wassname/moral-maps
1
0
Fork 0
mirror of https://github.com/wassname/moral-maps.git synced 2026-09-22 13:20:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Clarify direct-choice audit checks

Browse Source 
Co-Authored-By: PI[gpt-5.6-terra] <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-09-17 11:54:00 +08:00
co-authored by PI[gpt-5.6-terra]
parent 84aca5b60c
commit a015bbf7ea
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/wvs_direct_choice_production_audit.py
+1 -1
View File
@@ -290,7 +290,7 @@ def main() -> None:
"",
"## Decision",
"",
"1. Resolve-condition verdict: **met for mechanics and preregistered diagnostics; not yet met for a broad construct migration.** The complete panel, parser, balance, usage, and cache gates pass. The panel also records schedule/direction variation rather than hiding it.",
"1. Resolve-condition verdict: **met for mechanics and preregistered diagnostics; not yet met for a broad construct migration.** The complete panel, parser, position-balance, usage, and cache-replay checks pass. The panel also records schedule/direction variation rather than hiding it.",
"2. Prediction check: the balanced schedule predicted exact position exposure for n=2/4/10, nearest balance for n=3, and no automatic hard exclusion from the position TV diagnostic. These are supported. No prediction claimed that all items would have low schedule-half TV.",
"3. Earliest unsupported link: direct choices under this assistant-behavior prompt are a stable substitute for dense-rated WVS coordinates.",
"4. Validity: define invalid as a result suitable for merging into published rated coordinates or for authorizing the wider paid expansion. P(invalid for that use) is highly likely, about 0.80. The ledger and direct-choice behavioral observations are credible under the exact protocol.",
slop/audits/20260917_wvs_gemini37_direct_choice_production_task_1631.md
+1 -1
View File
@@ -102,7 +102,7 @@ No item crosses the preregistered position-bias warning TV >0.25. The largest se
## Decision
1. Resolve-condition verdict: **met for mechanics and preregistered diagnostics; not yet met for a broad construct migration.** The complete panel, parser, balance, usage, and cache gates pass. The panel also records schedule/direction variation rather than hiding it.
1. Resolve-condition verdict: **met for mechanics and preregistered diagnostics; not yet met for a broad construct migration.** The complete panel, parser, position-balance, usage, and cache-replay checks pass. The panel also records schedule/direction variation rather than hiding it.
2. Prediction check: the balanced schedule predicted exact position exposure for n=2/4/10, nearest balance for n=3, and no automatic hard exclusion from the position TV diagnostic. These are supported. No prediction claimed that all items would have low schedule-half TV.
3. Earliest unsupported link: direct choices under this assistant-behavior prompt are a stable substitute for dense-rated WVS coordinates.
4. Validity: define invalid as a result suitable for merging into published rated coordinates or for authorizing the wider paid expansion. P(invalid for that use) is highly likely, about 0.80. The ledger and direct-choice behavioral observations are credible under the exact protocol.