Repository navigation
M-J.4.4 A1+A2 -- calibrate half-lives (judge-validated) + write values - #17
Merged
Merged
Conversation
…lidated values A1 (paced multi-day blind judge) + A2 (write the validated values, re-run the pre-filter): - SUSPICION half-life CALIBRATED 48h -> 24h. eval/moral_multiday.render_suspicion_multiday renders a passing- suspicion persistence arc; a blind Sonnet agent ranked 24h>48h>14d -- 24h "fast enough to feel healthy, slow enough to feel real", 48h "slightly rigid", 14d "pathological" without evidence. Written into the overlay (relabelled CALIBRATED). - GUILT closed via the A3 Secret moral_weight drip (serious feel ~72h), so the 18h base stays the MINOR anchor (relabelled CALIBRATED). - perceived_injustice NOT judged -- CONFOUNDED (blame_other vents it, so it discharges rather than decays); its behavior is covered by the Gate-C corpus (accusation 3.75). Spec anchor kept. - fast states (exposure_anxiety/avoidance 60min) and derived slow states (rumination/repair/load) keep spec anchors -- low independent criticality. Re-ran the deterministic pre-filter after the magnitude change: still 18/81 (no litmus broken). Full suite 399 passed +4 skipped; goldens byte-identical (overlay opt-in); ruff clean. Evidence-grade suspicion (14d) split is a future follow-up (like minor/serious guilt). Verdict: calibration/moral_halflife_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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
A1+A2 — calibrate the remaining half-lives + write the validated values
Continues the moral calibration: judge the remaining slow half-lives (paced, single agents — never a big burst), then write the validated values into the overlay and re-run the deterministic pre-filter (the calibration lesson: re-check after any magnitude change).
Calibrated
render_suspicion_multidayrenders a passing-suspicion persistence arc; a blind Sonnet judge ranked 24h > 48h > 14d: 24h "fast enough to feel healthy, slow enough to feel real", 48h "slightly rigid", 14d "pathological" without evidence. Written into the overlay (relabelled CALIBRATED).moral_weightdrip (serious feel ~72h) — so the 18h base stays the MINOR anchor (relabelled CALIBRATED).Not judged (honestly)
blame_other-vents it, so it discharges rather than purely decaying → multi-day persistence isn't a clean signal. Its behaviour is covered by the Gate-C corpus (accusation 3.75). Spec anchor 24h kept.Status
Deterministic pre-filter still 18/81 after the change (no litmus broken). Full suite 399 passed, 4 skipped; goldens byte-identical (overlay opt-in); ruff clean. Evidence-grade suspicion (14d) split is a future follow-up like minor/serious guilt. Verdict:
calibration/moral_halflife_verdict.json.🤖 Generated with Claude Code