wassname

wassname synced commits to refs/heads/main at wassname/ml-debug from mirror 2026-07-25 13:20:34 +08:00

38ddb544d9 Make 'Numbers worth knowing' quote-anchored with source-credence + calibrated language

3f35addb7f Expand 'Numbers worth knowing' bullets with point + context for readability

23d3331ee2 Fix two varglite nits: complete Lost-in-Middle quote, add context to config block

f10151e661 Add varglite 2026 judge lit-review + operational-numbers digest

3b7e82d402 Add 2026 lit-search batch to judge-bias evidence (position/self/reasoning/context-rot + benchmark index)

Compare 5 commits »

wassname synced commits to refs/heads/main at wassname/adapters_as_hypotheses from mirror 2026-07-20 12:10:36 +08:00

eee889676a add flatness-seeking adapter family

wassname synced commits to refs/heads/main at wassname/persona-steering-template-library from mirror 2026-07-19 11:26:32 +08:00

a64f0438a0 Document Inspect validation outputs

dd218c8223 Point persona runbooks at the Inspect evaluator

56a35d92a7 Run persona validation through Inspect AI

Compare 3 commits »

wassname synced commits to refs/heads/main at wassname/moral-maps from mirror 2026-07-19 11:25:43 +08:00

01e9026641 guided: expose top-k sampling control

d665493119 guided: emit exactly one open think marker

00105855dc journal: random-baseline confound + raw per-choice output question (2026-07-17a)

Compare 3 commits »

wassname synced commits to refs/heads/main at wassname/ml_debug from mirror 2026-07-19 11:25:31 +08:00

9d00a069c6 Require reading both sides of LLM evaluations

wassname synced commits to refs/heads/main at wassname/ml-debug from mirror 2026-07-19 11:25:27 +08:00

9d00a069c6 Require reading both sides of LLM evaluations

wassname synced commits to refs/heads/main at wassname/adapters_as_hypotheses from mirror 2026-07-19 11:20:42 +08:00

7bfcdf06ce restore catalog structure for low-curvature evidence

38e3ab4be1 Merge remote-tracking branch 'origin/main'

a91f2912f9 add quote-anchored low-curvature adapter section

f4750fcfd1 match

3ed3d8cb83 rtvn

Compare 5 commits »

wassname synced new reference refs/heads/main to wassname/soul-ab-test-scope-test from mirror 2026-07-18 12:40:46 +08:00

wassname synced commits to refs/heads/main at wassname/soul-ab-test-scope-test from mirror 2026-07-18 12:40:46 +08:00

wassname synced commits to refs/heads/main at wassname/persona-steering-template-library from mirror 2026-07-18 12:30:36 +08:00

64fafcf301 Remove prose repeated by the steering example

81096c6d4c Show steering from paired prompts to inference

90ccc40391 Introduce steering before the runbook

baac06a2be Replace Quarto site with concise README

bdebe3fb25 Simplify persona scenario and template selection

Compare 5 commits »

wassname synced commits to refs/heads/main at wassname/moral-maps from mirror 2026-07-16 11:20:13 +08:00

43349a3361 metrics: fail on invalid coherence inputs and unpaired rows

a9a01327e7 Rename showcase readout dlogit->dclr to match steering-lite

7f873ffb42 Add shared gated_selectivity + si_flips metric

Compare 3 commits »

wassname synced commits to refs/heads/fix/chat-transient-retries at wassname/openrouter-python-sdk-retry-er… from mirror 2026-07-15 11:22:48 +08:00

47ba6304b3 Merge branch 'main' into fix/chat-transient-retries

8b505afe3d chore: 🐝 Update SDK - Generate (spec change merged) 0.11.33 (#454)

9408179f89 chore: update OpenAPI spec from monorepo (#453)

a3d8d7084d chore: 🐝 Update SDK - Generate (spec change merged) 0.11.32 (#452)

f07b760c24 chore: update OpenAPI spec from monorepo (#451)

Compare 5 commits »

wassname synced commits to refs/heads/main at wassname/moral-maps from mirror 2026-07-15 11:22:17 +08:00

9237aa9faa readouts: add clr (centered log-ratio) categorical readout

de5c1488c2 guided: drop stale comment claiming descs were shortened (they are verbatim Clifford options)

Compare 2 commits »

wassname synced commits to refs/heads/main at wassname/jsteer from mirror 2026-07-15 11:21:14 +08:00

6836cf5783 Dedup marimo-pair/retro-marimo-pair into home skills

35230357bb Sync marimo-pair provenance header with home skill copy

Compare 2 commits »

wassname synced commits to refs/heads/main at wassname/ml_debug from mirror 2026-07-14 03:50:09 +08:00

cdab6cbbb9 Record ML debugging benchmark results

a0fcfa291e Add blinded ML debugging benchmark harness

Compare 2 commits »

wassname synced commits to refs/heads/benchmark/skill-ab at wassname/ml_debug from mirror 2026-07-14 03:50:08 +08:00

cdab6cbbb9 Record ML debugging benchmark results

wassname synced commits to refs/heads/main at wassname/ml-debug from mirror 2026-07-13 07:30:52 +08:00

cdab6cbbb9 Record ML debugging benchmark results

a0fcfa291e Add blinded ML debugging benchmark harness

Compare 2 commits »

wassname synced commits to refs/heads/benchmark/skill-ab at wassname/ml-debug from mirror 2026-07-13 07:30:51 +08:00

cdab6cbbb9 Record ML debugging benchmark results

wassname synced commits to refs/heads/main at wassname/jsteer from mirror 2026-07-13 07:29:37 +08:00

5670cef162 Resolve model commit from cached config

b927dc64fd Use the shared full-precision edge evaluator

1c927e95ba Keep demo reproduction on the fitted layer band

77998d6bad Make steering demo runs selectable and append-only

89fc7b1068 wip

Compare 5 commits »

wassname synced commits to refs/heads/main at wassname/ml_debug from mirror 2026-07-13 03:40:11 +08:00

9774c4bb1d Record skill audit evidence

0be4323312 Add fail-fast skill audit

fa534cf44e Correct overconfident debugging advice

Compare 3 commits »