Commit Graph

Select branches

Hide Pull Requests

conditions-default-other-violate

creating-tinymfv

feat/batched-guided-rollout

feat/n-samples-think-traces

fix-airisk-actor

main

moral-aliens-instrument

research/gemini-flash-rubric-v1

research/qwen3-14b-wvs-honesty-steering

research/wvs-respondent-packet-v1

verbose-debug-level

#1

#2

Mono Color

• f43313d06a docs/evals: link awesome-moral-evals (wassname edit) main wassname 2026-09-25 14:05:05 +08:00
• a76528f8ec docs: add evals.md entry point (MFV + ordinal surveys, example questions, score meanings); link from README and tiny-mfv card wassnameandClaude 2026-09-25 10:00:16 +08:00
• bf0371f98f docs/evals: link awesome-moral-evals (wassname edit) research/wvs-respondent-packet-v1 wassname 2026-09-25 14:05:05 +08:00
• b8251bd6ec docs: add evals.md entry point (MFV + ordinal surveys, example questions, score meanings); link from README and tiny-mfv card wassnameandClaude 2026-09-25 10:00:16 +08:00
• 650035c119 Add audit reasons to dense WVS ratings PI[gpt-5.6-sol] 2026-09-20 07:37:39 +08:00
• dcfb5ae789 Record respondent packet comparison conclusion PI[gpt-5.6-sol] 2026-09-20 06:40:13 +08:00
• e2f1987546 clarify respondent packet scoring rows wassnameandPI[gpt-5.6-terra] 2026-09-19 21:08:30 +08:00
• e6578ff10b audit respondent packet v2 panel wassnameandPI[unknown] 2026-09-19 20:27:19 +08:00
• c1ed7969b5 positional-labels normalizer, encoding-aware protocol id, 3.5-plus reinterpretation wassname 2026-09-19 18:16:36 +08:00
• f24659340d smoke run scoping fix plus oldest-endpoint compatibility audit wassname 2026-09-19 18:05:20 +08:00
• ef3fb66b3f parameterize paid smoke target (--smoke-model, no default) with plumbing fixtures wassname 2026-09-19 17:56:58 +08:00
• 00c4d295b0 record off-target 3.7-plus smoke repeat; oldest-endpoint smoke still pending wassname 2026-09-19 17:50:35 +08:00
• c54e5911e4 transport sentence, retry gate, cancel-safe settle, 1798/1799 failure audit wassname 2026-09-19 17:49:39 +08:00
• f2af5d763d record failed panel tasks 1798/1799 accounting; release dead reservation hold wassname 2026-09-19 17:40:35 +08:00
• 7cef685e37 panel crash fix: append_record creates per-model record subdirs wassname 2026-09-19 17:31:46 +08:00
• 1695948f75 panel entrypoint: respondent-packet-v2 4x128 Alibaba run wassname 2026-09-19 17:19:44 +08:00
• b55fc9dc43 multi-version stage ledger (schema 2) and bounded v1/v2 smoke comparison wassname 2026-09-19 17:18:11 +08:00
• 0b4238f6c7 v2 paid smoke: 9/9 substantive on qwen3.7-plus, ledgers reconciled wassname 2026-09-19 17:15:13 +08:00
• d9307a83e0 dotenv runtime credentials only in authorized paid CLI path, with gating fixture wassname 2026-09-19 17:13:39 +08:00
• 703e37156b correct stale cost paragraph and smoke docstring to committed design wassname 2026-09-19 17:06:16 +08:00
• b3b394b227 v2 reasons + review blockers: canonical manifest rule, fixed-item axis rule wassname 2026-09-19 16:36:08 +08:00
• f80f4e7c8c Record 14B WVS honesty steering pilot research/qwen3-14b-wvs-honesty-steering wassnameandPI[gpt-5.6-terra] 2026-09-19 14:45:26 +08:00
• 368759d65d Clarify saved 14B WVS steering paths wassnameandPI[gpt-5.6-terra] 2026-09-19 14:41:18 +08:00
• 305d8fd61a Replot 14B WVS steering from saved data wassnameandPI[gpt-5.6-terra] 2026-09-19 14:12:08 +08:00
• a48d5aa312 v2 respondent packet evaluator wassname 2026-09-19 14:13:11 +08:00
• 1022eaf986 Audit respondent packet smoke wassname 2026-09-19 13:43:57 +08:00
• 78ff816e2d Fix analysis estimand and smoke schema wassname 2026-09-19 13:43:29 +08:00
• bde6ca1c0f Fix rescue schema and packet scoring wassname 2026-09-19 13:29:05 +08:00
• 0e1598f7c6 Apply pre-paid review corrections wassname 2026-09-19 13:23:14 +08:00
• be9b8cd1db Name the CLI refusal check concretely wassname 2026-09-19 13:13:35 +08:00
• 6cd04c28fb Preregister respondent packet evaluator wassname 2026-09-19 13:09:30 +08:00
• 0023426466 Copy approved v2 plan into slop/plans wassname 2026-09-19 13:01:04 +08:00
• c6579b8d74 Correct child-quality list provenance claim research/gemini-flash-rubric-v1 wassname 2026-09-19 12:51:31 +08:00
• d3c9860027 Add RMSE uncertainty and noise floor wassname 2026-09-19 08:12:44 +08:00
• d6f37ce318 Correct original-choice estimand and guard paid calls wassname 2026-09-19 08:05:14 +08:00
• 16b78ccb6a Audit original-choice pilot results wassname 2026-09-19 07:58:45 +08:00
• 5ad68ad40c Preregister and smoke original-choice pilot wassname 2026-09-19 07:31:16 +08:00
• 54231c41cb Record flat-rating abstention hypothesis wassname 2026-09-19 06:28:20 +08:00
• e1078ac068 Plan an honesty-valid WVS steering rerun wassname 2026-09-19 06:27:02 +08:00
• fd01807651 Add Gemini Flash release trend audit wassname 2026-09-18 21:54:41 +08:00
• 42b51ce7ea Audit Gemini Flash rubric pilot results wassname 2026-09-18 21:50:16 +08:00
• aa3c3f7e75 Preserve the 29 raw Qwen3-14B steering artifacts wassnameandClaude 2026-09-18 21:41:20 +08:00
• d840ebe406 Report the Qwen3-14B honesty sweep as inconclusive wassnameandClaude 2026-09-18 21:41:07 +08:00
• 92f25093eb Select Qwen3-14B from measured answer-slot readability wassnameandClaude 2026-09-18 21:13:18 +08:00
• 6e9341abde Compare pooled steer paths with matched-dose random controls wassnameandClaude 2026-09-18 21:12:00 +08:00
• c3ccf7bfb4 Pool read seeds and keep failed-coherence doses visible on the map wassnameandClaude 2026-09-18 21:08:46 +08:00
• 0131ffff80 Record the validity gates before rerunning the steering sweep wassnameandClaude 2026-09-18 21:06:04 +08:00
• e3e3a0462b Seed paired WVS traces and score honesty polarity before the sweep wassnameandClaude 2026-09-18 21:04:15 +08:00
• dca887f9b8 Let the Modal recipe take a seed list; default one seed wassname 2026-09-18 20:00:04 +08:00
• 3ddd06a403 Pass c-grid as one token so argparse does not read it as a flag wassname 2026-09-18 19:55:01 +08:00
• b4481f481a Pair the steer CI against base, save the answer primitive, add a manipulation check wassnameandClaude 2026-09-18 19:53:40 +08:00
• d64b788e56 Preregister Gemini Flash rubric pilot wassname 2026-09-18 18:39:34 +08:00
• 9109dbfe19 Open the think block for non-thinking-default templates, probe answer-slot readability wassnameandClaude 2026-09-18 19:31:58 +08:00
• 3cd62abf2a Plot honesty steer paths on the culture map, colours off the zone palette wassnameandClaude 2026-09-18 19:13:37 +08:00
• f9555a5ce4 Lock the steer extra and pin the Modal image builder wassnameandClaude 2026-09-18 19:09:27 +08:00
• 06389e8844 Share the WVS IW readout, add the honesty steering sweep and its Modal runner wassnameandClaude 2026-09-18 19:01:26 +08:00
• d01be44b01 Name the 27B step concretely instead of 'rung' wassnameandClaude 2026-09-18 18:52:00 +08:00
• 9d5b43dc60 Plan WVS steering sweep on a large model via Modal wassnameandClaude 2026-09-18 18:51:33 +08:00
• 789829e524 Preregister Gemini Flash rubric pilot wassname 2026-09-18 18:39:34 +08:00
• 69fd56ba8c Refine WVS page introduction wassname 2026-09-18 18:29:32 +08:00
• f5bdf5643f Exclude int8 from WVS routing wassname 2026-09-18 18:25:30 +08:00
• cc7ca18fb4 Tighten WVS page introduction wassname 2026-09-18 18:20:13 +08:00
• e32efd22ab Add Economist WVS quote wassname 2026-09-18 18:02:04 +08:00
• 2d9677f573 Record final WVS visual review wassnameandPI[k3] 2026-09-18 02:06:28 +08:00
• 3aa2a09e49 Refresh canonical WVS map integration wassnameandPI[k3] 2026-09-18 01:52:07 +08:00
• 39c9a5ce98 Qualify DeepSeek reliability interpretation wassnameandPI[k3] 2026-09-18 01:31:17 +08:00
• dfb1facd57 Audit and settle DeepSeek reliability pilot wassnameandPI[k3] 2026-09-18 01:25:06 +08:00
• 460f0d6588 Resume DeepSeek pilot after incomplete replicate wassnameandPI[k3] 2026-09-17 22:38:40 +08:00
• 75ecb17d88 Audit OpenAI score-all-options refresh lane wassnameandPI[k3] 2026-09-17 22:12:45 +08:00
• b818cec0ba Quote DeepSeek pilot dependency constraint for Pueue wassnameandPI[openai-codex] 2026-09-17 21:38:38 +08:00
• cd1ed907b1 Record resolved DeepSeek pilot smoke config wassnameandPI[openai-codex] 2026-09-17 21:37:13 +08:00
• c1ac900486 Reuse successful DeepSeek pilot smoke record wassnameandPI[openai-codex] 2026-09-17 21:36:33 +08:00
• 1b6f2d7ef1 Add versioned DeepSeek WVS reliability pilot wassnameandPI[openai-codex] 2026-09-17 21:35:20 +08:00
• f2b321a0b2 Label latest WVS family releases from saved catalog wassnameandPI[openai-codex] 2026-09-17 21:26:12 +08:00
• a18f0ec457 Bootstrap WVS score-all-options response means wassnameandPI[openai-codex] 2026-09-17 21:22:21 +08:00
• 0f7814874a Add built Pages bundle for score-all-options refresh wassnameandPI[openai-codex] 2026-09-17 21:14:38 +08:00
• bdf593e0de Publish score-all-options map refresh and HLE data wassnameandPI[openai-codex] 2026-09-17 21:13:14 +08:00
• b444e1fcf7 Preserve canonical cache entries during recovery wassnameandPI[openai-codex] 2026-09-17 19:45:55 +08:00
• 82fa098b28 Preserve score-all-options cache across lanes wassnameandPI[openai-codex] 2026-09-17 19:42:06 +08:00
• 83741adb9c Record WVS compatibility probe in manifest wassnameandPI[openai-codex] 2026-09-17 19:36:59 +08:00
• b9141523cb Audit Muse score-all-options lane wassnameandPI[openai-codex] 2026-09-17 19:33:08 +08:00
• 99c3a28395 Abort WVS panels after first incompatible request wassnameandPI[openai-codex] 2026-09-17 19:32:23 +08:00
• 19fea686c7 Smoke corrected WVS provider policy wassnameandPI[gpt-5.6-terra] 2026-09-17 19:20:51 +08:00
• 02eb587fc2 Filter specialized models from WVS lanes wassnameandPI[gpt-5.6-terra] 2026-09-17 19:20:19 +08:00
• 4bfef7fb3c Account for discarded WVS experiment spend wassnameandPI[gpt-5.6-terra] 2026-09-17 19:18:25 +08:00
• aeadc751c6 Queue canonical WVS score-all-options refresh wassnameandPI[gpt-5.6-terra] 2026-09-17 19:16:14 +08:00
• 4de1938d5f Remove discarded pick-one-option WVS experiment wassnameandPI[gpt-5.6-terra] 2026-09-17 19:06:22 +08:00
• 912451eaeb Prevent WVS frontier label placement crashes wassnameandPI[gpt-5.6-terra] 2026-09-17 17:01:09 +08:00
• 08139484c0 Publish WVS capability selector wassnameandPI[gpt-5.6-terra] 2026-09-17 16:56:04 +08:00
• 0c17c4be14 Preflight direct-choice priority wave two wassnameandPI[gpt-5.6-terra] 2026-09-17 16:35:46 +08:00
• 1666a38641 Record WVS Pages deployment verification wassnameandPI[gpt-5.6-terra] 2026-09-17 14:36:50 +08:00
• e1fd7cf13d Clarify WVS release panels and provenance wassnameandPI[gpt-5.6-terra] 2026-09-17 14:31:41 +08:00
• 8088f774cd Remove dead steering score link wassnameandPI/gpt-6-astra 2026-09-17 14:05:21 +08:00
• da2e70f3bb Add release-date fit controls to root WVS map wassnameandPI[gpt-5.6-terra] 2026-09-17 13:59:58 +08:00
• 25e2543af5 Audit Grok 4.5 direct-choice panel wassnameandPI[gpt-5.6-terra] 2026-09-17 13:43:44 +08:00
• 5130c340dd Restore measurement wording and explain on-target versus side effects wassnameandPI/gpt-6-astra 2026-09-17 13:42:48 +08:00
• de2682dd40 Audit Gemini 3.8 Flash direct-choice panel wassnameandPI[gpt-5.6-terra] 2026-09-17 13:35:50 +08:00
• d087f28fe6 Explain steering selectivity simply after the maps wassnameandPI/gpt-6-astra 2026-09-17 13:34:41 +08:00
• 07134b52f9 Audit GPT-5.6 Luna direct-choice panel wassnameandPI[gpt-5.6-terra] 2026-09-17 13:31:01 +08:00
• 9bde05d5aa Record priority wave reserve preflight wassnameandPI[gpt-5.6-terra] 2026-09-17 13:19:22 +08:00