violetxi/wm-rl-notes-v4

wm-rl notes v4 — two note banks from a recursive self-experience loop

Continuation update (2026-09-14): rounds 6–8 appended. The recursive bank now contains 162,729 notes / 71,907,847 training tokens. The original 108,081-row round-5 bank remains an exact prefix. Round-4/5 task lists were available for duplicate rejection, but their trajectory manifests were not published; the continuation therefore seeds round 6 from the published recall sessions and carries complete histories from round 6 onward.

continuation round	emitted tasks	notes added	tokens added	pair-coverage gain
r6	7,461	12,360	5,599,950	n/a pt
r7	12,500	20,823	9,407,473	n/a pt
r8	13,000	21,465	9,819,676	n/a pt

Data from a study of scaling internalization from a model's own experience. A Qwen3.5-9B agent works inside a synthetic law firm's document-management system (the "C&H DMS": 9,288 markdown documents across 266 matters, commit 8a578c22) and writes notes about what it finds. The notes are then the only text a LoRA adapter is trained on, and the adapter is measured on a frozen 250-task open-book agentic benchmark and on ~7,900 closed-book probes.

Updated 2026-09-13: rounds 4 and 5 appended. The recursive bank grew from 74,976 notes / 32,276,156 tokens to 108,081 notes / 47,080,748 tokens. The bank is append-only and in generation order, so the first 74,976 notes are byte-identical to the previous snapshot (sha256 c3dac284…, verified) and every published n-1m … n-30m checkpoint remains a valid prefix of this file.

Rounds 4 and 5 were run at a fixed 10,000 tasks each — deliberately, so the per-round series is controlled (rounds 1–3 varied: 4,622 / 11,591 / 16,000). Yield came out near-identical across the two: 16,519 and 16,586 notes, 7.38M and 7.43M tokens, 1.65 and 1.66 notes per task.

round	tasks	notes added	tokens added	pair-coverage gain	near-dup rate
seed	—	23,174	9.24M	—	—
r1	4,622	6,050	2.73M	1.97 pt	0.0007
r2	11,591	19,063	8.46M	3.69 pt	0.0006
r3	16,000	26,689	11.85M	2.90 pt	0.0005
r4	10,000	16,519	7.38M	1.17 pt	0.0001
r5	10,000	16,586	7.43M	0.98 pt	0.0004

Note what the controlled rounds expose: token yield stayed flat while coverage gain halved. The protocol requires the agent to write ≥1 grounded note per rollout, so note count has a floor independent of whether there is new ground to cover — which means note count is a poor saturation signal and coverage is the one to read. Pair coverage reached 61.4%, document coverage 96.1%.

The two banks

file	notes	training tokens	how it was made
banks/notes_recursive.jsonl	162,729	71,907,847	Recursive. Round 0 (the seed) + 8 recursive rounds. Each round proposes tasks from prior sessions, acts with a live note bank, and writes connective notes over the new sessions.
banks/notes_inherited_only.jsonl	112,814	44,206,421	Not recursive. The same seed, then more of the same: connective notes written over bundles of the 49,017 inherited sessions (trajectories from earlier waves). No self-proposed tasks, no new sessions, no agent-written notes — every note is about a session that existed before this study.

The two banks share their first 23,174 notes (9,235,631 tokens, the seed — also published standalone as seed_notes_accepted.jsonl) and nothing else. Both are in generation order; the training conditions are prefixes.

Each bank is published twice, same rows in the same order:

• banks/<bank>.jsonl — the raw file, byte-identical to what the adapters were trained from (sha256 below).
• banks/<bank>.parquet — the browsable copy the dataset viewer shows (the "Viewer" tab above). Columns reordered for reading — note_type, title, text, written_by, generation_round, generation_batch, evidence, … with training_text last — plus one added column, written_by = agent (written mid-task) or connective (post-round pass); prompts_version is null for agent-written notes. Nothing else differs.

Composition

notes_recursive.jsonl by generation_batch:

batch	round	notes	of which agent-written (during rollouts)	connective (post-round pass)
seed	0	23,174	0	23,174
r1	1	6,050	3,861	2,189
r2	2	19,063	10,949	8,114
r3	3	26,689	15,275	11,414

notes_inherited_only.jsonl: seed 23,174 + r0extra 89,640 (73,388 bundles proposed, 235,097 notes proposed, 38.1% accepted).

Note types in both banks: chronology, matter_profile, pitfall, document_map, practice_recipe.

How the training conditions slice them

condition (HF adapter violetxi/qwen35-9b-wmrl-v4-<name>)	source	notes	tokens
n-1m	first 2,546 notes of notes_recursive (all seed)	2,546	1.0M
n-3m	first 7,597 (all seed)	7,597	3.0M
n-10m	first 24,754 (seed + 1,580 of r1)	24,754	10.0M
n-30m	first 68,833 (seed + r1 + r2 + 20,546 of r3)	68,833	30.0M
n-1m-hi	greedy pair-coverage pick over the whole recursive bank (not a prefix)	2,383	1.0M
r0-30m	first 76,350 notes of notes_inherited_only (seed + 53,176 of r0extra)	76,350	30.0M
null-30m	the n-30m slice with every name, number, date and path replaced by a placeholder (derived; not uploaded)	68,587	30.0M
raw-30m	the source documents themselves, not notes	2,319 docs	30.0M
b5-30m	rendered fact items from an earlier wave, not notes	966,315	30.0M

n-1m ⊂ n-3m ⊂ n-10m ⊂ n-30m (nesting verified at build time).

Schema (both banks)

note_id, repository, commit, note_type, title, text, training_text, evidence (list of {path, start_line, end_line} the note cites), supporting_trajectory_ids, related_question_ids, generation_round, generation_batch, prompts_version.

• Only training_text is trained on (title + repository/commit header + body). Documents are packed with one EOS separator each into 16,384-token rows.
• Agent-written vs connective: notes written by the agent mid-task have no prompts_version key and their supporting_trajectory_ids are the round's own task ids (n1_…, n2_…, n3_…). Connective notes carry prompts_version (notes-prompts-5 for the seed, -8 for round 1, -9 after) and cite inherited-session ids (g3_…, d2_…, r4_…, …).

Gates

Every note passed 13 static checks before entering a bank (gates versions notes-gates-6 for the seed, -9 for round 1, -10 after): schema, note type, length (≤1,024 tokens, body ≥30), support, cited path was actually observed in the session, cited line range exists and was read, observed range, entity containment (every named entity/number/date in the note appears in the documents the session read), coverage-claim, firm-scope, near-duplicate (Jaccard ≥0.8), anti-scaffold, voice, and a fast lexical gate. Acceptance: seed 29.7%, round 1 47.1%, rounds 2–3 55.9% / 55.7%, r0extra 38.1%.

What the banks did (summary of the study's readout)

• Closed-book knowledge rises ≈ +9 points per decade of note tokens along n-1m → n-30m (n = 4 points, r = +0.94).
• On the open-book agentic benchmark, n-30m keeps protocol integrity (0.875 of rows answered, no-think) while r0-30m (0.233) and null-30m (0.012) collapse at the same step count — the model starts emitting note-shaped text instead of tool calls. What is written, not how many steps, decides whether the adapter survives.
• Notes beat the source documents at equal tokens (raw-30m) on both evals.

Caveats

• Notes are not scrubbed against the held-out benchmark. They come from the model's own reads, so some overlap is inherent: 638 of 1,369 held-out fact pairs are contained somewhere in the n-30m slice (633 in r0-30m). Conditions trained on notes are tagged DIRECTIONAL rather than EXACT in the scorecard.
• The seed bank's grounding audit did not clear its bar (80.0% vs an 85% target, absence-type over-claims 17.3%). It is published as measured; the study is about what a model internalises from its own experience, and filtering to audit-clean would measure something else.
• Round-1 tasks were regenerated twice before the round ran; both failed task sets are kept (round1_tasks_attempt*.jsonl, audits/) because the failures are informative.

Files kept from the earlier snapshot

seed_notes_accepted.jsonl (the shared 23,174-note seed), seed_notes_rejected.jsonl (54,979 seed rejections with verdict strings), round1_tasks_attempt{1,2}_*.jsonl, round1_taskgen_report_attempt{1,2}.md, audits/.

Integrity

file	sha256
banks/notes_recursive.jsonl	3b94c2cdd2de6bc729e4ad9964ba72a4c23faf9db9c8656730c65d2cb0588bc5
banks/notes_inherited_only.jsonl	218c3ed9cf40b5ddcb64d228caa7a6f9c3ca8ad15e681bbde841f58d1713a2f5