Releases: DataScience-EngineeringExperts/conclave
Release list
v1.2.0 — Elite Decision Protocol
Conclave v1.2.0
This release adds the quality-first Elite Decision Protocol, constrained-choice vote mode,
true every-mode manifest enforcement, and the guarded paid-exploratory evaluation runner.
Highlights:
- Elite: independent answers, claim audits, revisions, synthesis, and structured verdict.
- Fixed three-success gates, phased execution receipts, and decision-readiness tri-state.
- Capped live evaluation: dry-run default, exact USD 10 approval, authenticated checkpoints,
bounded estimates, one in-flight call, and no-repeat resume. - OpenAI strict verdict extraction compatibility with recursive replay-contract validation.
- Package, security, OIDC, and Sigstore identities migrated to
DataScience-EngineeringExperts/conclave.
The live smoke establishes operational correctness only; it does not establish decision-quality
or efficiency superiority. Install with pip install conclave-cli; the command and import remain
conclave.
See CHANGELOG.md for the full release history.
v1.1.0 — The Auditable Council
The auditable council. Every run can now produce a structured, agreement-scored, fully auditable verdict plus a redacted execution manifest — on top of the existing synthesize / raw / debate / adversarial modes.
conclave ask "<decision>" now returns a verdict: a headline + recommendation, a consensus score computed by reproducible arithmetic over the models' own clustering (never an LLM-emitted number), conflicts citing which providers disagreed (with evidence), and minority reports — alongside a manifest recording which providers were called and how the verdict was made. Streaming and --json carry it too.
Highlights
- CouncilResult v2 —
verdict,consensus_score/method/label,conflicts,provider_votes,minority_reports,manifest(all additive, backward-compatible). - ModelHarnessManifest with verdict-extraction provenance — no secrets, ever.
- Deterministic consensus (
position_cluster_ratio_v1) — the auditability fix: the number is arithmetic over the model's clustering, not an LLM figure. - Native structured output across OpenAI / Anthropic / Gemini, threaded through
call_model. - Static provider capability catalog; CLI verdict rendering +
--jsonmanifest; streaming verdict parity.
The verdict is optional — open-ended prompts, single-member councils, and extraction failures degrade gracefully to synthesis-only. Default-on; opt out with Council(extract_verdict=False).
Full notes in CHANGELOG.md. Install: pip install -U conclave-cli
v1.0.0
First stable release. conclave is a bring-your-own-keys multi-model council: it fans one prompt to N foundation models concurrently and merges their answers. 4 modes (synthesize / raw / debate / adversarial), 9 providers, streaming, optional result cache, debate convergence early-stop.
Install
pip install conclave-cli
The PyPI distribution is conclave-cli; the import package, CLI command, and repo all stay conclave:
from conclave import Councilconclave ask "your prompt" --mode synthesize
Highlights
- Distribution + release engineering — published via OIDC Trusted Publishing with Sigstore keyless signing and PEP 740 attestations (no stored token); hash-pinned lockfile;
RELEASING.mdrunbook. - Key-leak hardening — cause-chain fix so a key-bearing
httpxtransport exception can't surface through a traceback; transport-logging guard default-on (drops the httpx/httpcore DEBUG records that emit the auth header);SECURITY.mdBYO-keys threat model; dedicatedtests/test_keyleak_audit.pyregression suite + gitleaks. - Synthesizer behavior — observable degradation (an unkeyed/failed synthesizer always surfaces on
synthesis_error, never silent); versioned synthesis prompt stamped onto everyCouncilResult.prompt_version; documented selection precedence. - Supply-chain CI — fail-closed
pip-auditjob.
Full notes in CHANGELOG.md.