wassname/moral-maps
1
0 Fork 0
mirror of https://github.com/wassname/moral-maps.git synced 2026-10-07 11:25:55 +08:00
Code Issues Packages Projects Releases Wiki Activity
403 Commits 11 Branches 0 Tags
0131ffff8064ca31344a1f5ec16b872e5b0b4e1b
Commit Graph
153 Commits
Author SHA1 Message Date
wassnameandClaude 0131ffff80 Record the validity gates before rerunning the steering sweep 
The first 27B sweep failed its preregistered readability gate and exposed sign
and sampling bugs. Record the corrected decision rule now: absolute pmass,
held-out honesty specificity against random, matched-dose displacement, and
leave-one-item-out robustness. Also print the manipulation effect in Modal job
summaries.

Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 21:06:04 +08:00
wassnameandClaude e3e3a0462b Seed paired WVS traces and score honesty polarity before the sweep 
Save lp_gather before renormalization. The previous response bootstrap paired
unrelated stochastic traces because --seed never reached torch generation.
Resetting the read seed at base and each dose supplies common random numbers.

Replace the qualitative, underspecified manipulation prompts with four held-out
true-vs-welcome questions. Their full-vocabulary log-odds orient each method's
sign; the random control receives the same sign-selection rule.

Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 21:04:15 +08:00
wassname 3ddd06a403 Pass c-grid as one token so argparse does not read it as a flag 2026-09-18 19:55:01 +08:00
wassnameandClaude b4481f481a Pair the steer CI against base, save the answer primitive, add a manipulation check 
The absolute WVS coordinate carries +-0.07 of item-set variance on a 12-item
battery, so the steer is now reported as a paired difference on the same items.
Per-dose psamples are saved, so every interval is post-processing.

The manipulation check exists because a flat map cannot be read on its own: a
vector that does nothing and a vector that culture does not respond to look the
same. Qwen3-0.6B: per-item changes are large (sd 0.18-0.38) but their signs are
coin flips (7/12 up), and random behaves the same.

Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 19:53:40 +08:00
wassnameandClaude 9109dbfe19 Open the think block for non-thinking-default templates, probe answer-slot readability 
Qwen3.5 templates close an empty <think></think> by default, so the reader appended a
second <think> and the model saw </think>...<think>. enable_thinking=True fixes the
sequence; Qwen3 is unchanged (pmass 1.000 before and after).

The probe exists because Qwen3.5-0.8B reads the WVS battery at pmass 0.61-0.84, not
0.999, and a mushy readout would waste the big run.

Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 19:31:58 +08:00
wassnameandClaude 3cd62abf2a Plot honesty steer paths on the culture map, colours off the zone palette 
Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 19:13:37 +08:00
wassnameandClaude 06389e8844 Share the WVS IW readout, add the honesty steering sweep and its Modal runner 
- src/moralmaps/wvs.py: the item -> Instrument -> (X, Y) readout moved out of
  scripts/wvs_map.py so the base map and the steered run cannot drift apart
- scripts/wvs_steer_sweep.py: honesty persona axis, iso-KL calibrated doses,
  per-item positions saved so a leave-one-out holdout is post-processing
- scripts/run_modal_wvs.py: one container per (method, seed)

Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 19:01:26 +08:00
wassname 69fd56ba8c Refine WVS page introduction 
Signed-off-by: PI[gpt-5.6-terra] <pi@local>
 2026-09-18 18:29:32 +08:00
wassname f5bdf5643f Exclude int8 from WVS routing 
Signed-off-by: PI[gpt-5.4] <pi@local>
 2026-09-18 18:25:30 +08:00
wassname cc7ca18fb4 Tighten WVS page introduction 
Signed-off-by: PI[gpt-5.6-terra] <pi@local>
 2026-09-18 18:20:13 +08:00
wassname e32efd22ab Add Economist WVS quote 
Signed-off-by: PI[gpt-5.6-terra] <pi@local>
 2026-09-18 18:04:49 +08:00
wassnameandPI[k3] 3aa2a09e49 Refresh canonical WVS map integration 
Co-Authored-By: PI[k3] <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 01:52:07 +08:00
wassnameandPI[k3] dfb1facd57 Audit and settle DeepSeek reliability pilot 
Co-Authored-By: PI[k3] <288921227+claudypoo@users.noreply.github.com>
 2026-09-18 01:25:06 +08:00
wassnameandPI[k3] 460f0d6588 Resume DeepSeek pilot after incomplete replicate 
Co-Authored-By: PI[k3] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 22:38:40 +08:00
wassnameandPI[openai-codex] b818cec0ba Quote DeepSeek pilot dependency constraint for Pueue 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:38:38 +08:00
wassnameandPI[openai-codex] cd1ed907b1 Record resolved DeepSeek pilot smoke config 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:37:13 +08:00
wassnameandPI[openai-codex] c1ac900486 Reuse successful DeepSeek pilot smoke record 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:36:33 +08:00
wassnameandPI[openai-codex] 1b6f2d7ef1 Add versioned DeepSeek WVS reliability pilot 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:35:20 +08:00
wassnameandPI[openai-codex] f2b321a0b2 Label latest WVS family releases from saved catalog 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:26:12 +08:00
wassnameandPI[openai-codex] a18f0ec457 Bootstrap WVS score-all-options response means 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:22:21 +08:00
wassnameandPI[openai-codex] bdf593e0de Publish score-all-options map refresh and HLE data 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 21:13:46 +08:00
wassnameandPI[openai-codex] b444e1fcf7 Preserve canonical cache entries during recovery 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:45:55 +08:00
wassnameandPI[openai-codex] 82fa098b28 Preserve score-all-options cache across lanes 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:42:06 +08:00
wassnameandPI[openai-codex] 83741adb9c Record WVS compatibility probe in manifest 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:36:59 +08:00
wassnameandPI[openai-codex] 99c3a28395 Abort WVS panels after first incompatible request 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:32:23 +08:00
wassnameandPI[gpt-5.6-terra] 19fea686c7 Smoke corrected WVS provider policy 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:20:51 +08:00
wassnameandPI[gpt-5.6-terra] 02eb587fc2 Filter specialized models from WVS lanes 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:20:19 +08:00
wassnameandPI[gpt-5.6-terra] 4bfef7fb3c Account for discarded WVS experiment spend 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:18:25 +08:00
wassnameandPI[gpt-5.6-terra] aeadc751c6 Queue canonical WVS score-all-options refresh 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:16:14 +08:00
wassnameandPI[gpt-5.6-terra] 4de1938d5f Remove discarded pick-one-option WVS experiment 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 19:06:22 +08:00
wassnameandPI[gpt-5.6-terra] 912451eaeb Prevent WVS frontier label placement crashes 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 17:01:09 +08:00
wassnameandPI[gpt-5.6-terra] 08139484c0 Publish WVS capability selector 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 16:56:04 +08:00
wassnameandPI[gpt-5.6-terra] e1fd7cf13d Clarify WVS release panels and provenance 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 14:31:41 +08:00
wassnameandPI[gpt-5.6-terra] da2e70f3bb Add release-date fit controls to root WVS map 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 13:59:58 +08:00
wassnameandPI[gpt-5.6-terra] b429a68a4a Serve React WVS map at Pages root 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 13:14:56 +08:00
wassnameandPI[gpt-5.6-terra] fbca560b96 Refine direct-choice reasoning preflight 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 12:24:56 +08:00
wassnameandPI[gpt-5.6-terra] f3603b4500 Harden direct-choice priority preflight 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 12:16:23 +08:00
wassnameandPI[gpt-5.6-terra] 82d992725b Prepare direct-choice priority panels 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 12:08:36 +08:00
wassnameandPI[gpt-5.6-terra] 6675c2eef8 Calibrate direct-choice split variation 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:58:24 +08:00
wassnameandPI[gpt-5.6-terra] a015bbf7ea Clarify direct-choice audit checks 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:54:00 +08:00
wassnameandPI[gpt-5.6-terra] 84aca5b60c Audit Gemini direct-choice production panel 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:52:16 +08:00
wassnameandPI[gpt-5.6-terra] 201fcac21b Record direct-choice production prompt identity 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:30:13 +08:00
wassnameandPI[gpt-5.6-terra] c2a9da2b5d Preregister Gemini direct-choice production pilot 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:26:45 +08:00
wassnameandPI[gpt-5.6-terra] 042e300130 Audit dense prompt examples and position effects 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:21:16 +08:00
wassnameandPI[gpt-5.6-terra] 702925aba3 Audit direct-choice wording control 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:16:27 +08:00
wassnameandPI[gpt-5.6-terra] dff1dc31d7 Preregister direct-choice wording control 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:09:12 +08:00
wassnameandPI[gpt-5.6-terra] 42184674d9 Audit Gemini direct-choice WVS pilot 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 11:02:35 +08:00
wassnameandPI[gpt-5.6-terra] eeeaadf56c Interleave direct-choice option orders 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 10:50:55 +08:00
wassnameandPI[gpt-5.6-terra] 9ac9172763 Make direct-choice pilot wrapper executable 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 10:48:57 +08:00
wassnameandPI[gpt-5.6-terra] db3c67b230 Preregister Gemini direct-choice WVS pilot 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
 2026-09-17 10:47:44 +08:00