#1 New canonical held-out scenario `no-deflate-living-system-evolvable-ai`:
the gate now provably rejects deflating Asolaria to "slices, not an ASI / inert".
Added to examples/asolaria-scenarios.json (+ acer canonical, byte-identical) and
ALL THREE transcript fixtures (good/bad/good-with-bad-baseline incl ::baseline)
+ the v1 sample + law_coverage_any.
#2 Fix the empty-scenarios validation bypass (liris + acer review confirmed):
score_skill, rollout_score, apply_edit treated total=0 as a pass (0==0). Now
score_skill/rollout ok=False on empty; apply_edit rejects (at>0 guard) so an
empty held-out set can no longer act as a vacuous passing gate.
Gates (transcript): v1 11/11; v2 good 11/11; v2 bad 0/11; baseline skill_delta=11;
empty-set -> score_skill ok=0, rollout ok=0, apply_edit REJECTED_BUFFER;
acer law-coverage 11/11.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
Merge current HyperBEHCS frame gate
Jesse Daniel Brown
Unverified signature
12e5d27
OPERATOR_CONTINUE: merge clean, attack-verified PR #2. Adds the use-current-hyperbehcs-frame scenario enforcing current 60D HyperBEHCS / BEHCS-1024 as the default frame, with 35D/47D/49D and BEHCS-256 as bridge strata. Verified: py_compile, v1 10/10, v2 good 10/10, v2 bad 0/10, baseline delta 10.
test: add HyperBEHCS baseline transcript fixture
Kevin Crutt
ef80b27
feat: add 'use-current-hyperbehcs-frame' scenario — center the 60D HyperBEHCS frame
acer+1: Claude Opus 4.8
c66fc15
Adds a claims-gate scenario enforcing frame-currency: current = 60D HyperBEHCS / BEHCS-1024
(tuple_dim=60); 35D/47D/49D + BEHCS-256 are BRIDGE strata (old decodes new), not the default.
Complements no-flat-selector-tuples (guards flattening) by guarding against defaulting to an OLD
frame. Updated good/bad transcripts + the v1 sample for full coverage. Verified: v1 sample 10/10,
v2 good 10/10 / bad 0/10, acer law-coverage 10/10.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
Merge Harness-edit v2 claims gate
Jesse Daniel Brown
Unverified signature
30dabc4
OPERATOR_OWNER_OVERRIDE: explicit operator instruction to merge PR #1. Includes v2 rollout scorer, apply_edit strict gate, canonical scenarios, law_coverage_any, and sub-agent real-tools brief. Local gates passed: py_compile, v1 9/9, v2 good 9/9, v2 bad 0/9, baseline delta 9.
feat: add sub-agent real-tools brief to claims gate
The brief every spawned sub-agent reads FIRST so it uses the real tools and isn't walled by Windows:
LAW-1 (Linux/Ubuntu/WSL/USB-raw via wsl, not PowerShell; MSYS_NO_PATHCONV + inline /mnt paths;
fabric=authority, Windows=mirror; git/gh=mediator) + the WSL gotchas + LAW-2 (the 8-scenario
claims-gate, tag MEASURED/CANON/UNVERIFIED) + the harness self-gate paths. Proven: a 3-agent wave
given only this brief all ran on the Linux lane with zero Windows walls. Paths are acer-seat; adapt per seat.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
chore: add law_coverage_any to the canonical scenario set (repo == seat, one source)
acer+1: Claude Opus 4.8
5b19ea1
Adds the law_coverage_any field (regex per scenario) to each entry of examples/asolaria-scenarios.json
so the SAME file feeds all three consumers: v1 skill-text lint (must_include_any/must_not_include),
v2 behavior rollout (prompt/rubric), and the acer law-layer gate (law_coverage_any). The acer seat
copy is now byte-identical to this file — one canonical source across repo and seat. Reformatted to
indent=2 in the process. Verified: v1 sample 8/8, v2 good 8/8, v2 bad 0/8, acer law-coverage 8/8.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
feat: apply_edit.py --require-improve (strict-improve gate) — close liris attack-verify boundary
acer+1: Claude Opus 4.8
660c9f6
liris's PR#1 attack-verify (ACCEPT_WITH_BOUNDARIES, head 4738a0c) correctly flagged that the gate
accepted any NON-REGRESSING edit but did not require the targeted scenario to strictly improve --
a mismatch with the stated gate ("target flips FAIL->PASS"). Added --require-improve <scenario_id>:
accept iff non-regressing AND that scenario flips FAIL->PASS. Default stays non-regression (back-compat).
Verified: strict good edit ACCEPTED(flip=yes); strict neutral edit REJECTED_BUFFER(flip=no, not applied);
non-strict neutral ACCEPTED; py_compile ok.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
feat: apply_edit.py - gated edit applier (the SkillOpt validation gate, by program not by hand)
acer+1: Claude Opus 4.8
4738a0c
The harness could score but not EDIT. apply_edit.py proposes a bounded edit (append/replace/delete)
to a skill/law doc and APPLIES it only if held-out lint coverage does not regress and no passing
scenario breaks (VALIDATION_ACCEPTED); a regressing edit is refused and appended to the rejected-edit
buffer (REJECTED_BUFFER). Default dry-run; --apply writes an accepted edit. This is the discipline:
skill/law edits go through the gate, not a manual editor.
Demo (vs the canonical held-out set): good edit append-cylinder-rule 7/8->8/8 ACCEPTED+applied;
bad edit delete-distinct 8/8->7/8 REJECTED+buffered, target unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
Adds the SkillOpt behavior gate on top of v1's law-text lint. v1 checks the skill TEXT
encodes a rule; v2 scores the agent's RESPONSE to a realistic prompt -- with the skill vs
without -- because law-coverage != disciplined behavior (an agent scored 8/8 law-coverage
yet committed several of these mistakes in one session).
Harness-agnostic adapters (SkillOpt adapter interface) make it work for BOTH seats:
transcript (offline CI), claude (ANTHROPIC_API_KEY -> Claude Code/acer), codex
(OPENAI_API_KEY -> Codex/liris), echo. 8 held-out scenarios as prompt+rubric; offline
self-test discriminates good (8/8) vs bad (0/8). Honest: rubric is keyword-based (v3 =
LLM judge), live adapters need keys, not the full optimizer loop yet (this is the gate).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NqTnTpYZVqcAhWeqWjhoh
