Datasets:
Tasks:
Text Classification
Formats:
parquet
Sub-tasks:
semantic-similarity-classification
Languages:
code
Size:
1M - 10M
License:
sync EmbedEd artifacts (phase01-v8)
Browse files- artifacts/phase4_decisions.json +1 -1
- report/measurements.md +18 -1
- sync_manifest.json +4 -2
artifacts/phase4_decisions.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"recorded_utc": "2026-10-21
|
| 3 |
"version": "phase01-v8",
|
| 4 |
"decisions": {
|
| 5 |
"D4-1": {
|
|
|
|
| 1 |
{
|
| 2 |
+
"recorded_utc": "2026-10-21 05:11 UTC",
|
| 3 |
"version": "phase01-v8",
|
| 4 |
"decisions": {
|
| 5 |
"D4-1": {
|
report/measurements.md
CHANGED
|
@@ -777,7 +777,7 @@ Phase 2 checkpoints (C0 = base model) scored on the same s′ partition without
|
|
| 777 |
|
| 778 |
## Phase 4 — the one table
|
| 779 |
|
| 780 |
-
Assembled from the run dirs by the Phase-4 notebook — version `phase01-v8`, recorded 2026-10-21
|
| 781 |
|
| 782 |
### Panel A — benchmark (F1 mean ± sd over seeds 13-15; pinned = validation thresholds saturated at 1.0)
|
| 783 |
|
|
@@ -810,3 +810,20 @@ Assembled from the run dirs by the Phase-4 notebook — version `phase01-v8`, re
|
|
| 810 |
### Panel C — mechanism bookkeeping
|
| 811 |
|
| 812 |
Trigger = an FN that is simultaneously a functional clone and a J ≥ 0.40 near-duplicate; rate = dose; margin 0.10 = amplifier (§9.4 FINAL). The 40% cut removes ~40% of C2 contamination and 0 audited clean pairs. audit_c4 PASSED (30% ≤ 42%); audit_c5/c6 outcomes: see section 3 — no C5/C6 headline before its branch is recorded.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 777 |
|
| 778 |
## Phase 4 — the one table
|
| 779 |
|
| 780 |
+
Assembled from the run dirs by the Phase-4 notebook — version `phase01-v8`, recorded 2026-10-21 05:11 UTC. Sources: `runs/<run>/eval_metrics.json`, the audit sections below, and the s′ table. **The two panels are different rulers** (benchmark: natural 13.7% prevalence, all-negative ceiling 0.2406; s′: balanced slices) — absolute F1 is NOT comparable across panels; within a panel, orderings are the finding.
|
| 781 |
|
| 782 |
### Panel A — benchmark (F1 mean ± sd over seeds 13-15; pinned = validation thresholds saturated at 1.0)
|
| 783 |
|
|
|
|
| 810 |
### Panel C — mechanism bookkeeping
|
| 811 |
|
| 812 |
Trigger = an FN that is simultaneously a functional clone and a J ≥ 0.40 near-duplicate; rate = dose; margin 0.10 = amplifier (§9.4 FINAL). The 40% cut removes ~40% of C2 contamination and 0 audited clean pairs. audit_c4 PASSED (30% ≤ 42%); audit_c5/c6 outcomes: see section 3 — no C5/C6 headline before its branch is recorded.
|
| 813 |
+
|
| 814 |
+
## False-negative audit — C5 (blind; SCOPE P2-9)
|
| 815 |
+
|
| 816 |
+
`python -m scripts.audit_sample score` — version `phase01-v8`, conditions C5. Share of mined negatives judged to be clones (functional equivalence). Headline counts `unsure` as not_clone; the upper bound counts it as clone.
|
| 817 |
+
|
| 818 |
+
| condition | n | clone | not_clone | unsure | FN rate [95% CI] | upper bound incl. unsure |
|
| 819 |
+
|---|---|---|---|---|---|---|
|
| 820 |
+
| C5 | 50 | 15 | 34 | 1 | **30%** [19%, 44%] | 32% [21%, 46%] |
|
| 821 |
+
|
| 822 |
+
## False-negative audit — C6 (blind; SCOPE P2-9)
|
| 823 |
+
|
| 824 |
+
`python -m scripts.audit_sample score` — version `phase01-v8`, conditions C6. Share of mined negatives judged to be clones (functional equivalence). Headline counts `unsure` as not_clone; the upper bound counts it as clone.
|
| 825 |
+
|
| 826 |
+
| condition | n | clone | not_clone | unsure | FN rate [95% CI] | upper bound incl. unsure |
|
| 827 |
+
|---|---|---|---|---|---|---|
|
| 828 |
+
| C6 | 50 | 15 | 34 | 1 | **30%** [19%, 44%] | 32% [21%, 46%] |
|
| 829 |
+
|
sync_manifest.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"version": "phase01-v8",
|
| 3 |
-
"synced_at_utc": "2026-10-21T05:
|
| 4 |
"staged_files": [
|
| 5 |
"artifacts/audit/audit_key.csv",
|
| 6 |
"artifacts/audit/audit_labels.csv",
|
|
@@ -27,6 +27,7 @@
|
|
| 27 |
"artifacts/pairs_train.tsv",
|
| 28 |
"artifacts/pairs_valid.tsv",
|
| 29 |
"artifacts/phase3_decisions.json",
|
|
|
|
| 30 |
"artifacts/runs/C0_13/eval_metrics.json",
|
| 31 |
"artifacts/runs/C1_13/eval_metrics.json",
|
| 32 |
"artifacts/runs/C1_13/metrics.json",
|
|
@@ -298,7 +299,8 @@
|
|
| 298 |
"artifacts/triples_C3.jsonl",
|
| 299 |
"artifacts/triples_C4.jsonl",
|
| 300 |
"artifacts/triples_C5.jsonl",
|
| 301 |
-
"artifacts/triples_C6.jsonl"
|
|
|
|
| 302 |
],
|
| 303 |
"excluded_patterns": [
|
| 304 |
"sprime/*",
|
|
|
|
| 1 |
{
|
| 2 |
"version": "phase01-v8",
|
| 3 |
+
"synced_at_utc": "2026-10-21T05:11:16Z",
|
| 4 |
"staged_files": [
|
| 5 |
"artifacts/audit/audit_key.csv",
|
| 6 |
"artifacts/audit/audit_labels.csv",
|
|
|
|
| 27 |
"artifacts/pairs_train.tsv",
|
| 28 |
"artifacts/pairs_valid.tsv",
|
| 29 |
"artifacts/phase3_decisions.json",
|
| 30 |
+
"artifacts/phase4_decisions.json",
|
| 31 |
"artifacts/runs/C0_13/eval_metrics.json",
|
| 32 |
"artifacts/runs/C1_13/eval_metrics.json",
|
| 33 |
"artifacts/runs/C1_13/metrics.json",
|
|
|
|
| 299 |
"artifacts/triples_C3.jsonl",
|
| 300 |
"artifacts/triples_C4.jsonl",
|
| 301 |
"artifacts/triples_C5.jsonl",
|
| 302 |
+
"artifacts/triples_C6.jsonl",
|
| 303 |
+
"report/measurements.md"
|
| 304 |
],
|
| 305 |
"excluded_patterns": [
|
| 306 |
"sprime/*",
|