Repository navigation
M-J.4.4 B -- close the two Gate-C corpus residuals - #18
Merged
Merged
Conversation
…rer polish)
The blind judge left two diminishing-returns renderer residuals on the otherwise 4.0/4.0 corpus:
- suspicion-avoidant: the strong "stays withdrawn and wary" beat could fire once PER withdrawal run
(the adjacent-dup guard only catches consecutive lines), reading as a verbatim repeat. The beat now
lands ONCE; a continuing withdrawal collapses into "keeps to himself, the wariness not lifting".
- welf-betrayal: a single REACT at the very end left the arc closing on a flash of temper, the chill
never landing. A betrayal now always settles -- if the flare fired but never went cold, a closing
note lands the withdrawn trust ("the cold stays in him; the trust does not come back").
Deterministic renderer-only change (no engine/overlay touch); 399 passed +4 skipped; ruff clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016rWVvvhWijSuLMEhzjiUhu
…confirms residuals closed)
A single paced blind Sonnet agent (NEVER a burst -- token rule) re-judged the 14 affected vignettes:
- betrayal FIXED: lands_chill true, believability 4.1 / curve 4.3 -- all 7 land on settled cold distrust,
never on a flash of temper.
- suspicion-avoidant: the verbatim repeat is gone; the judge's only residual note was that the wary
open/close felt semantically near-identical. Addressed by making the continuing-withdrawal beat DEVELOP
("by the end he has shut the man out entirely, giving him nothing") instead of restating the wary line.
Deterministic renderer-only; 399 passed +4 skipped; ruff clean; goldens untouched.
Verdict: calibration/moral_corpus_residual_verdict.json.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016rWVvvhWijSuLMEhzjiUhu
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
B — close the two minor Gate-C corpus residuals (renderer polish)
The blind judge left two diminishing-returns renderer residuals on the otherwise 4.0/4.0 corpus. Both now fixed; a single paced blind Sonnet agent (never a burst — token rule) confirmed on the 14 affected vignettes.
welf-betrayal — arc ended on a flare
A single REACT at the very end left the arc closing on a flash of temper. A betrayal now always lands the chill: if the flare fired but never went cold, a closing note settles the withdrawn trust.
→ judge:
lands_chill: true, betrayal believability 4.1 / curve 4.3, "all 7 land on settled cold distrust, never on a flash of temper".suspicion-avoidant — withdrawn-wary beat repeated
The strong "stays withdrawn and wary" beat could fire once per withdrawal run (the adjacent-dup guard only caught consecutive lines). It now lands once; a continuing withdrawal develops ("by the end he has shut the man out entirely, giving him nothing") rather than restating the wary line.
→ judge:
verbatim_repeat_feel: false; the only residual note (open/close felt near-identical) addressed by the escalating close.Status
Deterministic renderer-only change (no engine/overlay/goldens touch). 399 passed, 4 skipped; ruff clean. Verdict:
calibration/moral_corpus_residual_verdict.json.🤖 Generated with Claude Code