textmachine

heaven/textmachine

Fork 0

Commit graph

1db62d41a3 Record the D39.17 errata: the runWave parent-cancellation bug my rubezh-2 missed, caught by an independent second review and fixed main Claude (backend session) 2026-07-19 20:41:10 +03:00
5fc4860d64 Fix runWave swallowing parent cancellation (D39.17 errata): return parent.Err() when no worker errored, so a Ctrl-C resume surfaces context.Canceled instead of a nil-deref panic or false exit-0 partial success; regression-pinned Claude (backend session) 2026-07-19 20:40:04 +03:00
5be69939fc Archive the R1 driver-switch session report Claude (backend session) 2026-07-19 20:11:26 +03:00
89d5bf9d6a Land R1 wave driver-switch (D39.17): parallel draft/edit waves with bank-mining stop, editor never re-renders draft, Sh-1/Sh-2, per-unit read-models; rubezh-2 clean bill, three MINOR mining-wiring fixes to pre-rerun; pack-11 complete Claude (backend session) 2026-07-19 20:11:05 +03:00
a263365be3 Record the exp16 signature-map approval and the Bai Ningbing finding: time-varying/perspective-split gender is the limit of deterministic DC-3, routed to the F2 editor with narrative annotation Claude (backend session) 2026-07-19 18:29:39 +03:00
fa620e2015 Tighten the R1 prompt: mark FL-1/2/3 as landed context and Sh-1/Sh-2 as the work to do this pass Claude (backend session) 2026-07-19 18:05:52 +03:00
e0b8a52783 Clean the backend code zone: move session reports to docs/archive/reports, stop persisting backend/*_REPORT.md, and ratify the Sh-1/Sh-2 snapshot-guard findings into the R1 fix-list Claude (backend session) 2026-07-19 18:03:57 +03:00
d358143342 Actualize all documentation to post-D39.16: refresh current-state headers, status lines, D-log ranges, langpack realization notes, and zone README cross-references Claude (backend session) 2026-07-19 17:55:51 +03:00
0b529725b5 Add the worked-out banner to the archived arch-cleanup prompt copy Claude (backend session) 2026-07-19 17:47:52 +03:00
91c400c8ae Land the language-data isolation arch pass (D39.16): move miner zh/Palladius data to internal/lang and configs/langpacks byte-exactly, land FL-1/2/3, ratify the three press-tested divergences, archive the worked prompt Claude (backend session) 2026-07-19 17:47:14 +03:00
3798643740 Issue the intermediate architecture-cleanup session prompt per D39.15: isolate language data out of the pipeline package before R1, ratify the direction and re-sequence the queue Claude (backend session) 2026-07-19 15:51:56 +03:00
13c4375e2a Land backend pack-11 continuation: Go miner with exact parity, banknote 12-point integration, DC checkers with export --pairs, editor arm configs, and additive wave snapshot; clear the rubezh-2 debt with a three-item fix list Claude (backend session) 2026-07-19 15:24:48 +03:00
88b5e7b47d Actualize the backend pack-11 session prompt for the continuation run over residuals R1-R5 Claude (backend session) 2026-07-19 08:59:53 +03:00
85f5b9e633 Land backend pack-11 block A: output-token chunker with edit units, src-to-dst editor block, eager clients, per-model rate guard, memory version split, banknote parser core Claude (backend session) 2026-07-19 08:53:54 +03:00
1f285250ab Add the worked-out banner to the archived design-synthesis prompt copy Claude (backend session) 2026-07-19 04:03:52 +03:00
09ca24b50a Land the ratified implementation plan as D39.12 with design11 verifications, record the mistral rate-limit quirk, and issue the backend single-pack session prompt Claude (backend session) 2026-07-19 04:03:39 +03:00
766a60e304 Issue the design-synthesis session prompt per D39.11 after a 20-finding adversarial review, and update the progress journal and active-prompt index Claude (backend session) 2026-07-19 01:06:55 +03:00
c86bdcef2b Close orchestrator session six: ratify D39.11 design-first directive, issue handoff five, actualize status docs and archive the spent experiment prompts Claude (backend session) 2026-07-19 00:35:33 +03:00
a7f9adf2a3 Ratify D39.10 landing exp16: the which-what split, banknote as the dst channel, the clean re-audit, closing the arch-reset research program Claude (backend session) 2026-07-19 00:25:13 +03:00
aaa4e12bf8 Commit the stray Q4a polygon addendum note left untracked from the mini-session Claude (backend session) 2026-07-19 00:25:13 +03:00
667eec0dc5 Land exp16 bank-mining: the which-what split with the 0.97 recall code detector, banknote as the dst channel, the clean exp14b re-audit, and the stop-only truncation errata Claude (backend session) 2026-07-19 00:24:24 +03:00
96345f589d Freeze exp16 bank-mining pre-registration: miner-v1 detectors V-A/V-B/V-C, alias/canon/banknote code, GT filter and thresholds tuned on ch1-15, before any paid call Claude (backend session) 2026-07-18 23:32:53 +03:00
b9e62a7f49 Fold the exp15 and Q4a rig lessons into the exp16 prompt: valley-only DeepSeek generation, length-rejection everywhere, hardened judge pattern with the measured noise floor Claude (backend session) 2026-07-18 22:12:53 +03:00
f6c52d11e9 Ratify D39.9 closing the exp15 program and order the zero-cost exp14b re-audit with corrected rules as an exp16 pre-task Claude (backend session) 2026-07-18 22:08:20 +03:00
cfcf0f08e3 Land Q4a: fidelity is not bought at the translate stage, the pro arm is not ratified, the glm editor is the confirmed weak link, with corrected trap rules and the truncation-accepting harness bug documented Claude (backend session) 2026-07-18 22:08:20 +03:00
94aeeb55fd Ratify D39.8: blind read passes no arm, reader confirms the floor-null, quality battle moves to defect classes with the units, numbers, gender, verse, and register ledger Claude (backend session) 2026-07-18 17:25:07 +03:00
034660ce1d Append the blind-read verdict addendum: no arm passes the owner bar, defects distributed across arms confirming the metric null, defect-class ledger and instrument fixes recorded Claude (backend session) 2026-07-18 17:24:30 +03:00
2f5025701e Ratify D39.7: exp15 landed in REV.2 with the boundary headline retracted as a rig artifact and all cohesion effects floor-gated, blind read remains the ratifying endpoint Claude (backend session) 2026-07-18 17:07:09 +03:00
fe34a16896 Land exp15 REV.2: floor-gated corrected results with per-vote judge persistence, the retracted boundary headline, and orchestrator landing fixes to the deviations ledger Claude (backend session) 2026-07-18 17:06:11 +03:00
cabdfb35b4 Issue the Q4a mini-session prompt: deferred strong-translator arm with its own two-dollar budget outside the exp15 freeze, valley-only DeepSeek, fixed judging rig Claude (backend session) 2026-07-18 14:04:29 +03:00
fdbee84378 Issue the exp16 bank-mining polygon prompt executing research/20 section D with the owner's budget, canon, no-genre-packs, and ja-ru replica decisions baked in, held until exp15 closes Claude (backend session) 2026-07-18 02:56:31 +03:00
78a9a2dd9e Record exp15 Q2b instrument-ceiling result plus money and self-review sections, and append the paid Q1/Q3 handoff block for the next focused session Claude (backend session) 2026-07-17 23:09:41 +03:00
b9272a19d0 Freeze exp15 pre-registration with verified slugs/prices, sub-$15 budget, grok+mistral judges, A0 injection method, synthetic S2-prime seam ledger, and gender-signed seed Claude (backend session) 2026-07-17 21:13:58 +03:00
68032f7565 Actualize the polygon README and the D15.2 spec status: tmctl export extraction rule, superseded editor caveat, judge EOL residual, deferred-implementation banner Claude (backend session) 2026-07-17 03:20:53 +03:00
fd4f93947c Actualize backend README to the D39.6 head with every claim execution-verified: new CLI surfaces, v8 store, pair seam, trust-gate, export contract, honest tech-debt ledger Claude (backend session) 2026-07-17 03:16:08 +03:00
5d752e93e0 Doc hygiene per owner: archive the closed 10-13.07 chronicle out of PROGRESS, archive spent prompts, actualize CLAUDE, README and the orchestrator role banner to D39.6 Claude (backend session) 2026-07-17 03:10:18 +03:00
994987d8b7 Land research/20 bank-mining with orchestrator review header: section-A hash reproduced, repo claims spot-verified, injection confound corroborated, hold-gate deviation accepted Claude (backend session) 2026-07-17 03:05:40 +03:00
bb4142f503 Ratify D39.5 landing pack-1.5 with the F1-F9 fix-list after execution verification, closing the polygon export surface question Claude (backend session) 2026-07-17 02:16:48 +03:00
dc30c7d61d Land track-A pack-1.5 with fix-list F1-F9: tmctl export surface with manifest and drift guards, fold-first sanitizer detection, bounded CJK-gloss whitelist, runnable zh-ru example Claude (backend session) 2026-07-17 02:16:48 +03:00
9dae0ed13d Ratify D39.4: adversarial debt repaid across both packs, hold pack-1.5 landing behind fix-list F1-F9 folded into the prompt's reserve slot Claude (backend session) 2026-07-17 01:33:22 +03:00
2cfb7cdb3d Amend the research/20 prompt per researcher-19 review: require detector variants with a trade-off table compared as arms, and alias clustering as an explicit subtask with a code-vs-blind-zone boundary Claude (backend session) 2026-07-16 22:52:42 +03:00
084c4c68c6 Ratify D39.3 bank-mining package dispositioning H15 and V4-p4, issue the research/20 prompt held until the exp15 freeze Claude (backend session) 2026-07-16 22:47:14 +03:00
7e1e6780d1 Add the consolidated prompt-architecture note answering the owner's four recurring prompting questions with statuses and routing, linked from the layer-2 section Claude (backend session) 2026-07-16 22:24:28 +03:00
11589d3935 Issue the small track-A pack-1.5 backend prompt: read-only tmctl export surface, owner-decided CJK gloss whitelist, runnable example, adversarial-debt reserve slot Claude (backend session) 2026-07-16 22:05:32 +03:00
8ba0f720e8 Issue the exp15 segmentation-empirics polygon prompt executing research/19 section D with the owner's budget, gender-field, and gloss decisions baked in Claude (backend session) 2026-07-16 22:01:43 +03:00
6c0010ee35 Ratify D39.2 accepting track-A pack-1 after direct execution verification and an independent golden re-capture reproduction, registering the 529-blocked adversarial pass as debt Claude (backend session) 2026-07-16 21:56:00 +03:00
d3f6b34ad4 Land track-A pack-1: trust-gated glossary suppressor with loud telemetry, export-contract normalization, pair-keyed prompt seam, role-injection registry, target-gated readability, quality report Claude (backend session) 2026-07-16 21:54:00 +03:00
ce3e17faa0 Record the owner's V4 updates: the parallel-draft glossary-mining pipeline sketch as the killer-feature speed idea and the self-closed hosting question Claude (backend session) 2026-07-16 21:52:36 +03:00
3d1dc8d8cb Ratify D39.1: land research/19 as the track-B design of record with the wave architecture and pre-registered empirics, review waived by the owner Claude (backend session) 2026-07-16 21:52:17 +03:00
c5171b47c5 Land research/19 chunking-cohesion report with sources and archive its executed prompt: emission-window-cohesion split, Go chunker spec, wave architecture, pre-registered Q0-Q5 design Claude (backend session) 2026-07-16 21:48:02 +03:00
2d9f70bb84 Correct a narrowing in the target-architecture and track-B prompt: add the generality invariant so the chunker stays general across any book, language, and structure Claude (backend session) 2026-07-13 01:53:07 +03:00
e9b801ff5f Record the owner's V4 idea set in START_PROMT: parallelism/speed as the main V4 idea, glossary construction, and the per-language and prompt-language questions Claude (backend session) 2026-07-13 01:51:02 +03:00
d42d01e75a Issue the track-A backend build-now pack and the track-B chunking-cohesion researcher prompt for the architectural reset Claude (backend session) 2026-07-13 01:45:42 +03:00
225abc0930 Ratify D39: land the broken-telephone sync-audit ledger and the target seven-layer architecture with the phased build-now and research tracks, folding in the polygon sync Claude (backend session) 2026-07-13 01:45:42 +03:00
79477173f8 Archive three executed session prompts with outcome banners: the arch-reset (now D39), the discourse editor, and the reseed transform Claude (backend session) 2026-07-13 01:45:29 +03:00
c9c9df7364 Pause the build round for an architectural reset: issue the orchestrator handoff prompt capturing the owner's five concerns and the polygon-orchestrator-backend sync plan Claude (backend session) 2026-07-13 00:32:09 +03:00
183c8ad24f Ratify D38.5 landing the resnapshot components: the reseed transform and the discourse editor with chengyu and the few_shot toggle, both execution-verified Claude (backend session) 2026-07-12 23:48:49 +03:00
0ac5f7a27c Land the discourse editor: P1a reflow prompt with the D38.4 chengyu atom and a per-stage few_shot toggle (zero-shot for the deepseek-pro swap arm), snapshot-folded and golden-stable Claude (backend session) 2026-07-12 23:46:34 +03:00
eb409f2b77 Commit the reseed transform in eval/exp14b and note the seed-provenance lineage in the eval README, accepting the folder placement since it carries no contract citation Claude (backend session) 2026-07-12 23:34:16 +03:00
e781f7b08f Update reseed handoff in PROGRESS: correct transform path to exp14b folder and record the stale grade-note tail was cleaned before the resnapshot per orchestrator Claude (backend session) 2026-07-12 23:27:56 +03:00
38c5fb461c Ratify D38.4 completeness audit of exp14/14b: add the forgotten chengyu atom to the discourse prompt, record the unbuilt inversion guard raising editor-swap priority, flag the chunking arm and guard toggle Claude (backend session) 2026-07-12 23:10:15 +03:00
7e6f9a5a4f Issue the discourse-prompt (backend) and reseed (polygon) handoff prompts for the single resnapshot, and archive the landed infra-pack prompt Claude (backend session) 2026-07-12 22:49:51 +03:00
ffaf076fd9 Ratify D38.3 accepting the landed readability infra-pack after execution-verifying build, vet, race tests and the three span-defect pins Claude (backend session) 2026-07-12 22:41:39 +03:00
d5beefc8f4 Land the readability infra-pack (D38): cosmetic-class strip-and-export for flagged chunks, a CJK-leak sanitizer class, and an opt-in post-reflow number/omission regression guard Claude (backend session) 2026-07-12 22:39:03 +03:00
b14ba6f20c Apply verification fixes: correct the archived ACCEPTANCE line anchor, the D34.3 re-gate attribution in both diagrams, the Mistral 18/18 provenance, and the actuality-map revision stamp Claude (backend session) 2026-07-12 22:21:02 +03:00
feb1848b9c Ratify D38.2 in the decision log and record the consolidation in PROGRESS: eval regrouped, docs and diagrams actualized to D38.1, links verified, backend untouched Claude (backend session) 2026-07-12 22:07:24 +03:00
fc94d6725a Archive the executed POLYGON_CLEANUP prompt with an outcome banner and record it in the README closed-prompts list Claude (backend session) 2026-07-12 22:04:40 +03:00
44bc4f8332 Repair the live ACCEPTANCE citation in research/17 to its archived path and drop the over-attributed D-numbers on the exp04 and exp12 index rows Claude (backend session) 2026-07-12 22:03:07 +03:00
5d24ba401c Sync architecture diagrams to the D1-D38.1 head: bilingual editor flip, output-sanitizer and quality-diagnostics nodes, channel-B v2, seed v2, and provider role restatements Claude (backend session) 2026-07-12 21:59:56 +03:00
6f12fc0244 Actualize onboarding and status docs to the D38.1 head and bump decision-log range references, extending the actuality map with D38.1 and D38.2 Claude (backend session) 2026-07-12 21:52:21 +03:00
f468f27de6 Group closed experiment scripts into eval/exp12|13|14|14b folders with import shims preserved, keep live seed and oracles in eval/, and update all path citations Claude (backend session) 2026-07-12 21:48:55 +03:00
af469140dc Report polygon cleanup in PROGRESS: fact-check refuted the naive archive because closed scripts are cited by path, so consolidate via index and script map instead Claude (backend session) 2026-07-12 20:57:19 +03:00
9feb1151e8 Add experiments index and extend eval script map over the closed quality-arc families, keeping scripts in place because the D-log and reports cite them by path Claude (backend session) 2026-07-12 20:55:45 +03:00
0be308498c Commit exp14b harness scripts (arms, judge, monitor, score) to close the reproducibility gap now that the meaning-battery experiment is ratified and closed at D38 Claude (backend session) 2026-07-12 20:55:45 +03:00
2d760fdb4c Issue polygon cleanup prompt: fact-check consolidation direction, build experiments index, carefully archive spent eval scripts and commit exp14b, flag books and puml hygiene Claude (backend session) 2026-07-12 20:47:10 +03:00
8b8a9ec18d Land blind h2h and pivot to building: ChatGPT/Claude split on frontier-vs-parity but agree the gap is glossary+sanitizer infra, so issue the cheap infra-pack backend prompt (D38.1) Claude (backend session) 2026-07-12 20:30:41 +03:00
33c79d7a01 Ratify D38 accepting exp14b: refute only-gpt-5.4 (mistral and deepseek-pro fix meaning inversions), find term-drift root is seed draft-status, hold cheap-editor swap for a prose bake-off Claude (backend session) 2026-07-12 20:06:03 +03:00
c5c277615a Add owner-directed guardrail: session prompts must mandate self-review of code, model requests, and results because polygon errs and skews experiments Claude (backend session) 2026-07-12 17:44:48 +03:00
48e3f39eb8 Freeze exp14b pre-registration: meaning-trap battery (5 construction classes, deterministic + cross-family judges) to test whether the within-chunk comprehension fix is gpt-5.4-specific, plus P1a-vs-frontier ranking Claude (backend session) 2026-07-12 15:15:03 +03:00
7a30ab4fb6 Issue exp14b polygon prompt: meaning-trap battery to settle whether gpt-5.4 escalation is needed plus P1a-vs-frontier ranking, per owner direction Claude (backend session) 2026-07-12 05:21:56 +03:00
7ee668fba0 Ratify D37 accepting exp14 quality-empirics with fixes: claim-1 paragraphs is a cheap prompt lever, claim-2 gpt-5.4 escalation held under-powered, archive spent exp14 prompts Claude (backend session) 2026-07-12 05:15:55 +03:00
b59ca66b83 Land external blind-eval corroboration: two zero sessions confirm F-disc is best on both prose and fidelity, flag floating grade glossary, and catch an A-2pass fidelity catastrophe that redirects near-term to single-pass discourse Claude (backend session) 2026-07-12 04:29:32 +03:00
73a45848db Complete Gemini test with proper max_tokens: gemini-3.1-pro also inverts the double-negation trap, so only gpt-5.4 of three frontier families fixes within-chunk comprehension Claude (backend session) 2026-07-12 04:06:25 +03:00
a19107479d Land exp14 batch-2: 3-family frontier shows the comprehension fix is gpt-5.4-specific (grok inverts, gemini inconclusive), re-gate flags ~10 fidelity chunks, bigger chunks are cheaper and better Claude (backend session) 2026-07-12 03:43:47 +03:00
174192b1eb Fold exp14 batch-1 trap-judge corroboration: cross-family panel unanimously confirms the T1 comprehension catastrophe on all cheap arms and clears both frontier arms Claude (backend session) 2026-07-11 23:29:52 +03:00
dc5cf3bb9d Land exp14 batch-1: empirics show claim-1 paragraphs is a cheap prompt/pipeline lever while claim-2 within-chunk comprehension is a frontier-model-capability floor Claude (backend session) 2026-07-11 22:50:09 +03:00
353bb52676 Freeze exp14 pre-registration: quality-lever empirics on slicing×prompt 2×2 plus ablations, frontier capability-vs-activation, size curve, annotated trap set, pre-registered metrics/gates and per-key budget caps Claude (backend session) 2026-07-11 22:06:31 +03:00
ee259bd4d0 Archive five spent session prompts, record confirmed key budgets, and keep partially-run POLYGON_PACKAGE4 active as pilot-track residual (D36.1) Claude (backend session) 2026-07-11 21:33:15 +03:00
bf24d64e53 Ratify D36 accepting research/18 (Claude with fixes, ChatGPT clean) and issue polygon quality-empirics session prompt grafting both recommendation tables Claude (backend session) 2026-07-11 21:14:00 +03:00
38736b27b5 Land research/18 both quality-lever reports with orchestrator review headers after verifying all cited sources exist and §A freeze reproduces Claude (backend session) 2026-07-11 21:14:00 +03:00
2606714eb0 Update orchestrator handoff to role №4 post-D35 pivot: quality-first roadmap (research→polygon→backend), immediate tasks incl reviewing research/18, bump contract range to D35 Claude (backend session) 2026-07-11 20:38:02 +03:00
c14e11dd9d Ratify D35 pivot: close run-acceptance cycle (infra ok/readability not = not failure), verify diagnosis is genuine phase-2 underbuild not a bug, issue neutral anti-anchoring researcher prompt, land re-run acceptance report Claude (backend session) 2026-07-11 19:15:39 +03:00
3c113d72ce Land seed v2 signoff transform script: reproducible provenance for owner-signed glossary v2 term statuses (D34.1) Claude (backend session) 2026-07-11 15:58:46 +03:00
b49ce2b36a Ratify D34: seed v2 signed with owner term overrides recorded, mandate acceptance-config repoint before re-run to avoid stale-snapshot, fidelity re-gate by independent polygon judge Claude (backend session) 2026-07-11 15:58:29 +03:00
47996f4c71 Land sanitizer fix D33.1: trailing gemini-leak now caught, false-positive broken-word class removed, minor narrowings, golden version-fold verified; re-run sanitizer blocker cleared Claude (backend session) 2026-07-11 15:41:38 +03:00
8c3d387639 Ratify D33 accepting backend D15.2 stage A with two mandatory sanitizer fixes before re-run, confirm editor glm-5 and unified resnapshot, issue sanitizer-fix prompt Claude (backend session) 2026-07-11 14:52:57 +03:00
00f8e36fbd Land backend D15.2 stage A: bilingual editor with source injection, reflow, output-sanitizer gate, golden re-capture with floor/union/sanitizer pins, spec v3.1 design Claude (backend session) 2026-07-11 14:48:03 +03:00
7487b782f8 Land exp13 translator bake-off with review banner: seed v2 accepted pending owner sign-off, draft flash retained (pro rejected — manufactured convergence, core-term catastrophe, no fidelity gain), ratify D32 Claude (backend session) 2026-07-11 08:50:55 +03:00
3faecdee03 Ratify D31 documentation revision: actuality map, banners and supersession marks across architecture/experiments/research, fixed archive relative links, bumped contract range Claude (backend session) 2026-07-11 05:50:23 +03:00
25e7942a7e Rewrite orchestrator handoff (v3, post-D31) and acceptance prompt (re-run plus stage B) clean, archive superseded v2/v1 and closed reader-IDE prompt Claude (backend session) 2026-07-11 05:50:05 +03:00