hi": 1.0,
2791
+ "lo": 0.9,
2792
+ "mean_confidence": 0.9637182950973511,
2793
+ "n": 33
2794
+ }
2795
+ ],
2796
+ "ece": 0.11412504853473769,
2797
+ "n": 72
2798
+ },
2799
+ "latency": {
2800
+ "n": 72,
2801
+ "p50_s": 0.06066564843058586,
2802
+ "p95_s": 0.12478509345091895
2803
+ },
2804
+ "latency_failures": {
2805
+ "n": 0,
2806
+ "p50_s": null,
2807
+ "p95_s": null
2808
+ },
2809
+ "n_attempted": 72,
2810
+ "n_correct": 50,
2811
+ "n_planned": 72,
2812
+ "n_renormalized": 0,
2813
+ "n_scorable": 72,
2814
+ "n_valid": 72,
2815
+ "operational_success": 1.0,
2816
+ "ordinal_mae": 0.556312729138881,
2817
+ "price_per_1000_decisions_usd": null,
2818
+ "schema_validity": 1.0,
2819
+ "schema_validity_strict": 1.0
2820
+ }
2821
+ },
2822
+ "version": "jevbench-v1"
2823
+ }
2824
+ }
probe_results.json ADDED
@@ -0,0 +1,155 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "T": 1.1242100350620867,
3
+ "null_offset": 0.0,
4
+ "calib_sets": [
5
+ "val"
6
+ ],
7
+ "val_nll": 0.37577103864577555,
8
+ "args": {
9
+ "name": "probe_ts1b",
10
+ "backbone": "Qwen/Qwen3-1.7B-Base",
11
+ "lora_layers": 8,
12
+ "lora_r": 16,
13
+ "steps": 12000,
14
+ "bs": 64,
15
+ "grad_accum": 1,
16
+ "lr": 0.0003,
17
+ "lora_lr": 0.0001,
18
+ "seed": 0,
19
+ "no_hybrid": false,
20
+ "no_cand_null": false,
21
+ "no_null": false,
22
+ "null": "factored",
23
+ "hard_only": false,
24
+ "mix": "full",
25
+ "eval_every": 4000,
26
+ "val_every": 1000,
27
+ "ckpt_every": 2000,
28
+ "eval_bs": 128,
29
+ "device": "auto",
30
+ "wandb": false,
31
+ "hf_repo": null,
32
+ "ckpt_upload": false,
33
+ "data": "data_probe",
34
+ "extra_data": null,
35
+ "eval_cap": null,
36
+ "calib_sets": null,
37
+ "family_weights": null,
38
+ "bucket_map": null,
39
+ "null_aug": null,
40
+ "smoke": false,
41
+ "eval_limit": 0,
42
+ "tap_layer": 20,
43
+ "zscore": false,
44
+ "eval_only": true,
45
+ "dump_logits": null,
46
+ "tower_d": 512,
47
+ "tower_layers": 2,
48
+ "tower_heads": 8,
49
+ "joint": false,
50
+ "listwise": false,
51
+ "extra_tap": 0,
52
+ "init_from": null,
53
+ "readout": "native",
54
+ "nc_head": "n3",
55
+ "nc_render": "letters_nonull",
56
+ "score_head": "choice",
57
+ "noul_head": "bern",
58
+ "qtype_filter": null,
59
+ "ordinal_smooth": 0.0,
60
+ "max_state": 1024,
61
+ "perm_lambda": 0.0,
62
+ "no_shuffle": false,
63
+ "zero_shot": false,
64
+ "shots": 0,
65
+ "cand_encoder": "backbone",
66
+ "head": "mlp",
67
+ "z_dim": 128,
68
+ "z_probes": 8,
69
+ "tiny_layers": 2,
70
+ "distill_alpha": 1.0,
71
+ "distill_beta": 0.0,
72
+ "distill_T": 2.0,
73
+ "delta_gamma": 0.0
74
+ },
75
+ "eval": {
76
+ "mmlu_choicesonly": {
77
+ "raw": {
78
+ "nll": 2.132304014679386,
79
+ "brier": 0.855733086948981,
80
+ "jsd": 0.4787010918172532,
81
+ "acc": 0.22666666666666666,
82
+ "acc_k": 0.22666666666666666,
83
+ "ece": 0.05177726411571105,
84
+ "sel_acc80": 0.25625,
85
+ "coverage": 0.9975,
86
+ "false_abstain": 0.0025,
87
+ "null_recall": "n/a"
88
+ },
89
+ "scaled": {
90
+ "nll": 2.119891606203886,
91
+ "brier": 0.8533568513213663,
92
+ "jsd": 0.4808686117250174,
93
+ "acc": 0.22666666666666666,
94
+ "acc_k": 0.22666666666666666,
95
+ "ece": 0.03572918727373083,
96
+ "sel_acc80": 0.25625,
97
+ "coverage": 0.9975,
98
+ "false_abstain": 0.0025,
99
+ "null_recall": "n/a"
100
+ }
101
+ },
102
+ "mmlu_pro": {
103
+ "raw": {
104
+ "nll": 1.870684202959633,
105
+ "brier": 0.7778605752958676,
106
+ "jsd": 0.4253066819218272,
107
+ "acc": 0.3425,
108
+ "acc_k": 0.3433333333333333,
109
+ "ece": 0.04600312157223621,
110
+ "sel_acc80": 0.39166666666666666,
111
+ "coverage": 0.995,
112
+ "false_abstain": 0.005,
113
+ "null_recall": "n/a"
114
+ },
115
+ "scaled": {
116
+ "nll": 1.8650622347991608,
117
+ "brier": 0.7769553003598431,
118
+ "jsd": 0.4300648104283429,
119
+ "acc": 0.3425,
120
+ "acc_k": 0.3433333333333333,
121
+ "ece": 0.04282940872013569,
122
+ "sel_acc80": 0.39166666666666666,
123
+ "coverage": 0.995,
124
+ "false_abstain": 0.005,
125
+ "null_recall": "n/a"
126
+ }
127
+ },
128
+ "mmlu_shuffledq": {
129
+ "raw": {
130
+ "nll": 2.2841442391906717,
131
+ "brier": 0.8968269319049337,
132
+ "jsd": 0.49137674516409957,
133
+ "acc": 0.21083333333333334,
134
+ "acc_k": 0.21666666666666667,
135
+ "ece": 0.09170151713614662,
136
+ "sel_acc80": 0.23125,
137
+ "coverage": 0.9775,
138
+ "false_abstain": 0.0225,
139
+ "null_recall": "n/a"
140
+ },
141
+ "scaled": {
142
+ "nll": 2.2550489918898133,
143
+ "brier": 0.8906454120623264,
144
+ "jsd": 0.492517543247486,
145
+ "acc": 0.21083333333333334,
146
+ "acc_k": 0.21666666666666667,
147
+ "ece": 0.0703738091637691,
148
+ "sel_acc80": 0.23125,
149
+ "coverage": 0.9766666666666667,
150
+ "false_abstain": 0.023333333333333334,
151
+ "null_recall": "n/a"
152
+ }
153
+ }
154
+ }
155
+ }
results.json ADDED
@@ -0,0 +1,2248 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "T": 1.1242100350620867,
3
+ "null_offset": 0.0,
4
+ "calib_sets": [
5
+ "val"
6
+ ],
7
+ "val_nll": 0.37577103864577555,
8
+ "args": {
9
+ "name": "ts1b",
10
+ "backbone": "Qwen/Qwen3-1.7B-Base",
11
+ "lora_layers": 8,
12
+ "lora_r": 16,
13
+ "steps": 12000,
14
+ "bs": 64,
15
+ "grad_accum": 1,
16
+ "lr": 0.0003,
17
+ "lora_lr": 0.0001,
18
+ "seed": 0,
19
+ "no_hybrid": false,
20
+ "no_cand_null": false,
21
+ "no_null": false,
22
+ "null": "factored",
23
+ "hard_only": false,
24
+ "mix": "full",
25
+ "eval_every": 4000,
26
+ "val_every": 1000,
27
+ "ckpt_every": 2000,
28
+ "eval_bs": 128,
29
+ "device": "auto",
30
+ "wandb": false,
31
+ "hf_repo": null,
32
+ "ckpt_upload": false,
33
+ "data": "data_v5",
34
+ "extra_data": "data_kb,data_wf,data_wf_hf,data_wf_long,data_wh,data_u",
35
+ "eval_cap": 1500,
36
+ "calib_sets": null,
37
+ "family_weights": "E:0.40,K:0.15,W:0.35,U:0.10",
38
+ "bucket_map": "data_wf_long=W,data_wh=W,data_u=U",
39
+ "null_aug": "W:0.20",
40
+ "smoke": false,
41
+ "eval_limit": 0,
42
+ "tap_layer": 20,
43
+ "zscore": false,
44
+ "eval_only": true,
45
+ "dump_logits": null,
46
+ "tower_d": 512,
47
+ "tower_layers": 2,
48
+ "tower_heads": 8,
49
+ "joint": false,
50
+ "listwise": false,
51
+ "extra_tap": 0,
52
+ "init_from": null,
53
+ "readout": "native",
54
+ "nc_head": "n3",
55
+ "nc_render": "letters_nonull",
56
+ "score_head": "choice",
57
+ "noul_head": "bern",
58
+ "qtype_filter": null,
59
+ "ordinal_smooth": 0.0,
60
+ "max_state": 1024,
61
+ "perm_lambda": 0.0,
62
+ "no_shuffle": false,
63
+ "zero_shot": false,
64
+ "shots": 0,
65
+ "cand_encoder": "backbone",
66
+ "head": "mlp",
67
+ "z_dim": 128,
68
+ "z_probes": 8,
69
+ "tiny_layers": 2,
70
+ "distill_alpha": 1.0,
71
+ "distill_beta": 0.0,
72
+ "distill_T": 2.0,
73
+ "delta_gamma": 0.0
74
+ },
75
+ "eval": {
76
+ "anli_test": {
77
+ "raw": {
78
+ "nll": 1.1289180740916946,
79
+ "brier": 0.6692979299798677,
80
+ "jsd": 0.2759239949154016,
81
+ "acc": 0.496875,
82
+ "acc_k": 0.496875,
83
+ "ece": 0.19597449547611173,
84
+ "sel_acc80": 0.516015625,
85
+ "coverage": 1.0,
86
+ "false_abstain": 0.0,
87
+ "null_recall": "n/a"
88
+ },
89
+ "scaled": {
90
+ "nll": 1.0832124084790902,
91
+ "brier": 0.6531642279580523,
92
+ "jsd": 0.2744905806400761,
93
+ "acc": 0.496875,
94
+ "acc_k": 0.496875,
95
+ "ece": 0.1716771886870265,
96
+ "sel_acc80": 0.51640625,
97
+ "coverage": 1.0,
98
+ "false_abstain": 0.0,
99
+ "null_recall": "n/a"
100
+ }
101
+ },
102
+ "banking77_k": {
103
+ "raw": {
104
+ "nll": 0.8717148169361592,
105
+ "brier": 0.4279012928249291,
106
+ "jsd": 0.1819415852368955,
107
+ "acc": 0.7100130039011704,
108
+ "acc_k": 0.8606340819022457,
109
+ "ece": 0.10486168391101536,
110
+ "sel_acc80": 0.7724502234863877,
111
+ "conf_wrong": 0.09282970550576185,
112
+ "coverage": 0.6501950585175553,
113
+ "false_abstain": 0.08586525759577279,
114
+ "null_recall": 0.6056338028169014,
115
+ "auroc_null": 0.8788824577101132,
116
+ "auroc_conf": 0.8729193341869398
117
+ },
118
+ "scaled": {
119
+ "nll": 0.8565775965678362,
120
+ "brier": 0.42266858395142376,
121
+ "jsd": 0.18248111365256584,
122
+ "acc": 0.711963589076723,
123
+ "acc_k": 0.8606340819022457,
124
+ "ece": 0.09658078322252464,
125
+ "sel_acc80": 0.7757009345794392,
126
+ "conf_wrong": 0.08450704225352113,
127
+ "coverage": 0.6449934980494147,
128
+ "false_abstain": 0.08982826948480846,
129
+ "null_recall": 0.6120358514724712,
130
+ "auroc_null": 0.8788824577101132,
131
+ "auroc_conf": 0.8727353915736523
132
+ }
133
+ },
134
+ "banking77_test": {
135
+ "raw": {
136
+ "nll": 1.8774157424809907,
137
+ "brier": 0.5718742123352633,
138
+ "jsd": 0.2792573519790713,
139
+ "acc": 0.5929778933680104,
140
+ "acc_k": 0.6134590377113134,
141
+ "ece": 0.11646726103535868,
142
+ "sel_acc80": 0.6834620073141,
143
+ "coverage": 0.9307542262678804,
144
+ "false_abstain": 0.06924577373211964,
145
+ "null_recall": "n/a"
146
+ },
147
+ "scaled": {
148
+ "nll": 1.7716108606732313,
149
+ "brier": 0.565376058808758,
150
+ "jsd": 0.28325008418116127,
151
+ "acc": 0.5923276983094928,
152
+ "acc_k": 0.6134590377113134,
153
+ "ece": 0.09275831222069124,
154
+ "sel_acc80": 0.6822429906542056,
155
+ "coverage": 0.9245773732119635,
156
+ "false_abstain": 0.0754226267880364,
157
+ "null_recall": "n/a"
158
+ }
159
+ },
160
+ "boolq_val": {
161
+ "raw": {
162
+ "nll": 0.43645680106840656,
163
+ "brier": 0.2618663419126855,
164
+ "jsd": 0.11080636584099614,
165
+ "acc": 0.8235474006116208,
166
+ "acc_k": 0.825993883792049,
167
+ "ece": 0.047257785731499316,
168
+ "sel_acc80": 0.8799694189602446,
169
+ "coverage": 0.9957186544342508,
170
+ "false_abstain": 0.004281345565749235,
171
+ "null_recall": "n/a"
172
+ },
173
+ "scaled": {
174
+ "nll": 0.42464742925907756,
175
+ "brier": 0.2590639951058971,
176
+ "jsd": 0.11250903904001809,
177
+ "acc": 0.8235474006116208,
178
+ "acc_k": 0.825993883792049,
179
+ "ece": 0.03589964046937609,
180
+ "sel_acc80": 0.8799694189602446,
181
+ "coverage": 0.9957186544342508,
182
+ "false_abstain": 0.004281345565749235,
183
+ "null_recall": "n/a"
184
+ }
185
+ },
186
+ "chaos_mnli": {
187
+ "raw": {
188
+ "nll": 1.2939700895771997,
189
+ "brier": 0.24465358531101472,
190
+ "jsd": 0.0949094297745448
191
+ },
192
+ "scaled": {
193
+ "nll": 1.212854225335715,
194
+ "brier": 0.22637623031446555,
195
+ "jsd": 0.08678660793866103
196
+ }
197
+ },
198
+ "clinc_heldout": {
199
+ "raw": {
200
+ "nll": 0.23143039751182512,
201
+ "brier": 0.09878319172558447,
202
+ "jsd": 0.04650508883232555,
203
+ "acc": 0.9320712694877505,
204
+ "acc_k": 0.9632516703786191,
205
+ "ece": 0.024297224585887893,
206
+ "sel_acc80": 0.9958275382475661,
207
+ "coverage": 0.9576837416481069,
208
+ "false_abstain": 0.042316258351893093,
209
+ "null_recall": "n/a"
210
+ },
211
+ "scaled": {
212
+ "nll": 0.2261652916434549,
213
+ "brier": 0.09926372397049718,
214
+ "jsd": 0.04731882559377883,
215
+ "acc": 0.9320712694877505,
216
+ "acc_k": 0.9632516703786191,
217
+ "ece": 0.024705316879701483,
218
+ "sel_acc80": 0.9958275382475661,
219
+ "coverage": 0.9565701559020044,
220
+ "false_abstain": 0.043429844097995544,
221
+ "null_recall": "n/a"
222
+ }
223
+ },
224
+ "clinc_k": {
225
+ "raw": {
226
+ "nll": 0.3231793638375014,
227
+ "brier": 0.1685297802997737,
228
+ "jsd": 0.07669964194250567,
229
+ "acc": 0.8873651079136691,
230
+ "acc_k": 0.9664729548502459,
231
+ "ece": 0.018513263035699633,
232
+ "sel_acc80": 0.9530767069401517,
233
+ "conf_wrong": 0.04115784712799638,
234
+ "coverage": 0.5613758992805755,
235
+ "false_abstain": 0.042914617791685296,
236
+ "null_recall": 0.8389868837630031,
237
+ "auroc_null": 0.9699786110290582,
238
+ "auroc_conf": 0.9684840721009897
239
+ },
240
+ "scaled": {
241
+ "nll": 0.3216351683687122,
242
+ "brier": 0.16711785999498344,
243
+ "jsd": 0.07704117516737093,
244
+ "acc": 0.8887140287769785,
245
+ "acc_k": 0.9664729548502459,
246
+ "ece": 0.01572394735856555,
247
+ "sel_acc80": 0.9530767069401517,
248
+ "conf_wrong": 0.03844414292175486,
249
+ "coverage": 0.5584532374100719,
250
+ "false_abstain": 0.044702726866338846,
251
+ "null_recall": 0.8430574400723655,
252
+ "auroc_null": 0.9699786110290585,
253
+ "auroc_conf": 0.9683471940092281
254
+ }
255
+ },
256
+ "clinc_oos": {
257
+ "raw": {
258
+ "nll": 0.5756657426096591,
259
+ "brier": 0.2989822245764066,
260
+ "jsd": 0.16344119860379913,
261
+ "acc": 0.8268268268268268,
262
+ "ece": 0.14330651938497604,
263
+ "sel_acc80": 0.8525,
264
+ "conf_wrong": 0.02002002002002002,
265
+ "coverage": 0.17317317317317316,
266
+ "false_abstain": "n/a",
267
+ "null_recall": 0.8268268268268268
268
+ },
269
+ "scaled": {
270
+ "nll": 0.5756657362777501,
271
+ "brier": 0.28923916904188607,
272
+ "jsd": 0.1634411964162575,
273
+ "acc": 0.8328328328328328,
274
+ "ece": 0.15310207864006717,
275
+ "sel_acc80": 0.8625,
276
+ "conf_wrong": 0.02002002002002002,
277
+ "coverage": 0.16716716716716717,
278
+ "false_abstain": "n/a",
279
+ "null_recall": 0.8328328328328328
280
+ }
281
+ },
282
+ "clinc_test": {
283
+ "raw": {
284
+ "nll": 0.7784299687760449,
285
+ "brier": 0.28365467007926637,
286
+ "jsd": 0.14319114862151858,
287
+ "acc": 0.8037320143884892,
288
+ "acc_k": 0.8340827338129496,
289
+ "ece": 0.028675001282110875,
290
+ "sel_acc80": 0.9033436358527677,
291
+ "coverage": 0.9332284172661871,
292
+ "false_abstain": 0.06677158273381295,
293
+ "null_recall": "n/a"
294
+ },
295
+ "scaled": {
296
+ "nll": 0.7536637095175133,
297
+ "brier": 0.2846120396481659,
298
+ "jsd": 0.14652556864953595,
299
+ "acc": 0.8005845323741008,
300
+ "acc_k": 0.8340827338129496,
301
+ "ece": 0.02309598260820451,
302
+ "sel_acc80": 0.9013767912334926,
303
+ "coverage": 0.9280575539568345,
304
+ "false_abstain": 0.07194244604316546,
305
+ "null_recall": "n/a"
306
+ }
307
+ },
308
+ "cse_banking77": {
309
+ "raw": {
310
+ "nll": 0.9038375048530173,
311
+ "brier": 0.36887379787688285,
312
+ "jsd": 0.16302877525225848,
313
+ "acc": 0.7076666666666667,
314
+ "acc_k": 0.768,
315
+ "ece": 0.1064401980837186,
316
+ "sel_acc80": 0.78875,
317
+ "conf_wrong": 0.116,
318
+ "coverage": 0.8143333333333334,
319
+ "false_abstain": 0.0952,
320
+ "null_recall": 0.638,
321
+ "auroc_null": 0.8856160000000001,
322
+ "auroc_conf": 0.8466096000000001
323
+ },
324
+ "scaled": {
325
+ "nll": 0.8694398602835821,
326
+ "brier": 0.36427865239843843,
327
+ "jsd": 0.16326378234678152,
328
+ "acc": 0.7063333333333334,
329
+ "acc_k": 0.768,
330
+ "ece": 0.09903382284442586,
331
+ "sel_acc80": 0.7875,
332
+ "conf_wrong": 0.1,
333
+ "coverage": 0.8113333333333334,
334
+ "false_abstain": 0.0984,
335
+ "null_recall": 0.64,
336
+ "auroc_null": 0.8856160000000001,
337
+ "auroc_conf": 0.84588
338
+ },
339
+ "cse": {
340
+ "add_irr/p_irr": 0.0005295094670840995,
341
+ "add_irr/dlo_top2": 0.13645354047403815,
342
+ "remove_gold/dp_null": 0.4498149465108654,
343
+ "dup/mass_err": 0.0331556449693544,
344
+ "dup/slot_gap": 0.17666359726031608,
345
+ "dup/acc_dedup": 0.844,
346
+ "reorder/max_dp": 0.08150279471576005,
347
+ "near_dup/cluster_mass_err": 0.03125877748820778
348
+ }
349
+ },
350
+ "cse_clinc": {
351
+ "raw": {
352
+ "nll": 0.30950664569064634,
353
+ "brier": 0.12839356722975465,
354
+ "jsd": 0.05918977870640326,
355
+ "acc": 0.8746124698587668,
356
+ "acc_k": 0.9073535521395929,
357
+ "ece": 0.02284423466431781,
358
+ "sel_acc80": 0.9599655617735686,
359
+ "conf_wrong": 0.028225806451612902,
360
+ "coverage": 0.8284533241474337,
361
+ "false_abstain": 0.03489821354383049,
362
+ "null_recall": 0.8346774193548387,
363
+ "auroc_null": 0.9766046946406314,
364
+ "auroc_conf": 0.9623075170537546
365
+ },
366
+ "scaled": {
367
+ "nll": 0.3052672934138403,
368
+ "brier": 0.12660908099041643,
369
+ "jsd": 0.059311244384245086,
370
+ "acc": 0.875645883568722,
371
+ "acc_k": 0.9073535521395929,
372
+ "ece": 0.01759377376523214,
373
+ "sel_acc80": 0.9595350839431769,
374
+ "conf_wrong": 0.028225806451612902,
375
+ "coverage": 0.8274199104374784,
376
+ "false_abstain": 0.03489821354383049,
377
+ "null_recall": 0.8407258064516129,
378
+ "auroc_null": 0.9766046946406314,
379
+ "auroc_conf": 0.9617245399841859
380
+ },
381
+ "cse": {
382
+ "add_irr/p_irr": 0.001569051850666318,
383
+ "add_irr/dlo_top2": 0.16152436470361048,
384
+ "remove_gold/dp_null": 0.7188482229822042,
385
+ "dup/mass_err": 0.012801834568350948,
386
+ "dup/slot_gap": 0.16565532824922566,
387
+ "dup/acc_dedup": 0.9798387096774194,
388
+ "reorder/max_dp": 0.03195492150344274,
389
+ "near_dup/cluster_mass_err": 0.016617430864665605
390
+ }
391
+ },
392
+ "cse_hwu64": {
393
+ "raw": {
394
+ "nll": 0.5609047915995059,
395
+ "brier": 0.2507698667896729,
396
+ "jsd": 0.1167981555537652,
397
+ "acc": 0.7933333333333333,
398
+ "acc_k": 0.8428,
399
+ "ece": 0.04368377149105075,
400
+ "sel_acc80": 0.8616666666666667,
401
+ "conf_wrong": 0.068,
402
+ "coverage": 0.845,
403
+ "false_abstain": 0.0504,
404
+ "null_recall": 0.678,
405
+ "auroc_null": 0.9338304,
406
+ "auroc_conf": 0.89552
407
+ },
408
+ "scaled": {
409
+ "nll": 0.5507840743009456,
410
+ "brier": 0.24834142217870656,
411
+ "jsd": 0.11789211835745053,
412
+ "acc": 0.792,
413
+ "acc_k": 0.8428,
414
+ "ece": 0.0344938989629348,
415
+ "sel_acc80": 0.8608333333333333,
416
+ "conf_wrong": 0.062,
417
+ "coverage": 0.843,
418
+ "false_abstain": 0.0524,
419
+ "null_recall": 0.68,
420
+ "auroc_null": 0.9338304,
421
+ "auroc_conf": 0.8944143999999999
422
+ },
423
+ "cse": {
424
+ "add_irr/p_irr": 0.011361859426103703,
425
+ "add_irr/dlo_top2": 0.13446737565746777,
426
+ "remove_gold/dp_null": 0.4905430292236051,
427
+ "dup/mass_err": 0.027505983864113463,
428
+ "dup/slot_gap": 0.17891635448112775,
429
+ "dup/acc_dedup": 0.938,
430
+ "reorder/max_dp": 0.0636793900321718,
431
+ "near_dup/cluster_mass_err": 0.03313319766287608
432
+ }
433
+ },
434
+ "cse_snli": {
435
+ "raw": {
436
+ "nll": 0.40686523131470476,
437
+ "brier": 0.16987205012507944,
438
+ "jsd": 0.08462270844792318,
439
+ "acc": 0.8890134529147982,
440
+ "acc_k": 0.9016544117647058,
441
+ "ece": 0.07036731607087164,
442
+ "sel_acc80": 0.9271368519383466,
443
+ "conf_wrong": 0.032,
444
+ "coverage": 0.844170403587444,
445
+ "false_abstain": 0.0,
446
+ "null_recall": 0.834,
447
+ "auroc_null": 0.9999007352941176,
448
+ "auroc_conf": 0.9475193014705883
449
+ },
450
+ "scaled": {
451
+ "nll": 0.41134177711307934,
452
+ "brier": 0.17059116434583468,
453
+ "jsd": 0.08824297050300778,
454
+ "acc": 0.8893871449925261,
455
+ "acc_k": 0.9016544117647058,
456
+ "ece": 0.08119960839794711,
457
+ "sel_acc80": 0.9285380663241476,
458
+ "conf_wrong": 0.03,
459
+ "coverage": 0.843796711509716,
460
+ "false_abstain": 0.0,
461
+ "null_recall": 0.836,
462
+ "auroc_null": 0.9999007352941176,
463
+ "auroc_conf": 0.9443722426470589
464
+ },
465
+ "cse": {
466
+ "add_irr/p_irr": 0.01940807816808774,
467
+ "add_irr/dlo_top2": 0.09028009353347251,
468
+ "remove_gold/dp_null": 0.7515602824201983,
469
+ "dup/mass_err": 0.04235328816168476,
470
+ "dup/slot_gap": 0.1555379865783034,
471
+ "dup/acc_dedup": 0.904,
472
+ "reorder/max_dp": 0.01188999435165897,
473
+ "near_dup/cluster_mass_err": 0.04870520358202735
474
+ }
475
+ },
476
+ "hwu64_test": {
477
+ "raw": {
478
+ "nll": 0.9291563079845718,
479
+ "brier": 0.3372229757452212,
480
+ "jsd": 0.18254796335437554,
481
+ "acc": 0.7676579925650557,
482
+ "acc_k": 0.7843866171003717,
483
+ "ece": 0.03418452451940364,
484
+ "sel_acc80": 0.8641114982578397,
485
+ "coverage": 0.9563197026022305,
486
+ "false_abstain": 0.04368029739776952,
487
+ "null_recall": "n/a"
488
+ },
489
+ "scaled": {
490
+ "nll": 0.9142610185219991,
491
+ "brier": 0.3406498241436756,
492
+ "jsd": 0.18930016280648101,
493
+ "acc": 0.7611524163568774,
494
+ "acc_k": 0.7843866171003717,
495
+ "ece": 0.04919961878690576,
496
+ "sel_acc80": 0.8652729384436701,
497
+ "coverage": 0.9451672862453532,
498
+ "false_abstain": 0.05483271375464684,
499
+ "null_recall": "n/a"
500
+ }
501
+ },
502
+ "ksweep_banking77": {
503
+ "raw": {
504
+ "nll": 0.9349302343615308,
505
+ "brier": 0.4471019757458735,
506
+ "jsd": 0.19401309938809777,
507
+ "acc": 0.68875,
508
+ "acc_k": 0.8521428571428571,
509
+ "ece": 0.11756176422217063,
510
+ "sel_acc80": 0.7609375,
511
+ "conf_wrong": 0.08214285714285714,
512
+ "coverage": 0.6689285714285714,
513
+ "false_abstain": 0.08392857142857142,
514
+ "null_recall": 0.5782142857142857,
515
+ "auroc_null": 0.8703889030612244,
516
+ "auroc_conf": 0.8664147959183673
517
+ },
518
+ "scaled": {
519
+ "nll": 0.9189812145302394,
520
+ "brier": 0.4404521904122212,
521
+ "jsd": 0.19510637309634174,
522
+ "acc": 0.6907142857142857,
523
+ "acc_k": 0.8521428571428571,
524
+ "ece": 0.10724775899201631,
525
+ "sel_acc80": 0.7647321428571429,
526
+ "conf_wrong": 0.07571428571428572,
527
+ "coverage": 0.6648214285714286,
528
+ "false_abstain": 0.08714285714285715,
529
+ "null_recall": 0.5832142857142857,
530
+ "auroc_null": 0.8703889030612244,
531
+ "auroc_conf": 0.8663109056122449
532
+ },
533
+ "ksweep": {
534
+ "K2/p_null_absent": 0.8690049877914134,
535
+ "K2/p_null_present": 0.1643589718966723,
536
+ "K2/acc": 0.975,
537
+ "K2/auroc": 0.9450375,
538
+ "K3/p_null_absent": 0.7012581462779781,
539
+ "K3/p_null_present": 0.08284207384213005,
540
+ "K3/acc": 0.96,
541
+ "K3/auroc": 0.9444312499999999,
542
+ "K5/p_null_absent": 0.6039111433527432,
543
+ "K5/p_null_present": 0.08178361789279734,
544
+ "K5/acc": 0.935,
545
+ "K5/auroc": 0.9240437499999999,
546
+ "K10/p_null_absent": 0.5488354222249473,
547
+ "K10/p_null_present": 0.10999041035836854,
548
+ "K10/acc": 0.89,
549
+ "K10/auroc": 0.8832062500000001,
550
+ "K20/p_null_absent": 0.46607297291746363,
551
+ "K20/p_null_present": 0.13052993028948548,
552
+ "K20/acc": 0.84,
553
+ "K20/auroc": 0.8385499999999998,
554
+ "K50/p_null_absent": 0.2712666229833849,
555
+ "K50/p_null_present": 0.10264942378737033,
556
+ "K50/acc": 0.715,
557
+ "K50/auroc": 0.7575625,
558
+ "K77/p_null_absent": 0.22605883536394686,
559
+ "K77/p_null_present": 0.10786349694710225,
560
+ "K77/acc": 0.65,
561
+ "K77/auroc": 0.7228375,
562
+ "p_null_absent_range": 0.6429461524274666
563
+ }
564
+ },
565
+ "ksweep_clinc": {
566
+ "raw": {
567
+ "nll": 0.4705098322558086,
568
+ "brier": 0.24142269373822897,
569
+ "jsd": 0.10927668251896748,
570
+ "acc": 0.8366228070175439,
571
+ "acc_k": 0.943295739348371,
572
+ "ece": 0.03536469359791146,
573
+ "sel_acc80": 0.9007439310884886,
574
+ "conf_wrong": 0.05169172932330827,
575
+ "coverage": 0.5897556390977443,
576
+ "false_abstain": 0.05263157894736842,
577
+ "null_recall": 0.7678571428571429,
578
+ "auroc_null": 0.9494151654198153,
579
+ "auroc_conf": 0.9456508596051532
580
+ },
581
+ "scaled": {
582
+ "nll": 0.4667588320710177,
583
+ "brier": 0.23902214623321588,
584
+ "jsd": 0.11002446956487777,
585
+ "acc": 0.837875939849624,
586
+ "acc_k": 0.943295739348371,
587
+ "ece": 0.03253249252649176,
588
+ "sel_acc80": 0.9025058731401723,
589
+ "conf_wrong": 0.04887218045112782,
590
+ "coverage": 0.5850563909774436,
591
+ "false_abstain": 0.05670426065162907,
592
+ "null_recall": 0.7731829573934837,
593
+ "auroc_null": 0.9494151654198151,
594
+ "auroc_conf": 0.9453110768776578
595
+ },
596
+ "ksweep": {
597
+ "K2/p_null_absent": 0.9109689841002581,
598
+ "K2/p_null_present": 0.05722885545324601,
599
+ "K2/acc": 0.9974937343358395,
600
+ "K2/auroc": 0.9872048542408653,
601
+ "K3/p_null_absent": 0.8574261630425477,
602
+ "K3/p_null_present": 0.03833966100025525,
603
+ "K3/acc": 0.9949874686716792,
604
+ "K3/auroc": 0.9842714555813091,
605
+ "K5/p_null_absent": 0.8189782498793055,
606
+ "K5/p_null_present": 0.04019082001239549,
607
+ "K5/acc": 0.9899749373433584,
608
+ "K5/auroc": 0.9828707106111142,
609
+ "K10/p_null_absent": 0.7780514757080648,
610
+ "K10/p_null_present": 0.06639160508924383,
611
+ "K10/acc": 0.9799498746867168,
612
+ "K10/auroc": 0.9722363552992757,
613
+ "K20/p_null_absent": 0.7468801522364882,
614
+ "K20/p_null_present": 0.10790303507195388,
615
+ "K20/acc": 0.9649122807017544,
616
+ "K20/auroc": 0.9462566189910867,
617
+ "K50/p_null_absent": 0.6084068263040617,
618
+ "K50/p_null_present": 0.11852430529603011,
619
+ "K50/acc": 0.924812030075188,
620
+ "K50/auroc": 0.9115206562772847,
621
+ "K100/p_null_absent": 0.4608676842318143,
622
+ "K100/p_null_present": 0.10504461155228671,
623
+ "K100/acc": 0.8972431077694235,
624
+ "K100/auroc": 0.8819479777137078,
625
+ "K150/p_null_absent": 0.3821950357776126,
626
+ "K150/p_null_present": 0.13089805873773785,
627
+ "K150/acc": 0.7969924812030075,
628
+ "K150/auroc": 0.8273377679788443,
629
+ "p_null_absent_range": 0.5287739483226455
630
+ }
631
+ },
632
+ "mmlu_cf": {
633
+ "raw": {
634
+ "nll": 1.9652890876376825,
635
+ "brier": 0.7101132646736971,
636
+ "jsd": 0.40397400241739434,
637
+ "acc": 0.34933333333333333,
638
+ "acc_k": 0.36,
639
+ "ece": 0.06165287426610789,
640
+ "sel_acc80": 0.4008333333333333,
641
+ "coverage": 0.9833333333333333,
642
+ "false_abstain": 0.016666666666666666,
643
+ "null_recall": "n/a"
644
+ },
645
+ "scaled": {
646
+ "nll": 1.9540729414231996,
647
+ "brier": 0.7075911562875681,
648
+ "jsd": 0.40846717789843123,
649
+ "acc": 0.34933333333333333,
650
+ "acc_k": 0.36,
651
+ "ece": 0.05156281360983849,
652
+ "sel_acc80": 0.40166666666666667,
653
+ "coverage": 0.9833333333333333,
654
+ "false_abstain": 0.016666666666666666,
655
+ "null_recall": "n/a"
656
+ },
657
+ "cf": {
658
+ "orig/acc": 0.376,
659
+ "orig/among_k": 0.376,
660
+ "remove3/acc": 0.376,
661
+ "remove3/among_k": 0.432,
662
+ "add3_unrel/acc": 0.356,
663
+ "add3_unrel/among_k": 0.356,
664
+ "replace_hard/acc": 0.372,
665
+ "replace_hard/among_k": 0.376,
666
+ "add_neardup/acc": 0.272,
667
+ "add_neardup/among_k": 0.276,
668
+ "reorder/acc": 0.344,
669
+ "reorder/among_k": 0.344,
670
+ "add3_unrel/dlo_top2": 0.13401622239778724,
671
+ "reorder/max_dp": 0.12416648952616378,
672
+ "add_neardup/slot_gap": 0.12370567691186443,
673
+ "add_neardup/cluster_mass_err": 0.09759602142916993
674
+ }
675
+ },
676
+ "mmlu_cf_teacher": {
677
+ "raw": {
678
+ "nll": 1.9799224626151832,
679
+ "brier": 0.718764453695287,
680
+ "jsd": 0.411908770920552,
681
+ "acc": 0.33555555555555555,
682
+ "acc_k": 0.3458333333333333,
683
+ "ece": 0.04779739408546853,
684
+ "sel_acc80": 0.3767361111111111,
685
+ "coverage": 0.9838888888888889,
686
+ "false_abstain": 0.01611111111111111,
687
+ "null_recall": "n/a"
688
+ },
689
+ "scaled": {
690
+ "nll": 1.9711311573138646,
691
+ "brier": 0.716484928918705,
692
+ "jsd": 0.4160687618841204,
693
+ "acc": 0.33555555555555555,
694
+ "acc_k": 0.3458333333333333,
695
+ "ece": 0.036748226951393806,
696
+ "sel_acc80": 0.3770833333333333,
697
+ "coverage": 0.9838888888888889,
698
+ "false_abstain": 0.01611111111111111,
699
+ "null_recall": "n/a"
700
+ },
701
+ "cf": {
702
+ "orig/acc": 0.3425,
703
+ "orig/among_k": 0.3433333333333333,
704
+ "orig/kl_teacher": 0.25423722638876495,
705
+ "remove3/acc": 0.37916666666666665,
706
+ "remove3/among_k": 0.4325,
707
+ "remove3/kl_teacher": 0.9529805054700989,
708
+ "add3_unrel/acc": 0.34,
709
+ "add3_unrel/among_k": 0.3408333333333333,
710
+ "add3_unrel/kl_teacher": 0.32834803689669667,
711
+ "replace_hard/acc": 0.3475,
712
+ "replace_hard/among_k": 0.35,
713
+ "replace_hard/kl_teacher": 0.2868563655580847,
714
+ "add_neardup/acc": 0.26,
715
+ "add_neardup/among_k": 0.26166666666666666,
716
+ "add_neardup/kl_teacher": 0.2687358245348599,
717
+ "reorder/acc": 0.3441666666666667,
718
+ "reorder/among_k": 0.3466666666666667,
719
+ "reorder/kl_teacher": 0.24609989547374833,
720
+ "remove3/delta_mae": 1.1517046652173903,
721
+ "remove3/delta_teacher_abs": 1.135840148062828,
722
+ "add3_unrel/delta_mae": 0.7575456402215782,
723
+ "add3_unrel/delta_teacher_abs": 0.7390631803199154,
724
+ "replace_hard/delta_mae": 0.7262603232459742,
725
+ "replace_hard/delta_teacher_abs": 0.6681882120617108,
726
+ "add_neardup/delta_mae": 0.7387218217655764,
727
+ "add_neardup/delta_teacher_abs": 0.7305883395327554,
728
+ "reorder/delta_mae": 0.9534334357618881,
729
+ "reorder/delta_teacher_abs": 0.7763704285195197,
730
+ "add3_unrel/dlo_top2": 0.1328711094665931,
731
+ "reorder/max_dp": 0.11963965538597525,
732
+ "add_neardup/slot_gap": 0.10855303687150201,
733
+ "add_neardup/cluster_mass_err": 0.0905932489552591
734
+ }
735
+ },
736
+ "mmlu_choicesonly": {
737
+ "raw": {
738
+ "nll": 2.132304014679386,
739
+ "brier": 0.855733086948981,
740
+ "jsd": 0.4787010918172532,
741
+ "acc": 0.22666666666666666,
742
+ "acc_k": 0.22666666666666666,
743
+ "ece": 0.05177726411571105,
744
+ "sel_acc80": 0.25625,
745
+ "coverage": 0.9975,
746
+ "false_abstain": 0.0025,
747
+ "null_recall": "n/a"
748
+ },
749
+ "scaled": {
750
+ "nll": 2.119891606203886,
751
+ "brier": 0.8533568513213663,
752
+ "jsd": 0.4808686117250174,
753
+ "acc": 0.22666666666666666,
754
+ "acc_k": 0.22666666666666666,
755
+ "ece": 0.03572918727373083,
756
+ "sel_acc80": 0.25625,
757
+ "coverage": 0.9975,
758
+ "false_abstain": 0.0025,
759
+ "null_recall": "n/a"
760
+ }
761
+ },
762
+ "mmlu_pro": {
763
+ "raw": {
764
+ "nll": 1.870684202959633,
765
+ "brier": 0.7778605752958676,
766
+ "jsd": 0.4253066819218272,
767
+ "acc": 0.3425,
768
+ "acc_k": 0.3433333333333333,
769
+ "ece": 0.04600312157223621,
770
+ "sel_acc80": 0.39166666666666666,
771
+ "coverage": 0.995,
772
+ "false_abstain": 0.005,
773
+ "null_recall": "n/a"
774
+ },
775
+ "scaled": {
776
+ "nll": 1.8650622347991608,
777
+ "brier": 0.7769553003598431,
778
+ "jsd": 0.4300648104283429,
779
+ "acc": 0.3425,
780
+ "acc_k": 0.3433333333333333,
781
+ "ece": 0.04282940872013569,
782
+ "sel_acc80": 0.39166666666666666,
783
+ "coverage": 0.995,
784
+ "false_abstain": 0.005,
785
+ "null_recall": "n/a"
786
+ }
787
+ },
788
+ "mmlu_shuffledq": {
789
+ "raw": {
790
+ "nll": 2.2841442391906717,
791
+ "brier": 0.8968269319049337,
792
+ "jsd": 0.49137674516409957,
793
+ "acc": 0.21083333333333334,
794
+ "acc_k": 0.21666666666666667,
795
+ "ece": 0.09170151713614662,
796
+ "sel_acc80": 0.23125,
797
+ "coverage": 0.9775,
798
+ "false_abstain": 0.0225,
799
+ "null_recall": "n/a"
800
+ },
801
+ "scaled": {
802
+ "nll": 2.2550489918898133,
803
+ "brier": 0.8906454120623264,
804
+ "jsd": 0.492517543247486,
805
+ "acc": 0.21083333333333334,
806
+ "acc_k": 0.21666666666666667,
807
+ "ece": 0.0703738091637691,
808
+ "sel_acc80": 0.23125,
809
+ "coverage": 0.9766666666666667,
810
+ "false_abstain": 0.023333333333333334,
811
+ "null_recall": "n/a"
812
+ }
813
+ },
814
+ "mnli_val": {
815
+ "raw": {
816
+ "nll": 0.38298786152412717,
817
+ "brier": 0.2116872493208543,
818
+ "jsd": 0.09760297139339706,
819
+ "acc": 0.8591836734693877,
820
+ "acc_k": 0.8591836734693877,
821
+ "ece": 0.016026962665068034,
822
+ "sel_acc80": 0.9226190476190477,
823
+ "coverage": 1.0,
824
+ "false_abstain": 0.0,
825
+ "null_recall": "n/a"
826
+ },
827
+ "scaled": {
828
+ "nll": 0.3825671840790251,
829
+ "brier": 0.21233959847225758,
830
+ "jsd": 0.1018273888841336,
831
+ "acc": 0.8591836734693877,
832
+ "acc_k": 0.8591836734693877,
833
+ "ece": 0.02029442191123957,
834
+ "sel_acc80": 0.9226190476190477,
835
+ "coverage": 1.0,
836
+ "false_abstain": 0.0,
837
+ "null_recall": "n/a"
838
+ }
839
+ },
840
+ "ng20_test": {
841
+ "raw": {
842
+ "nll": 1.594920015134779,
843
+ "brier": 0.6067909948170968,
844
+ "jsd": 0.30844571120119146,
845
+ "acc": 0.5449077238550922,
846
+ "acc_k": 0.5807245386192754,
847
+ "ece": 0.07311073689328751,
848
+ "sel_acc80": 0.6211551606288448,
849
+ "coverage": 0.8474367737525632,
850
+ "false_abstain": 0.15256322624743676,
851
+ "null_recall": "n/a"
852
+ },
853
+ "scaled": {
854
+ "nll": 1.56044065734942,
855
+ "brier": 0.6067970962992972,
856
+ "jsd": 0.31531418996922944,
857
+ "acc": 0.5403964456596035,
858
+ "acc_k": 0.5807245386192754,
859
+ "ece": 0.051888617539112475,
860
+ "sel_acc80": 0.6134654818865345,
861
+ "coverage": 0.833902939166097,
862
+ "false_abstain": 0.16609706083390294,
863
+ "null_recall": "n/a"
864
+ }
865
+ },
866
+ "null_irrq_intent": {
867
+ "raw": {
868
+ "nll": 0.28509880355066847,
869
+ "brier": 0.12926910897196567,
870
+ "jsd": 0.07821245099591534,
871
+ "acc": 0.9241658240647118,
872
+ "acc_k": 0.9683673469387755,
873
+ "ece": 0.07439453771791758,
874
+ "sel_acc80": 0.9747315224257739,
875
+ "conf_wrong": 0.001002004008016032,
876
+ "coverage": 0.519211324570273,
877
+ "false_abstain": 0.04285714285714286,
878
+ "null_recall": 0.9108216432865731,
879
+ "auroc_null": 0.9787196842664921,
880
+ "auroc_conf": 0.9848288413561817
881
+ },
882
+ "scaled": {
883
+ "nll": 0.28319577466198464,
884
+ "brier": 0.1267643414616399,
885
+ "jsd": 0.07851246372610352,
886
+ "acc": 0.9282103134479271,
887
+ "acc_k": 0.9683673469387755,
888
+ "ece": 0.08138514553451921,
889
+ "sel_acc80": 0.9766266582438408,
890
+ "conf_wrong": 0.0,
891
+ "coverage": 0.5146612740141557,
892
+ "false_abstain": 0.04387755102040816,
893
+ "null_recall": 0.9188376753507014,
894
+ "auroc_null": 0.9787196842664921,
895
+ "auroc_conf": 0.9855026379289191
896
+ }
897
+ },
898
+ "null_irrq_squad": {
899
+ "raw": {
900
+ "nll": 0.10103190551973668,
901
+ "brier": 0.05798018973415887,
902
+ "jsd": 0.029012514334021317,
903
+ "acc": 0.9585,
904
+ "acc_k": 0.995,
905
+ "ece": 0.011917174428701351,
906
+ "sel_acc80": 0.998125,
907
+ "conf_wrong": 0.007,
908
+ "coverage": 0.522,
909
+ "false_abstain": 0.017,
910
+ "null_recall": 0.939,
911
+ "auroc_null": 0.996387,
912
+ "auroc_conf": 0.9962260000000001
913
+ },
914
+ "scaled": {
915
+ "nll": 0.1011740032118126,
916
+ "brier": 0.05775404134531006,
917
+ "jsd": 0.02910070699993618,
918
+ "acc": 0.959,
919
+ "acc_k": 0.995,
920
+ "ece": 0.011337469652295146,
921
+ "sel_acc80": 0.99875,
922
+ "conf_wrong": 0.007,
923
+ "coverage": 0.5215000000000001,
924
+ "false_abstain": 0.017,
925
+ "null_recall": 0.94,
926
+ "auroc_null": 0.996387,
927
+ "auroc_conf": 0.99622
928
+ }
929
+ },
930
+ "null_nearmiss_banking77": {
931
+ "raw": {
932
+ "nll": 0.9477015185824751,
933
+ "brier": 0.48575886111373834,
934
+ "jsd": 0.20246781100242447,
935
+ "acc": 0.672,
936
+ "acc_k": 0.88,
937
+ "ece": 0.14288236629962925,
938
+ "sel_acc80": 0.71625,
939
+ "conf_wrong": 0.12,
940
+ "coverage": 0.7090000000000001,
941
+ "false_abstain": 0.078,
942
+ "null_recall": 0.504,
943
+ "auroc_null": 0.8642439999999999,
944
+ "auroc_conf": 0.848276
945
+ },
946
+ "scaled": {
947
+ "nll": 0.9372510745375565,
948
+ "brier": 0.4799510588310396,
949
+ "jsd": 0.20350421208345068,
950
+ "acc": 0.674,
951
+ "acc_k": 0.88,
952
+ "ece": 0.13095481739938258,
953
+ "sel_acc80": 0.71875,
954
+ "conf_wrong": 0.116,
955
+ "coverage": 0.706,
956
+ "false_abstain": 0.08,
957
+ "null_recall": 0.508,
958
+ "auroc_null": 0.8642439999999999,
959
+ "auroc_conf": 0.847356
960
+ }
961
+ },
962
+ "null_nearmiss_clinc": {
963
+ "raw": {
964
+ "nll": 0.7261187274036407,
965
+ "brier": 0.382317397417127,
966
+ "jsd": 0.1502412147763253,
967
+ "acc": 0.756838905775076,
968
+ "acc_k": 0.9494949494949495,
969
+ "ece": 0.1270041166528259,
970
+ "sel_acc80": 0.7949367088607595,
971
+ "conf_wrong": 0.17886178861788618,
972
+ "coverage": 0.6808510638297872,
973
+ "false_abstain": 0.04242424242424243,
974
+ "null_recall": 0.5975609756097561,
975
+ "auroc_null": 0.9221318879855465,
976
+ "auroc_conf": 0.89660014781966
977
+ },
978
+ "scaled": {
979
+ "nll": 0.7226177400781414,
980
+ "brier": 0.3803847641898642,
981
+ "jsd": 0.15083584162884187,
982
+ "acc": 0.7588652482269503,
983
+ "acc_k": 0.9494949494949495,
984
+ "ece": 0.1236053464260507,
985
+ "sel_acc80": 0.7949367088607595,
986
+ "conf_wrong": 0.17886178861788618,
987
+ "coverage": 0.6788247213779128,
988
+ "false_abstain": 0.04242424242424243,
989
+ "null_recall": 0.6016260162601627,
990
+ "auroc_null": 0.9221318879855465,
991
+ "auroc_conf": 0.8943048369877638
992
+ }
993
+ },
994
+ "null_nearmiss_hwu64": {
995
+ "raw": {
996
+ "nll": 0.8017006188180456,
997
+ "brier": 0.44291562677417917,
998
+ "jsd": 0.18860549976513902,
999
+ "acc": 0.689,
1000
+ "acc_k": 0.906,
1001
+ "ece": 0.11632219871878623,
1002
+ "sel_acc80": 0.74125,
1003
+ "conf_wrong": 0.12,
1004
+ "coverage": 0.728,
1005
+ "false_abstain": 0.048,
1006
+ "null_recall": 0.496,
1007
+ "auroc_null": 0.8993880000000001,
1008
+ "auroc_conf": 0.8742839999999998
1009
+ },
1010
+ "scaled": {
1011
+ "nll": 0.8007376150333095,
1012
+ "brier": 0.43729978598013314,
1013
+ "jsd": 0.1897504311153405,
1014
+ "acc": 0.689,
1015
+ "acc_k": 0.906,
1016
+ "ece": 0.10554402610659597,
1017
+ "sel_acc80": 0.745,
1018
+ "conf_wrong": 0.108,
1019
+ "coverage": 0.725,
1020
+ "false_abstain": 0.052,
1021
+ "null_recall": 0.498,
1022
+ "auroc_null": 0.8993880000000001,
1023
+ "auroc_conf": 0.8721200000000001
1024
+ }
1025
+ },
1026
+ "snli_null": {
1027
+ "raw": {
1028
+ "nll": 0.36114479966903223,
1029
+ "brier": 0.19934539594747502,
1030
+ "jsd": 0.09212641990572218,
1031
+ "acc": 0.8649027861294849,
1032
+ "acc_k": 0.9433198380566802,
1033
+ "ece": 0.014997484241226794,
1034
+ "sel_acc80": 0.9321142284569138,
1035
+ "conf_wrong": 0.03136165144898769,
1036
+ "coverage": 0.581479254359591,
1037
+ "false_abstain": 0.022672064777327937,
1038
+ "null_recall": 0.8066693132195316,
1039
+ "auroc_null": 0.9770666015207499,
1040
+ "auroc_conf": 0.9634363453140746
1041
+ },
1042
+ "scaled": {
1043
+ "nll": 0.36211979227144847,
1044
+ "brier": 0.19853896379199165,
1045
+ "jsd": 0.09328289433312073,
1046
+ "acc": 0.8653036680697535,
1047
+ "acc_k": 0.9433198380566802,
1048
+ "ece": 0.014350581560996995,
1049
+ "sel_acc80": 0.9333667334669339,
1050
+ "conf_wrong": 0.02818578801111552,
1051
+ "coverage": 0.5810783724193225,
1052
+ "false_abstain": 0.022672064777327937,
1053
+ "null_recall": 0.8074632790789996,
1054
+ "auroc_null": 0.9770666015207501,
1055
+ "auroc_conf": 0.9622099573604974
1056
+ }
1057
+ },
1058
+ "snli_test": {
1059
+ "raw": {
1060
+ "nll": 0.284213361907045,
1061
+ "brier": 0.15704357356528473,
1062
+ "jsd": 0.07379276085972616,
1063
+ "acc": 0.894167167769092,
1064
+ "acc_k": 0.894167167769092,
1065
+ "ece": 0.008250219062326106,
1066
+ "sel_acc80": 0.9561623246492986,
1067
+ "coverage": 1.0,
1068
+ "false_abstain": 0.0,
1069
+ "null_recall": "n/a"
1070
+ },
1071
+ "scaled": {
1072
+ "nll": 0.2871341337130327,
1073
+ "brier": 0.1579343204560731,
1074
+ "jsd": 0.07789926571026709,
1075
+ "acc": 0.894167167769092,
1076
+ "acc_k": 0.894167167769092,
1077
+ "ece": 0.02058065520615534,
1078
+ "sel_acc80": 0.9556613226452906,
1079
+ "coverage": 1.0,
1080
+ "false_abstain": 0.0,
1081
+ "null_recall": "n/a"
1082
+ }
1083
+ },
1084
+ "snli_test_hyponly": {
1085
+ "raw": {
1086
+ "nll": 1.7306926384767332,
1087
+ "brier": 0.8538196615095388,
1088
+ "jsd": 0.3389282732869213,
1089
+ "acc": 0.39787532571657647,
1090
+ "acc_k": 0.39787532571657647,
1091
+ "ece": 0.3512127329314441,
1092
+ "sel_acc80": 0.4155811623246493,
1093
+ "coverage": 1.0,
1094
+ "false_abstain": 0.0,
1095
+ "null_recall": "n/a"
1096
+ },
1097
+ "scaled": {
1098
+ "nll": 1.605434509078102,
1099
+ "brier": 0.8275545011588878,
1100
+ "jsd": 0.3336658913312569,
1101
+ "acc": 0.39787532571657647,
1102
+ "acc_k": 0.39787532571657647,
1103
+ "ece": 0.3278745979004954,
1104
+ "sel_acc80": 0.4168336673346693,
1105
+ "coverage": 1.0,
1106
+ "false_abstain": 0.0,
1107
+ "null_recall": "n/a"
1108
+ }
1109
+ },
1110
+ "snli_test_paraphrase": {
1111
+ "raw": {
1112
+ "nll": 0.33001314107494534,
1113
+ "brier": 0.17626339669192392,
1114
+ "jsd": 0.093560031291203,
1115
+ "acc": 0.8879534976949288,
1116
+ "acc_k": 0.8879534976949288,
1117
+ "ece": 0.056906451303702234,
1118
+ "sel_acc80": 0.9493987975951904,
1119
+ "coverage": 1.0,
1120
+ "false_abstain": 0.0,
1121
+ "null_recall": "n/a"
1122
+ },
1123
+ "scaled": {
1124
+ "nll": 0.345213491059501,
1125
+ "brier": 0.18243332009021412,
1126
+ "jsd": 0.10055492307245627,
1127
+ "acc": 0.8879534976949288,
1128
+ "acc_k": 0.8879534976949288,
1129
+ "ece": 0.07956015918531549,
1130
+ "sel_acc80": 0.9483967935871743,
1131
+ "coverage": 1.0,
1132
+ "false_abstain": 0.0,
1133
+ "null_recall": "n/a"
1134
+ }
1135
+ },
1136
+ "snli_test_qpara": {
1137
+ "raw": {
1138
+ "nll": 0.28813321351334215,
1139
+ "brier": 0.15915075021311392,
1140
+ "jsd": 0.07611697804220868,
1141
+ "acc": 0.8916750250752257,
1142
+ "acc_k": 0.8916750250752257,
1143
+ "ece": 0.011274306875174358,
1144
+ "sel_acc80": 0.9530075187969925,
1145
+ "coverage": 1.0,
1146
+ "false_abstain": 0.0,
1147
+ "null_recall": "n/a"
1148
+ },
1149
+ "scaled": {
1150
+ "nll": 0.2928096177722551,
1151
+ "brier": 0.1605470109592482,
1152
+ "jsd": 0.08055557756133541,
1153
+ "acc": 0.8916750250752257,
1154
+ "acc_k": 0.8916750250752257,
1155
+ "ece": 0.03255550470670221,
1156
+ "sel_acc80": 0.9523809523809523,
1157
+ "coverage": 1.0,
1158
+ "false_abstain": 0.0,
1159
+ "null_recall": "n/a"
1160
+ }
1161
+ },
1162
+ "snli_test_soft": {
1163
+ "raw": {
1164
+ "nll": 0.5765725980284668,
1165
+ "brier": 0.11069785863315745,
1166
+ "jsd": 0.058387127716102265,
1167
+ "acc": 0.894167167769092,
1168
+ "acc_k": 0.894167167769092,
1169
+ "ece": 0.008250219062326106,
1170
+ "sel_acc80": 0.9561623246492986,
1171
+ "coverage": 1.0,
1172
+ "false_abstain": 0.0,
1173
+ "null_recall": "n/a"
1174
+ },
1175
+ "scaled": {
1176
+ "nll": 0.5471916195654647,
1177
+ "brier": 0.1071951342615467,
1178
+ "jsd": 0.05873654566137795,
1179
+ "acc": 0.894167167769092,
1180
+ "acc_k": 0.894167167769092,
1181
+ "ece": 0.02058065520615534,
1182
+ "sel_acc80": 0.9556613226452906,
1183
+ "coverage": 1.0,
1184
+ "false_abstain": 0.0,
1185
+ "null_recall": "n/a"
1186
+ }
1187
+ },
1188
+ "squad_null": {
1189
+ "raw": {
1190
+ "nll": 0.08034793192118085,
1191
+ "brier": 0.04045550145126374,
1192
+ "jsd": 0.01782847579404271,
1193
+ "acc": 0.97425,
1194
+ "acc_k": 0.997,
1195
+ "ece": 0.006439283616840791,
1196
+ "sel_acc80": 0.995,
1197
+ "conf_wrong": 0.0165,
1198
+ "coverage": 0.51125,
1199
+ "false_abstain": 0.013,
1200
+ "null_recall": 0.9645,
1201
+ "auroc_null": 0.994786,
1202
+ "auroc_conf": 0.9945622500000002
1203
+ },
1204
+ "scaled": {
1205
+ "nll": 0.08038613771297327,
1206
+ "brier": 0.0404679182690968,
1207
+ "jsd": 0.017924676484503386,
1208
+ "acc": 0.97425,
1209
+ "acc_k": 0.997,
1210
+ "ece": 0.006344715304672706,
1211
+ "sel_acc80": 0.995,
1212
+ "conf_wrong": 0.0165,
1213
+ "coverage": 0.51125,
1214
+ "false_abstain": 0.013,
1215
+ "null_recall": 0.9645,
1216
+ "auroc_null": 0.994786,
1217
+ "auroc_conf": 0.9945115000000001
1218
+ }
1219
+ },
1220
+ "trec_coarse": {
1221
+ "raw": {
1222
+ "nll": 1.4410608559203524,
1223
+ "brier": 0.5647949710098153,
1224
+ "jsd": 0.2594204842852714,
1225
+ "acc": 0.598,
1226
+ "acc_k": 0.64,
1227
+ "ece": 0.13209271228313446,
1228
+ "sel_acc80": 0.6525,
1229
+ "coverage": 0.858,
1230
+ "false_abstain": 0.142,
1231
+ "null_recall": "n/a"
1232
+ },
1233
+ "scaled": {
1234
+ "nll": 1.3737689758814966,
1235
+ "brier": 0.5630469420780047,
1236
+ "jsd": 0.2625928078499603,
1237
+ "acc": 0.598,
1238
+ "acc_k": 0.64,
1239
+ "ece": 0.1101637490391731,
1240
+ "sel_acc80": 0.65,
1241
+ "coverage": 0.856,
1242
+ "false_abstain": 0.144,
1243
+ "null_recall": "n/a"
1244
+ }
1245
+ },
1246
+ "trec_fine": {
1247
+ "raw": {
1248
+ "nll": 1.8286524021989985,
1249
+ "brier": 0.6856642809905157,
1250
+ "jsd": 0.33030271212248546,
1251
+ "acc": 0.516,
1252
+ "acc_k": 0.556,
1253
+ "ece": 0.11209276500344276,
1254
+ "sel_acc80": 0.5675,
1255
+ "coverage": 0.866,
1256
+ "false_abstain": 0.134,
1257
+ "null_recall": "n/a"
1258
+ },
1259
+ "scaled": {
1260
+ "nll": 1.7610366472313366,
1261
+ "brier": 0.6826349232338287,
1262
+ "jsd": 0.33616273860591916,
1263
+ "acc": 0.508,
1264
+ "acc_k": 0.556,
1265
+ "ece": 0.099832693696022,
1266
+ "sel_acc80": 0.56,
1267
+ "coverage": 0.85,
1268
+ "false_abstain": 0.15,
1269
+ "null_recall": "n/a"
1270
+ }
1271
+ },
1272
+ "unli_test": {
1273
+ "raw": {
1274
+ "nll": 0.5229310656880769,
1275
+ "brier": 0.12426689562340587,
1276
+ "jsd": 0.04685473064779179,
1277
+ "auroc_null": 0.8959275410935928,
1278
+ "auroc_conf": 0.8959275410935928
1279
+ },
1280
+ "scaled": {
1281
+ "nll": 0.5229310634122645,
1282
+ "brier": 0.124266895213907,
1283
+ "jsd": 0.04685473041870062,
1284
+ "auroc_null": 0.8959275410935928,
1285
+ "auroc_conf": 0.8959275410935928
1286
+ }
1287
+ },
1288
+ "data_kb_val": {
1289
+ "raw": {
1290
+ "nll": 0.9172184406482693,
1291
+ "brier": 0.4730059211402316,
1292
+ "jsd": 0.23839771802896959,
1293
+ "acc": 0.6173333333333333,
1294
+ "acc_k": 0.5984528832630098,
1295
+ "ece": 0.03232765065630277,
1296
+ "sel_acc80": 0.6975,
1297
+ "conf_wrong": 0.01282051282051282,
1298
+ "coverage": 0.9486666666666667,
1299
+ "false_abstain": 0.0007032348804500703,
1300
+ "null_recall": 0.9743589743589743,
1301
+ "auroc_null": 0.9997655883731833,
1302
+ "auroc_conf": 0.9838796927404522
1303
+ },
1304
+ "scaled": {
1305
+ "nll": 0.9193650865536224,
1306
+ "brier": 0.47460153079949197,
1307
+ "jsd": 0.24302611810488342,
1308
+ "acc": 0.6173333333333333,
1309
+ "acc_k": 0.5984528832630098,
1310
+ "ece": 0.03246050986647605,
1311
+ "sel_acc80": 0.6975,
1312
+ "conf_wrong": 0.01282051282051282,
1313
+ "coverage": 0.9486666666666667,
1314
+ "false_abstain": 0.0007032348804500703,
1315
+ "null_recall": 0.9743589743589743,
1316
+ "auroc_null": 0.9997655883731833,
1317
+ "auroc_conf": 0.9838255977496484
1318
+ }
1319
+ },
1320
+ "truthfulqa_mc1": {
1321
+ "raw": {
1322
+ "nll": 2.5604446606049165,
1323
+ "brier": 0.9727379180282026,
1324
+ "jsd": 0.44773304381342327,
1325
+ "acc": 0.25703794369645044,
1326
+ "acc_k": 0.32068543451652387,
1327
+ "ece": 0.27811915158422956,
1328
+ "sel_acc80": 0.2782874617737003,
1329
+ "coverage": 0.8763769889840881,
1330
+ "false_abstain": 0.12362301101591187,
1331
+ "null_recall": "n/a"
1332
+ },
1333
+ "scaled": {
1334
+ "nll": 2.48766464211813,
1335
+ "brier": 0.960296728999466,
1336
+ "jsd": 0.4468052354590374,
1337
+ "acc": 0.25703794369645044,
1338
+ "acc_k": 0.32068543451652387,
1339
+ "ece": 0.25495981387657224,
1340
+ "sel_acc80": 0.28134556574923547,
1341
+ "coverage": 0.8763769889840881,
1342
+ "false_abstain": 0.12362301101591187,
1343
+ "null_recall": "n/a"
1344
+ }
1345
+ },
1346
+ "data_wf_val": {
1347
+ "raw": {
1348
+ "nll": 0.37583161847228896,
1349
+ "brier": 0.20703477167290157,
1350
+ "jsd": 0.105633798882721,
1351
+ "acc": 0.8586666666666667,
1352
+ "acc_k": 0.8568507157464212,
1353
+ "ece": 0.05674472664793331,
1354
+ "sel_acc80": 0.9383333333333334,
1355
+ "conf_wrong": 0.0,
1356
+ "coverage": 0.9786666666666667,
1357
+ "false_abstain": 0.0006816632583503749,
1358
+ "null_recall": 0.9393939393939394,
1359
+ "auroc_null": 0.9999380306128771,
1360
+ "auroc_conf": 0.9979343537625747
1361
+ },
1362
+ "scaled": {
1363
+ "nll": 0.3865502429414146,
1364
+ "brier": 0.21285080875890586,
1365
+ "jsd": 0.11058233121499325,
1366
+ "acc": 0.8586666666666667,
1367
+ "acc_k": 0.8568507157464212,
1368
+ "ece": 0.07054323360323908,
1369
+ "sel_acc80": 0.9358333333333333,
1370
+ "conf_wrong": 0.0,
1371
+ "coverage": 0.9786666666666667,
1372
+ "false_abstain": 0.0006816632583503749,
1373
+ "null_recall": 0.9393939393939394,
1374
+ "auroc_null": 0.9999380306128771,
1375
+ "auroc_conf": 0.998223544235814
1376
+ },
1377
+ "ordinal": {
1378
+ "ordinal_mae": 0.2672811059907834,
1379
+ "exp_score_err": 0.4225480312823225,
1380
+ "n": 434
1381
+ },
1382
+ "noul_reversed": {
1383
+ "max_abs_diff": 0.0,
1384
+ "mean_abs_diff": 0.0
1385
+ }
1386
+ },
1387
+ "wf_heldout_choice": {
1388
+ "raw": {
1389
+ "nll": 0.02237159506845183,
1390
+ "brier": 0.009990251291453516,
1391
+ "jsd": 0.006968188747770273,
1392
+ "acc": 0.9946666666666667,
1393
+ "acc_k": 0.9957537154989384,
1394
+ "ece": 0.010020962238311813,
1395
+ "sel_acc80": 1.0,
1396
+ "conf_wrong": 0.0,
1397
+ "coverage": 0.9433333333333334,
1398
+ "false_abstain": 0.0,
1399
+ "null_recall": 0.9770114942528736,
1400
+ "auroc_null": 1.0,
1401
+ "auroc_conf": 0.9994143055860606
1402
+ },
1403
+ "scaled": {
1404
+ "nll": 0.02658096210267664,
1405
+ "brier": 0.0108593449895006,
1406
+ "jsd": 0.00840652061124444,
1407
+ "acc": 0.9946666666666667,
1408
+ "acc_k": 0.9957537154989384,
1409
+ "ece": 0.01406175897518795,
1410
+ "sel_acc80": 1.0,
1411
+ "conf_wrong": 0.0,
1412
+ "coverage": 0.9433333333333334,
1413
+ "false_abstain": 0.0,
1414
+ "null_recall": 0.9770114942528736,
1415
+ "auroc_null": 1.0,
1416
+ "auroc_conf": 0.9993492284289561
1417
+ }
1418
+ },
1419
+ "wf_heldout_noul": {
1420
+ "raw": {
1421
+ "nll": 0.6509058026223263,
1422
+ "brier": 0.4112349650016078,
1423
+ "jsd": 0.1613226242259403,
1424
+ "acc": 0.71,
1425
+ "acc_k": 0.71,
1426
+ "ece": 0.12073413105805718,
1427
+ "sel_acc80": 0.7566666666666667,
1428
+ "coverage": 1.0,
1429
+ "false_abstain": 0.0,
1430
+ "null_recall": "n/a"
1431
+ },
1432
+ "scaled": {
1433
+ "nll": 0.6225472483414641,
1434
+ "brier": 0.4031137822900303,
1435
+ "jsd": 0.1609770686535907,
1436
+ "acc": 0.71,
1437
+ "acc_k": 0.71,
1438
+ "ece": 0.10552653594811762,
1439
+ "sel_acc80": 0.7566666666666667,
1440
+ "coverage": 1.0,
1441
+ "false_abstain": 0.0,
1442
+ "null_recall": "n/a"
1443
+ },
1444
+ "noul_reversed": {
1445
+ "max_abs_diff": 0.0,
1446
+ "mean_abs_diff": 0.0
1447
+ }
1448
+ },
1449
+ "wf_heldout_score": {
1450
+ "raw": {
1451
+ "nll": 1.0077400803811332,
1452
+ "brier": 0.573195142899105,
1453
+ "jsd": 0.2761805759176359,
1454
+ "acc": 0.522,
1455
+ "acc_k": 0.5226666666666666,
1456
+ "ece": 0.040125139494736985,
1457
+ "sel_acc80": 0.5766666666666667,
1458
+ "coverage": 0.9993333333333333,
1459
+ "false_abstain": 0.0006666666666666666,
1460
+ "null_recall": "n/a"
1461
+ },
1462
+ "scaled": {
1463
+ "nll": 1.010836001086897,
1464
+ "brier": 0.5717227830762833,
1465
+ "jsd": 0.2801758694525442,
1466
+ "acc": 0.522,
1467
+ "acc_k": 0.5226666666666666,
1468
+ "ece": 0.05136033900578818,
1469
+ "sel_acc80": 0.575,
1470
+ "coverage": 0.9993333333333333,
1471
+ "false_abstain": 0.0006666666666666666,
1472
+ "null_recall": "n/a"
1473
+ },
1474
+ "ordinal": {
1475
+ "ordinal_mae": 0.5693333333333334,
1476
+ "exp_score_err": 0.5842081432726915,
1477
+ "n": 1500
1478
+ }
1479
+ },
1480
+ "wf_heldout_style": {
1481
+ "raw": {
1482
+ "nll": 0.375583335464391,
1483
+ "brier": 0.21027239215428048,
1484
+ "jsd": 0.10794280937778464,
1485
+ "acc": 0.8606666666666667,
1486
+ "acc_k": 0.8584969532836831,
1487
+ "ece": 0.07419633140166605,
1488
+ "sel_acc80": 0.9383333333333334,
1489
+ "conf_wrong": 0.0,
1490
+ "coverage": 0.9846666666666667,
1491
+ "false_abstain": 0.0,
1492
+ "null_recall": 1.0,
1493
+ "auroc_null": 1.0,
1494
+ "auroc_conf": 1.0
1495
+ },
1496
+ "scaled": {
1497
+ "nll": 0.38826847038172296,
1498
+ "brier": 0.21664383115975253,
1499
+ "jsd": 0.11323410183609288,
1500
+ "acc": 0.8606666666666667,
1501
+ "acc_k": 0.8584969532836831,
1502
+ "ece": 0.08574621800581617,
1503
+ "sel_acc80": 0.9358333333333333,
1504
+ "conf_wrong": 0.0,
1505
+ "coverage": 0.9846666666666667,
1506
+ "false_abstain": 0.0,
1507
+ "null_recall": 1.0,
1508
+ "auroc_null": 1.0,
1509
+ "auroc_conf": 1.0
1510
+ },
1511
+ "ordinal": {
1512
+ "ordinal_mae": 0.2709677419354839,
1513
+ "exp_score_err": 0.44317797320777375,
1514
+ "n": 465
1515
+ },
1516
+ "noul_reversed": {
1517
+ "max_abs_diff": 0.008650124073028564,
1518
+ "mean_abs_diff": 4.512667779076415e-05
1519
+ }
1520
+ },
1521
+ "wf_rubric_flip": {
1522
+ "raw": {
1523
+ "nll": 0.5544689662864979,
1524
+ "brier": 0.3214072550849167,
1525
+ "jsd": 0.14485513508589654,
1526
+ "acc": 0.7566666666666667,
1527
+ "acc_k": 0.7566666666666667,
1528
+ "ece": 0.03826767514149347,
1529
+ "sel_acc80": 0.8383333333333334,
1530
+ "coverage": 1.0,
1531
+ "false_abstain": 0.0,
1532
+ "null_recall": "n/a"
1533
+ },
1534
+ "scaled": {
1535
+ "nll": 0.547252401613635,
1536
+ "brier": 0.3193832961798829,
1537
+ "jsd": 0.14676300963290417,
1538
+ "acc": 0.7566666666666667,
1539
+ "acc_k": 0.7566666666666667,
1540
+ "ece": 0.04043171815077463,
1541
+ "sel_acc80": 0.8391666666666666,
1542
+ "coverage": 1.0,
1543
+ "false_abstain": 0.0,
1544
+ "null_recall": "n/a"
1545
+ },
1546
+ "flip": {
1547
+ "n_pairs": 750,
1548
+ "flip_rate": 0.7573333333333333,
1549
+ "both_correct": 0.5866666666666667,
1550
+ "acc": 0.7566666666666667
1551
+ },
1552
+ "ordinal": {
1553
+ "ordinal_mae": 0.547808764940239,
1554
+ "exp_score_err": 0.5861706615906308,
1555
+ "n": 502
1556
+ },
1557
+ "noul_reversed": {
1558
+ "max_abs_diff": 0.0,
1559
+ "mean_abs_diff": 0.0
1560
+ }
1561
+ },
1562
+ "wf_rubric_shuffled": {
1563
+ "raw": {
1564
+ "nll": 2.583553418746041,
1565
+ "brier": 0.928155771024346,
1566
+ "jsd": 0.3631805122798969,
1567
+ "acc": 0.376,
1568
+ "acc_k": 0.376,
1569
+ "ece": 0.36781600866715114,
1570
+ "sel_acc80": 0.3925,
1571
+ "coverage": 1.0,
1572
+ "false_abstain": 0.0,
1573
+ "null_recall": "n/a"
1574
+ },
1575
+ "scaled": {
1576
+ "nll": 2.364992132114415,
1577
+ "brier": 0.9104185657935674,
1578
+ "jsd": 0.3598282600140078,
1579
+ "acc": 0.376,
1580
+ "acc_k": 0.376,
1581
+ "ece": 0.3502550947666168,
1582
+ "sel_acc80": 0.39416666666666667,
1583
+ "coverage": 1.0,
1584
+ "false_abstain": 0.0,
1585
+ "null_recall": "n/a"
1586
+ },
1587
+ "ordinal": {
1588
+ "ordinal_mae": 0.9470899470899471,
1589
+ "exp_score_err": 0.8813899273937501,
1590
+ "n": 567
1591
+ },
1592
+ "noul_reversed": {
1593
+ "max_abs_diff": 0.0,
1594
+ "mean_abs_diff": 0.0
1595
+ }
1596
+ },
1597
+ "data_wf_hf_val": {
1598
+ "raw": {
1599
+ "nll": 0.019401891005933914,
1600
+ "brier": 0.009413470640887905,
1601
+ "jsd": 0.004283260887802352,
1602
+ "acc": 0.9946666666666667,
1603
+ "acc_k": 0.9946666666666667,
1604
+ "ece": 0.0031396085421244255,
1605
+ "sel_acc80": 1.0,
1606
+ "coverage": 1.0,
1607
+ "false_abstain": 0.0,
1608
+ "null_recall": "n/a"
1609
+ },
1610
+ "scaled": {
1611
+ "nll": 0.01934158623401211,
1612
+ "brier": 0.009498351599644392,
1613
+ "jsd": 0.004642312540112629,
1614
+ "acc": 0.9946666666666667,
1615
+ "acc_k": 0.9946666666666667,
1616
+ "ece": 0.0030115491151810297,
1617
+ "sel_acc80": 1.0,
1618
+ "coverage": 1.0,
1619
+ "false_abstain": 0.0,
1620
+ "null_recall": "n/a"
1621
+ }
1622
+ },
1623
+ "cua_s1_forms_demo": {
1624
+ "raw": {
1625
+ "nll": 0.0009300660857717562,
1626
+ "brier": 1.7813199239036247e-05,
1627
+ "jsd": 0.0003218505523506755,
1628
+ "acc": 1.0,
1629
+ "acc_k": 1.0,
1630
+ "ece": 0.0009250634787033585,
1631
+ "sel_acc80": 1.0,
1632
+ "coverage": 1.0,
1633
+ "false_abstain": 0.0,
1634
+ "null_recall": "n/a"
1635
+ },
1636
+ "scaled": {
1637
+ "nll": 0.0018790738508953008,
1638
+ "brier": 4.7467508047515636e-05,
1639
+ "jsd": 0.0006498546450564701,
1640
+ "acc": 1.0,
1641
+ "acc_k": 1.0,
1642
+ "ece": 0.0018648833644633322,
1643
+ "sel_acc80": 1.0,
1644
+ "coverage": 1.0,
1645
+ "false_abstain": 0.0,
1646
+ "null_recall": "n/a"
1647
+ }
1648
+ },
1649
+ "cua_s1_forms_test": {
1650
+ "raw": {
1651
+ "nll": 0.032046872477530054,
1652
+ "brier": 0.017368318078177302,
1653
+ "jsd": 0.006974812412770343,
1654
+ "acc": 0.9873333333333333,
1655
+ "acc_k": 0.9873333333333333,
1656
+ "ece": 0.006146776457627566,
1657
+ "sel_acc80": 1.0,
1658
+ "coverage": 1.0,
1659
+ "false_abstain": 0.0,
1660
+ "null_recall": "n/a"
1661
+ },
1662
+ "scaled": {
1663
+ "nll": 0.03077882151868218,
1664
+ "brier": 0.017053987064367247,
1665
+ "jsd": 0.007157656705056871,
1666
+ "acc": 0.9873333333333333,
1667
+ "acc_k": 0.9873333333333333,
1668
+ "ece": 0.006392546931902525,
1669
+ "sel_acc80": 1.0,
1670
+ "coverage": 1.0,
1671
+ "false_abstain": 0.0,
1672
+ "null_recall": "n/a"
1673
+ }
1674
+ },
1675
+ "jev4b_adversarial": {
1676
+ "raw": {
1677
+ "nll": 0.40543914653847096,
1678
+ "brier": 0.21213837393283155,
1679
+ "jsd": 0.10257905139738542,
1680
+ "acc": 0.8566666666666667,
1681
+ "acc_k": 0.8566666666666667,
1682
+ "ece": 0.04504729241132734,
1683
+ "sel_acc80": 0.9458333333333333,
1684
+ "coverage": 1.0,
1685
+ "false_abstain": 0.0,
1686
+ "null_recall": "n/a"
1687
+ },
1688
+ "scaled": {
1689
+ "nll": 0.40307816177058725,
1690
+ "brier": 0.21442104499654838,
1691
+ "jsd": 0.10539861463346857,
1692
+ "acc": 0.8566666666666667,
1693
+ "acc_k": 0.8566666666666667,
1694
+ "ece": 0.05155258302887279,
1695
+ "sel_acc80": 0.9458333333333333,
1696
+ "coverage": 1.0,
1697
+ "false_abstain": 0.0,
1698
+ "null_recall": "n/a"
1699
+ },
1700
+ "ordinal": {
1701
+ "ordinal_mae": 0.25833333333333336,
1702
+ "exp_score_err": 0.44842133761425884,
1703
+ "n": 120
1704
+ },
1705
+ "noul_reversed": {
1706
+ "max_abs_diff": 0.0,
1707
+ "mean_abs_diff": 0.0
1708
+ }
1709
+ },
1710
+ "jevlogs_triage": {
1711
+ "raw": {
1712
+ "nll": 1.371274447707115,
1713
+ "brier": 0.8498478108011697,
1714
+ "jsd": 0.29231064922915007,
1715
+ "acc": 0.5,
1716
+ "acc_k": 0.5,
1717
+ "ece": 0.4159393855730693,
1718
+ "sel_acc80": 0.5116666666666667,
1719
+ "coverage": 1.0,
1720
+ "false_abstain": 0.0,
1721
+ "null_recall": "n/a"
1722
+ },
1723
+ "scaled": {
1724
+ "nll": 1.252508511219465,
1725
+ "brier": 0.8168997236293741,
1726
+ "jsd": 0.28321615671899836,
1727
+ "acc": 0.5,
1728
+ "acc_k": 0.5,
1729
+ "ece": 0.39538653079668684,
1730
+ "sel_acc80": 0.5116666666666667,
1731
+ "coverage": 1.0,
1732
+ "false_abstain": 0.0,
1733
+ "null_recall": "n/a"
1734
+ },
1735
+ "noul_reversed": {
1736
+ "max_abs_diff": 0.0,
1737
+ "mean_abs_diff": 0.0
1738
+ }
1739
+ },
1740
+ "mind2web_choice": {
1741
+ "raw": {
1742
+ "nll": 1.5252458937008002,
1743
+ "brier": 0.7495051757484386,
1744
+ "jsd": 0.3487572217093445,
1745
+ "acc": 0.35133333333333333,
1746
+ "acc_k": 0.354,
1747
+ "ece": 0.20950158571203548,
1748
+ "sel_acc80": 0.38,
1749
+ "coverage": 0.9953333333333333,
1750
+ "false_abstain": 0.004666666666666667,
1751
+ "null_recall": "n/a"
1752
+ },
1753
+ "scaled": {
1754
+ "nll": 1.492309018165003,
1755
+ "brier": 0.7332142893414118,
1756
+ "jsd": 0.34893241092756905,
1757
+ "acc": 0.3506666666666667,
1758
+ "acc_k": 0.354,
1759
+ "ece": 0.19169397105276587,
1760
+ "sel_acc80": 0.38,
1761
+ "coverage": 0.994,
1762
+ "false_abstain": 0.006,
1763
+ "null_recall": "n/a"
1764
+ },
1765
+ "noul_reversed": {
1766
+ "max_abs_diff": 0.0,
1767
+ "mean_abs_diff": 0.0
1768
+ }
1769
+ },
1770
+ "pagerduty_trigger": {
1771
+ "raw": {
1772
+ "nll": 0.9310243398123302,
1773
+ "brier": 0.6155340506142242,
1774
+ "jsd": 0.23931832213418777,
1775
+ "acc": 0.56,
1776
+ "acc_k": 0.56,
1777
+ "ece": 0.25797174461682637,
1778
+ "sel_acc80": 0.5675,
1779
+ "coverage": 1.0,
1780
+ "false_abstain": 0.0,
1781
+ "null_recall": "n/a"
1782
+ },
1783
+ "scaled": {
1784
+ "nll": 0.8901599970352302,
1785
+ "brier": 0.5932938649817051,
1786
+ "jsd": 0.236096828884083,
1787
+ "acc": 0.56,
1788
+ "acc_k": 0.56,
1789
+ "ece": 0.27468403406937913,
1790
+ "sel_acc80": 0.5675,
1791
+ "coverage": 1.0,
1792
+ "false_abstain": 0.0,
1793
+ "null_recall": "n/a"
1794
+ },
1795
+ "noul_reversed": {
1796
+ "max_abs_diff": 0.0,
1797
+ "mean_abs_diff": 0.0
1798
+ }
1799
+ },
1800
+ "systemone_lite_hard": {
1801
+ "raw": {
1802
+ "nll": 0.4891307900271223,
1803
+ "brier": 0.30110738295596046,
1804
+ "jsd": 0.14204759917310972,
1805
+ "acc": 0.7786666666666666,
1806
+ "acc_k": 0.7786666666666666,
1807
+ "ece": 0.04153631351391474,
1808
+ "sel_acc80": 0.8391666666666666,
1809
+ "coverage": 1.0,
1810
+ "false_abstain": 0.0,
1811
+ "null_recall": "n/a"
1812
+ },
1813
+ "scaled": {
1814
+ "nll": 0.4946905537496238,
1815
+ "brier": 0.3048013448634015,
1816
+ "jsd": 0.1455159848651888,
1817
+ "acc": 0.7786666666666666,
1818
+ "acc_k": 0.7786666666666666,
1819
+ "ece": 0.054403663555781016,
1820
+ "sel_acc80": 0.8375,
1821
+ "coverage": 1.0,
1822
+ "false_abstain": 0.0,
1823
+ "null_recall": "n/a"
1824
+ },
1825
+ "ordinal": {
1826
+ "ordinal_mae": 0.332,
1827
+ "exp_score_err": 0.4491760475578015,
1828
+ "n": 500
1829
+ },
1830
+ "noul_reversed": {
1831
+ "max_abs_diff": 0.0,
1832
+ "mean_abs_diff": 0.0
1833
+ }
1834
+ },
1835
+ "tree_choice_cap": {
1836
+ "raw": {
1837
+ "nll": 1.9912322046858726,
1838
+ "brier": 0.776633232130246,
1839
+ "jsd": 0.3516807077694661,
1840
+ "acc": 0.4736111111111111,
1841
+ "acc_k": 0.7055555555555556,
1842
+ "ece": 0.2546793927076376,
1843
+ "sel_acc80": 0.5295138888888888,
1844
+ "coverage": 0.6069444444444445,
1845
+ "false_abstain": 0.39305555555555555,
1846
+ "null_recall": "n/a"
1847
+ },
1848
+ "scaled": {
1849
+ "nll": 2.037652770209596,
1850
+ "brier": 0.7879233374142699,
1851
+ "jsd": 0.36161560719221636,
1852
+ "acc": 0.46944444444444444,
1853
+ "acc_k": 0.7055555555555556,
1854
+ "ece": 0.25013160817842517,
1855
+ "sel_acc80": 0.5208333333333334,
1856
+ "coverage": 0.5916666666666667,
1857
+ "false_abstain": 0.4083333333333333,
1858
+ "null_recall": "n/a"
1859
+ }
1860
+ },
1861
+ "typed_decisions_test": {
1862
+ "raw": {
1863
+ "nll": 1.308004753461238,
1864
+ "brier": 0.2680494162037552,
1865
+ "jsd": 0.12193913841109022,
1866
+ "acc": 0.4786666666666667,
1867
+ "acc_k": 0.4786666666666667,
1868
+ "ece": 0.11097379342714944,
1869
+ "sel_acc80": 0.5066666666666667,
1870
+ "coverage": 1.0,
1871
+ "false_abstain": 0.0,
1872
+ "null_recall": "n/a"
1873
+ },
1874
+ "scaled": {
1875
+ "nll": 1.2651857313405155,
1876
+ "brier": 0.2577007533013979,
1877
+ "jsd": 0.11840589262700886,
1878
+ "acc": 0.4786666666666667,
1879
+ "acc_k": 0.4786666666666667,
1880
+ "ece": 0.09702249457438787,
1881
+ "sel_acc80": 0.5058333333333334,
1882
+ "coverage": 1.0,
1883
+ "false_abstain": 0.0,
1884
+ "null_recall": "n/a"
1885
+ },
1886
+ "ordinal": {
1887
+ "ordinal_mae": 0.8266666666666667,
1888
+ "exp_score_err": 0.8259128936029989,
1889
+ "n": 600
1890
+ },
1891
+ "noul_reversed": {
1892
+ "max_abs_diff": 0.0,
1893
+ "mean_abs_diff": 0.0
1894
+ }
1895
+ },
1896
+ "typed_decisions_train": {
1897
+ "raw": {
1898
+ "nll": 1.3072811380004457,
1899
+ "brier": 0.21419036544099046,
1900
+ "jsd": 0.09930688225243353,
1901
+ "acc": 0.362,
1902
+ "acc_k": 0.362,
1903
+ "ece": 0.14027928841114043,
1904
+ "sel_acc80": 0.38,
1905
+ "coverage": 0.9993333333333333,
1906
+ "false_abstain": 0.0006666666666666666,
1907
+ "null_recall": "n/a"
1908
+ },
1909
+ "scaled": {
1910
+ "nll": 1.2813463910992928,
1911
+ "brier": 0.20369022460075886,
1912
+ "jsd": 0.09572205414893548,
1913
+ "acc": 0.362,
1914
+ "acc_k": 0.362,
1915
+ "ece": 0.12121769414345425,
1916
+ "sel_acc80": 0.38,
1917
+ "coverage": 0.9993333333333333,
1918
+ "false_abstain": 0.0006666666666666666,
1919
+ "null_recall": "n/a"
1920
+ },
1921
+ "ordinal": {
1922
+ "ordinal_mae": 1.03,
1923
+ "exp_score_err": 0.8523423746745429,
1924
+ "n": 600
1925
+ },
1926
+ "noul_reversed": {
1927
+ "max_abs_diff": 0.0,
1928
+ "mean_abs_diff": 0.0
1929
+ }
1930
+ },
1931
+ "data_wh_val": {
1932
+ "raw": {
1933
+ "nll": 0.234013824826771,
1934
+ "brier": 0.13775511773788127,
1935
+ "jsd": 0.06296474492075899,
1936
+ "acc": 0.8973333333333333,
1937
+ "acc_k": 0.932381667918858,
1938
+ "ece": 0.020509268840153996,
1939
+ "sel_acc80": 0.9766666666666667,
1940
+ "conf_wrong": 0.0,
1941
+ "coverage": 0.8686666666666667,
1942
+ "false_abstain": 0.03906836964688205,
1943
+ "null_recall": 0.8579881656804734,
1944
+ "auroc_null": 0.9832799114426577,
1945
+ "auroc_conf": 0.9764780673871583
1946
+ },
1947
+ "scaled": {
1948
+ "nll": 0.23435895126674677,
1949
+ "brier": 0.1380138971177509,
1950
+ "jsd": 0.06413277962415562,
1951
+ "acc": 0.8973333333333333,
1952
+ "acc_k": 0.932381667918858,
1953
+ "ece": 0.01645974771181742,
1954
+ "sel_acc80": 0.9758333333333333,
1955
+ "conf_wrong": 0.0,
1956
+ "coverage": 0.8686666666666667,
1957
+ "false_abstain": 0.03906836964688205,
1958
+ "null_recall": 0.8579881656804734,
1959
+ "auroc_null": 0.9832799114426577,
1960
+ "auroc_conf": 0.9760735132635959
1961
+ }
1962
+ },
1963
+ "wh_heldout_family": {
1964
+ "raw": {
1965
+ "nll": 0.5463474661337482,
1966
+ "brier": 0.25969181917455386,
1967
+ "jsd": 0.1057542581702437,
1968
+ "acc": 0.8199052132701422,
1969
+ "acc_k": 0.8461538461538461,
1970
+ "ece": 0.078609590917402,
1971
+ "sel_acc80": 0.9023668639053254,
1972
+ "conf_wrong": 0.0,
1973
+ "coverage": 0.8767772511848342,
1974
+ "false_abstain": 0.034482758620689655,
1975
+ "null_recall": 0.8666666666666667,
1976
+ "auroc_null": 0.9859711170055997,
1977
+ "auroc_conf": 0.9732390215148836
1978
+ },
1979
+ "scaled": {
1980
+ "nll": 0.515709105678624,
1981
+ "brier": 0.25745610922373874,
1982
+ "jsd": 0.10622799758517455,
1983
+ "acc": 0.8175355450236966,
1984
+ "acc_k": 0.8461538461538461,
1985
+ "ece": 0.08652745971182513,
1986
+ "sel_acc80": 0.9053254437869822,
1987
+ "conf_wrong": 0.0,
1988
+ "coverage": 0.8744075829383886,
1989
+ "false_abstain": 0.03713527851458886,
1990
+ "null_recall": 0.8666666666666667,
1991
+ "auroc_null": 0.9859711170055997,
1992
+ "auroc_conf": 0.9724727379899794
1993
+ }
1994
+ },
1995
+ "wh_heldout_grammar": {
1996
+ "raw": {
1997
+ "nll": 0.24869155709362914,
1998
+ "brier": 0.14870406423381147,
1999
+ "jsd": 0.06852834314444578,
2000
+ "acc": 0.883578431372549,
2001
+ "acc_k": 0.9226519337016574,
2002
+ "ece": 0.0202937848939031,
2003
+ "sel_acc80": 0.9693721286370597,
2004
+ "conf_wrong": 0.0,
2005
+ "coverage": 0.866421568627451,
2006
+ "false_abstain": 0.04419889502762431,
2007
+ "null_recall": 0.8369565217391305,
2008
+ "auroc_null": 0.9835755464809033,
2009
+ "auroc_conf": 0.9737719192889743
2010
+ },
2011
+ "scaled": {
2012
+ "nll": 0.250087791983527,
2013
+ "brier": 0.1492171979474074,
2014
+ "jsd": 0.06994133383184256,
2015
+ "acc": 0.883578431372549,
2016
+ "acc_k": 0.9226519337016574,
2017
+ "ece": 0.022323701019380575,
2018
+ "sel_acc80": 0.9663093415007658,
2019
+ "conf_wrong": 0.0,
2020
+ "coverage": 0.8639705882352942,
2021
+ "false_abstain": 0.04558011049723757,
2022
+ "null_recall": 0.8478260869565217,
2023
+ "auroc_null": 0.9835755464809033,
2024
+ "auroc_conf": 0.9725708623588758
2025
+ }
2026
+ },
2027
+ "wh_heldout_style": {
2028
+ "raw": {
2029
+ "nll": 0.27159999686744934,
2030
+ "brier": 0.15087037945392912,
2031
+ "jsd": 0.0702443405562412,
2032
+ "acc": 0.8883374689826302,
2033
+ "acc_k": 0.8883374689826302,
2034
+ "ece": 0.02315892800503868,
2035
+ "sel_acc80": 0.9814241486068112,
2036
+ "coverage": 1.0,
2037
+ "false_abstain": 0.0,
2038
+ "null_recall": "n/a"
2039
+ },
2040
+ "scaled": {
2041
+ "nll": 0.26962945683446005,
2042
+ "brier": 0.1503456861985424,
2043
+ "jsd": 0.07193390448881899,
2044
+ "acc": 0.8883374689826302,
2045
+ "acc_k": 0.8883374689826302,
2046
+ "ece": 0.0237619669058778,
2047
+ "sel_acc80": 0.9814241486068112,
2048
+ "coverage": 1.0,
2049
+ "false_abstain": 0.0,
2050
+ "null_recall": "n/a"
2051
+ }
2052
+ },
2053
+ "wh_level7": {
2054
+ "raw": {
2055
+ "nll": 1.0941978558128005,
2056
+ "brier": 0.6996876956981805,
2057
+ "jsd": 0.2755464850522858,
2058
+ "acc": 0.49318181818181817,
2059
+ "acc_k": 0.49318181818181817,
2060
+ "ece": 0.21915304772555827,
2061
+ "sel_acc80": 0.5014204545454546,
2062
+ "coverage": 1.0,
2063
+ "false_abstain": 0.0,
2064
+ "null_recall": "n/a"
2065
+ },
2066
+ "scaled": {
2067
+ "nll": 1.0441702379342557,
2068
+ "brier": 0.6830152541002035,
2069
+ "jsd": 0.27213881565412834,
2070
+ "acc": 0.49318181818181817,
2071
+ "acc_k": 0.49318181818181817,
2072
+ "ece": 0.2006248111074621,
2073
+ "sel_acc80": 0.5,
2074
+ "coverage": 1.0,
2075
+ "false_abstain": 0.0,
2076
+ "null_recall": "n/a"
2077
+ }
2078
+ },
2079
+ "wh_rubric_flip": {
2080
+ "raw": {
2081
+ "nll": 0.4852592661059105,
2082
+ "brier": 0.2752466288148543,
2083
+ "jsd": 0.11332718659046732,
2084
+ "acc": 0.8026666666666666,
2085
+ "acc_k": 0.8235719450469993,
2086
+ "ece": 0.06838929454485576,
2087
+ "sel_acc80": 0.8791666666666667,
2088
+ "conf_wrong": 0.0,
2089
+ "coverage": 0.906,
2090
+ "false_abstain": 0.02964569775849602,
2091
+ "null_recall": 0.8547008547008547,
2092
+ "auroc_null": 0.9890427721230324,
2093
+ "auroc_conf": 0.9718931345829395
2094
+ },
2095
+ "scaled": {
2096
+ "nll": 0.4660126084095091,
2097
+ "brier": 0.2717966241329151,
2098
+ "jsd": 0.11361927024392757,
2099
+ "acc": 0.8013333333333333,
2100
+ "acc_k": 0.8235719450469993,
2101
+ "ece": 0.06078574814399086,
2102
+ "sel_acc80": 0.8783333333333333,
2103
+ "conf_wrong": 0.0,
2104
+ "coverage": 0.9046666666666666,
2105
+ "false_abstain": 0.03109182935647144,
2106
+ "null_recall": 0.8547008547008547,
2107
+ "auroc_null": 0.9890427721230324,
2108
+ "auroc_conf": 0.970243061349352
2109
+ }
2110
+ },
2111
+ "wh_rubric_shuffled": {
2112
+ "raw": {
2113
+ "nll": 2.9113006100472916,
2114
+ "brier": 0.9620256743702243,
2115
+ "jsd": 0.3528958761487168,
2116
+ "acc": 0.398,
2117
+ "acc_k": 0.4255469301340861,
2118
+ "ece": 0.4284619954824448,
2119
+ "sel_acc80": 0.3975,
2120
+ "conf_wrong": 0.12048192771084337,
2121
+ "coverage": 0.9359999999999999,
2122
+ "false_abstain": 0.04163726182074806,
2123
+ "null_recall": 0.4457831325301205,
2124
+ "auroc_null": 0.9583244764520326,
2125
+ "auroc_conf": 0.7902747192014352
2126
+ },
2127
+ "scaled": {
2128
+ "nll": 2.6478036458048013,
2129
+ "brier": 0.9498309075359751,
2130
+ "jsd": 0.3500576999611389,
2131
+ "acc": 0.39866666666666667,
2132
+ "acc_k": 0.4255469301340861,
2133
+ "ece": 0.41628739579518637,
2134
+ "sel_acc80": 0.39916666666666667,
2135
+ "conf_wrong": 0.12048192771084337,
2136
+ "coverage": 0.9353333333333333,
2137
+ "false_abstain": 0.04163726182074806,
2138
+ "null_recall": 0.4578313253012048,
2139
+ "auroc_null": 0.9583244764520324,
2140
+ "auroc_conf": 0.7830730118781406
2141
+ }
2142
+ },
2143
+ "data_u_val": {
2144
+ "raw": {
2145
+ "nll": 0.49041288665397725,
2146
+ "brier": 0.04423868525139621,
2147
+ "jsd": 0.01857737823312269,
2148
+ "acc": 0.8953333333333333,
2149
+ "acc_k": 0.8953333333333333,
2150
+ "ece": 0.12396540609995522,
2151
+ "sel_acc80": 0.9591666666666666,
2152
+ "coverage": 1.0,
2153
+ "false_abstain": 0.0,
2154
+ "null_recall": "n/a"
2155
+ },
2156
+ "scaled": {
2157
+ "nll": 0.49306508783951614,
2158
+ "brier": 0.04551667209993445,
2159
+ "jsd": 0.019483829755564286,
2160
+ "acc": 0.8953333333333333,
2161
+ "acc_k": 0.8953333333333333,
2162
+ "ece": 0.1426694559454918,
2163
+ "sel_acc80": 0.9591666666666666,
2164
+ "coverage": 1.0,
2165
+ "false_abstain": 0.0,
2166
+ "null_recall": "n/a"
2167
+ }
2168
+ },
2169
+ "u_chaosnli": {
2170
+ "raw": {
2171
+ "nll": 0.9992255345818895,
2172
+ "brier": 0.15592243687493582,
2173
+ "jsd": 0.06276678856554775,
2174
+ "acc": 0.548,
2175
+ "acc_k": 0.548,
2176
+ "ece": 0.05644560275475185,
2177
+ "sel_acc80": 0.5625,
2178
+ "coverage": 1.0,
2179
+ "false_abstain": 0.0,
2180
+ "null_recall": "n/a"
2181
+ },
2182
+ "scaled": {
2183
+ "nll": 0.989961406583137,
2184
+ "brier": 0.150204655222619,
2185
+ "jsd": 0.062144876814656025,
2186
+ "acc": 0.548,
2187
+ "acc_k": 0.548,
2188
+ "ece": 0.06379163038730622,
2189
+ "sel_acc80": 0.5641666666666667,
2190
+ "coverage": 1.0,
2191
+ "false_abstain": 0.0,
2192
+ "null_recall": "n/a"
2193
+ }
2194
+ },
2195
+ "u_real_heldout": {
2196
+ "raw": {
2197
+ "nll": 0.6606320054345184,
2198
+ "brier": 0.13458670329345712,
2199
+ "jsd": 0.05980471728660867,
2200
+ "acc": 0.6494661921708185,
2201
+ "acc_k": 0.6494661921708185,
2202
+ "ece": 0.062161013006739735,
2203
+ "sel_acc80": 0.6911111111111111,
2204
+ "coverage": 1.0,
2205
+ "false_abstain": 0.0,
2206
+ "null_recall": "n/a"
2207
+ },
2208
+ "scaled": {
2209
+ "nll": 0.6586741027584431,
2210
+ "brier": 0.13248622800527662,
2211
+ "jsd": 0.060664774416809425,
2212
+ "acc": 0.6494661921708185,
2213
+ "acc_k": 0.6494661921708185,
2214
+ "ece": 0.0787051947409572,
2215
+ "sel_acc80": 0.6933333333333334,
2216
+ "coverage": 1.0,
2217
+ "false_abstain": 0.0,
2218
+ "null_recall": "n/a"
2219
+ }
2220
+ },
2221
+ "u_synthetic_heldout": {
2222
+ "raw": {
2223
+ "nll": 0.6666258308359944,
2224
+ "brier": 0.031941487551321104,
2225
+ "jsd": 0.014305540425327239,
2226
+ "acc": 0.9157894736842105,
2227
+ "acc_k": 0.9157894736842105,
2228
+ "ece": 0.16438576234014407,
2229
+ "sel_acc80": 0.968421052631579,
2230
+ "coverage": 1.0,
2231
+ "false_abstain": 0.0,
2232
+ "null_recall": "n/a"
2233
+ },
2234
+ "scaled": {
2235
+ "nll": 0.6595814980676692,
2236
+ "brier": 0.029197536715013496,
2237
+ "jsd": 0.013290023567297079,
2238
+ "acc": 0.9157894736842105,
2239
+ "acc_k": 0.9157894736842105,
2240
+ "ece": 0.18903793711411326,
2241
+ "sel_acc80": 0.968421052631579,
2242
+ "coverage": 1.0,
2243
+ "false_abstain": 0.0,
2244
+ "null_recall": "n/a"
2245
+ }
2246
+ }
2247
+ }
2248
+ }