README: 2026-11-07 Pod B backup (e33, e35, e41-e45, e16/pykt remainders)

README.md

@@ -32,10 +32,18 @@ All E16-driver (S1) arms also mix 5 % replay rows from SargeDev/jev-distill-corp
 commercial Jev API, called through OpenRouter:
 - `typesafe/jev-1.13-20260917`: `e16/S1-ABCLODO16-*` (A+B+C LODO), `e16/S1-G-d*` (E18 grid, except `-a100`).
 - `typesafe/jev-1.14-20261014`: `e16/S1-E19-*` α 0.5 arms (`a1-n16k-1model`, `a1-n16k-nopersona`, `a2-n64k`,
-  `b-n64k`, `b-n128k`), `e16/S1-SS-hlr-n16k-syn`, `e16/S1-SS-mix-n16k-syn`.
+  `b-n64k`, `b-n128k`, `b-n256k`, `b-n4k-rt`, `b-n16k-rt`), `e16/S1-SS-hlr-n16k-syn`, `e16/S1-SS-mix-n16k-syn`,
+  the α 0.5 arms of `e33/`, `e35/` and `e42/` (LLM-student data, labels from `jev-latest` = 1.14).
+- `e41/S1-E41-FRESH-A050-ts0-syn`: `typesafe/jev-1.13-20260917` (E18 G-d512-n16k α 0.5 rows).
 - Both versions: `e16/S1-MIX-E18E19-n16k-syn`, `e16/S1-MIX3-n16k-syn`.
 - Indirect, through the Jev-distilled jev-distill corpus: `e15/F3-gen`.
-- Without Jev labels: E19 `-a100` / `-soft-a100` arms, `e15/F2-flat`, and all real-KT arms listed below.
+- Without Jev labels: E19 `-a100` / `-soft-a100` arms, `e15/F2-flat`, all `-a100` arms of `e33/` / `e35/`,
+  `e41/` except FRESH-A050, `e45/` (E18 α 1 rows, parametric simulator only), `e44/`, and all real-KT arms listed below.
+
+**Backbone note (`e41/`).** E41 "fresh" arms start from the plain Qwen/Qwen3.5-9B backbone (the JEV-9B backbone
+safetensors, verified bit-identical to Qwen3.5-9B's text tensors) with a NEW LoRA + head; the released JEV-9B adapter and
+head were not loaded. `e41/S1-E41-FRESH-B0-syn` is the untrained step-0 export. `S1-E41-FRESH-A100NR-ts0-syn` uses no
+replay rows (no Jev-derived parameter or row).
 
 **Research use; respect the source licences.**
 
@@ -45,9 +53,9 @@ commercial Jev API, called through OpenRouter:
 | `meta_only/podA/e16/S1-O1-near` | Duolingo HLR | **CC BY-NC 4.0** (config + log only) |
 | `meta_only/podA/e16/S1-Dmix-oth`, `meta_only/podC/e16/S1-T-mix`, `S1-O3-syn`, `S1-O3x16-syn` | ChaosNLI (via T-cls / D-mix) | **CC BY-NC 4.0** (configs + logs only) |
 | `e16/S1-E19-a2-*` | XES3G5M question text (KCQRL English translation) | unclear redistribution terms |
-| `e15/F1-LODO-*`, `e15/F1-cs/*`, `e15/P`, `e16/S1-ABCLODO16-*`, `e16/S1-F1pool64-rp`, `e16/S2-*`, `pykt/*`, `classical/*` | public KT benchmarks (ASSISTments 2009/2017, Eedi NIPS 2020 tasks 3–4, XES3G5M, DBE-KT22, KDD Cup 2010 Algebra 2005, Junyi) | each dataset's own terms |
+| `e15/F1-LODO-*`, `e15/F1-cs/*`, `e15/P`, `e16/S1-ABCLODO16-*`, `e16/S1-F1pool64-rp`, `e16/S2-*`, `pykt/*`, `classical/*`, `e43/akt*` | public KT benchmarks (ASSISTments 2009/2017, Eedi NIPS 2020 tasks 3–4, XES3G5M, DBE-KT22, KDD Cup 2010 Algebra 2005, Junyi) | each dataset's own terms |
 | `meta_only/podC/e16/S1-T-seq`, `S1-P-seq`, `S1-D-seq*` | KuaiRec (CC BY-SA 4.0), Lichess / football results (CC0) | ShareAlike + attribution (configs + logs only) |
-| `e16/S1-E19-*` | simulated-student answers from Llama 3.1/3.2, Gemma 3, Qwen 2.5 | Llama Community Licence and Gemma Terms of Use conditions on models trained on their outputs |
+| `e16/S1-E19-*`, `e42/*`, `e33/*`, `e35/*` | simulated-student answers from Llama 3.1/3.2, Gemma 3, Qwen 2.5 (E19 / E28 students, OpenRouter) and, for `e33/` / `e35/`, up to 30 further open ≤ 9B instruct models served locally (permissive licences; Llama 3 / Gemma via ungated byte-identical copies) | Llama Community Licence and Gemma Terms of Use conditions on models trained on their outputs |
 
 ## Layout
 
@@ -56,6 +64,9 @@ e15/<arm>/train_log.jsonl, e15/<arm>/ckpt-STEP/      # E15 arms (jevkt.jev9b_ft)
 e15/F1-cs/<ds>-s<seed>N<N>/ckpt-*/                   # one block per cold-start cell
 e16/<arm>/train_log.jsonl, e16/<arm>/ckpt-STEP/      # E16 driver (jevkt.e16_rl): S1 (incl. E17-E20, LODO), S2 RL arms
 meta_only/pod{A,C}/e1x/<arm>/                        # configs + train logs ONLY (weights not uploaded yet)
+e33/, e35/, e41/, e42/, e45/<arm>/ckpt-16000/       # E33 / E35 student-model sets, E41 fresh init, E42 1-vs-many students, E45 kNN-line formats
+e43/akt/<ds>/csN<N>-s<seed>/akt_tables.pt           # LLM-KT re-implementation: the in-cell AKT embedding tables only (its LoRA is never saved)
+e44/pykt/pretrain/<model>/                          # pyKT pretrained on E18 synthetic learners (simulator outcomes, α 1)
 pykt/<ds>/pykt-<model>-s<seed>/, pykt/<ds>/coldstart/ # pyKT DLKT baselines: qid_model.ckpt + train_info.json
 classical/<model>__<ds>__s<seed>.pt                  # leap-kt-toolkit baselines
 ```
@@ -80,4 +91,7 @@ Usage with JEV-9B served by vLLM (code in the Jev-KT repo):
 ## Not yet included
 
 - Weights for 68 checkpoints that are still only on stopped training pods. Their configs and logs are in `meta_only/`.
-- Arms that were still training at upload time: E19-b-n256k, E19-a2-n16k-a100, S2-grpo-log, E27 (27B).
+- Arms that were still training at upload time: E27 (27B) partials are in `e27/`; E44 K1 / A arms (training later).
+- 2026-11-07: every checkpoint that existed only on Pod B was added (`e15`/`e16` remainders incl. E19-b-n256k,
+  E19-a2-n16k-a100, S2-grpo-log, `e33`, `e35`, `e41`–`e45`, pod-trained `pykt/` models). Resume / optimizer states
+  are never uploaded.