chore: release v3.11.0 - #12
Merged
Merged
Conversation
Version bumped in all four declarations (plugin.json, marketplace.json, .claude-plugin/marketplace.json, pyproject.toml) — the catalog file is the one that drives installs, and the install path is version-scoped, so a missed bump there silently resolves updates to a stale cache slot. No skill or pipeline behaviour changes; the deliberation path is identical to 3.10.0. This release carries the benchmark-honesty work plus one user-facing correction: The marketplace catalog Claude Code shows on install still advertised "SID confidence scoring" and the "CortexDebate Trust Score" as headline features and claimed "3-round debates capture >95% of quality improvements". All three describe mechanisms this project retired: v3.8 demoted self-reported confidence to display-only and deleted the dynamic-rounds machinery as a placebo knob; v3.10 replaced the CRIS/CortexDebate rubric with trust derived mechanically from the verified-citation rate. Descriptions and the feature list now name what actually drives decisions — debate gate on claim agreement, citation verifier, mechanical trust, lossless claim ledger with mandatory dissent, anonymised review, and the opt-in execution arbiter. Unreleased/Planned now records the two things that must happen before any live benchmark number is trustworthy: validating the judgment-task set (separability + judge-human agreement) and fixing the LiveRunner topology mismatch, where Synod's in-session Claude is stood in for by a hardcoded API call and Phase 1's three solvers become two. 975 passed / 14 skipped; version-consistency guard green; ruff clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
1 participant
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.
What this is
Version bump to 3.11.0 across all four declarations, plus one user-facing
correction. No skill or pipeline behaviour changes — the deliberation path
is identical to 3.10.0.
.claude-plugin/marketplace.jsonplugin.jsonmarketplace.jsonpyproject.tomlMarketplace copy corrected
The description Claude Code shows on install advertised three retired mechanisms
as headline features:
T = 0.25 + 1.75 × verified-citation rateDescriptions and the feature list now name what actually drives decisions: the
debate gate on claim agreement, the citation verifier, mechanical trust, the
lossless claim ledger with mandatory Dissent, anonymised review, and the opt-in
execution arbiter.
Planned, recorded honestly
Two items now block any live benchmark number being trustworthy, and are written
into
Unreleased:judge-vs-human agreement. Standard from arXiv:2408.08808;
externally-validated candidates for a primary arm are CODAL-Bench /
CodeUltraFeedback,
CodeJudgeBench, and
Arena-Hard-Auto. Licence terms unchecked.
model (
synod-phase1-solver.mdwritesCLAUDE_SOLVER_RESPONSEfrom thesession), but LiveRunner calls the Anthropic API with a hardcoded
claude-sonnet-5/ 1024-token config and runs 2 solvers where Phase 1 has 3.So live numbers would not measure Synod as shipped, and
ANTHROPIC_API_KEYis a harness artifact rather than a Synod requirement.
Verification
tests/test_version_consistency.pygreen — all four declarations agree3.10.0references outside the CHANGELOG history975 passed / 14 skipped;ruff check+ruff format --checkclean🤖 Generated with Claude Code