Skip to content

release(v8.3.1): ultracode-superiority Wave A — multi-oracle / committed seeded fuzz / algorithm-differential - #88

Merged
aimsise merged 1 commit into
mainfrom
release/v8.3.1
Jun 3, 2026
Merged

release(v8.3.1): ultracode-superiority Wave A — multi-oracle / committed seeded fuzz / algorithm-differential#88
aimsise merged 1 commit into
mainfrom
release/v8.3.1

Conversation

@aimsise

@aimsise aimsise commented Jun 3, 2026

Copy link
Copy Markdown
Owner

TL;DR

A fail-open, criticality-gated refinement of v8.3.0's M1 (Gate 8 + the evidence_floor ladder) and M5 (criticality floor) — the Wave A verification-assurance pass that closes the main residual a 2026-06-02 A/B found a human-directed max-effort ("ultracode") build still held over the autonomous harness. It adds no new gate; it sharpens what the EXISTING thorough / exhaustive evidence_floor demands for a standard-backed computational AC. v8.4.0 remains reserved for M3 + M4.

Highlights

Item Change
H1 multi-oracle At thorough/exhaustive, a standard-backed computational AC's oracle evidence MUST be ≥2 mutually-validated oracles (≥1 first-principles), trusted only on agreement within tolerance.
H2 committed seeded fuzz test-authoring-guidance rule 7 promoted "encouraged" → depth-gated MUST (committed fixed-seed PRNG, tier-scaled).
H3 algorithm-differential EC-DIFFERENTIAL re-specced to algorithm-vs-algorithm where a 2nd algorithm exists (membership is necessary-not-sufficient).
H13 oracle harness new skills/impl/references/independent-oracle-harness.md (first-principles block + independent-library block + seeded PRNG + 2nd-algorithm differential), wired into producers + taxonomy + verifier.
H12 criticality tighten verification-depth.md criticality cue set widened (color-science cues + a shared-core input-boundary trigger) so the targeted AC class reliably escalates.

Fail-open (load-bearing safety property)

Every new requirement engages only at the thorough/exhaustive evidence_floor (already M5-criticality-gated) and only where a published spec / 2nd impl / 2nd algorithm exists; otherwise it degrades to the single natural channel + a Caveat — never a block. A routine S/M-conservative ticket is byte-identical to v8.3.0; the standard floor and every constraints.<x>: off path are unchanged. agents/ac-evaluator.mdagents/ac-evaluator-hi.md stay byte-identical (CT-EV-MODEL-1).

Verification

  • bash tests/test-skill-contracts.sh 797/797, bash tests/test-path-consistency.sh 142/142, full sweep 34/34.
  • New CT-EV-10..14 all pass; each signature token (mutually-validated, first-principles, fixed-seed, algorithm-vs-algorithm, independent-oracle-harness.md, shared-core) is HEAD=0 — a git stash flips the assert to FAIL, git stash pop restores PASS. The symmetry guards cover the full 9-file author→verify surface incl. the planner author and ticket-evaluator grader cells.
  • Every v8.3.0 contract intact (CT-AR-1/4/8/10/12/13/14, DEPTH-8, CT-EV-1..9, CT-EV-MODEL-1..4); Gates 1-8 carriers unchanged (no new gate). PII/English scan clean.

See CHANGELOG.md [8.3.1] for the full entry.

🤖 Generated with Claude Code

…ted seeded fuzz / algorithm-differential

Fail-open, criticality-gated refinement of v8.3.0's M1 (Gate 8 + evidence_floor) and M5 (criticality floor) — the Wave A verification-assurance pass closing the main residual a 2026-06-02 A/B found a human-directed max-effort build held over the harness. Adds NO new gate; sharpens what the EXISTING thorough/exhaustive evidence_floor demands for a standard-backed computational AC.

H1: oracle evidence must be >=2 mutually-validated oracles (>=1 first-principles). H2: committed fixed-seed property-fuzz promoted to a depth-gated MUST (test-authoring rule 7). H3: EC-DIFFERENTIAL re-specced to algorithm-vs-algorithm where a 2nd algorithm exists (membership necessary-not-sufficient). H13: new skills/impl/references/independent-oracle-harness.md (4-part gold standard), wired into producers + taxonomy + verifier. H12: criticality cue set widened (color-science cues + a shared-core trigger) so the targeted AC class reliably escalates.

Every requirement engages ONLY at thorough/exhaustive and ONLY where a published spec / 2nd impl / 2nd algorithm exists; else degrades to the single natural channel + a Caveat (never a block). A routine S/M-conservative ticket is byte-identical to v8.3.0; the standard floor and every constraints.<x>: off path are unchanged; ac-evaluator and ac-evaluator-hi stay byte-identical (CT-EV-MODEL-1). New CT-EV-10..14 (revert-flip, HEAD=0). Tests: skill-contracts 797/797, path-consistency 142/142, full sweep 34/34. v8.4.0 remains reserved for M3+M4.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@aimsise
aimsise merged commit 52b67a1 into main Jun 3, 2026
2 checks passed
@aimsise
aimsise deleted the release/v8.3.1 branch June 3, 2026 17:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant