v0.3.0: Offline Research Certification Workbench

[zrack]

Outcome

Ships the governed v0.3.0 Offline Research Certification Workbench release slice under SAED 1.3 work packet GEX-ORC-001.

What changed

• adds versioned model profiles and experiment specs/manifests with enforced point-in-time cutoffs and semantic reproduction
• adds append-only, hash-chained local corpus registration with integrity, rights, redaction, split, outcome, and cost metadata
• adds batch day/expiry/DTE-layer position-model comparisons without summing OI and trade-volume sources
• adds deterministic model-property, provider-fault, and generated-chain performance gates
• separates provider readiness from runtime connection state
• replaces causal/dealer UI language with explicit GEX-proxy, compatibility-level, and wall-proximity semantics
• expands installed-wheel CI across the offline certification workflows
• bumps source/package version to 0.3.0 and updates architecture, ADR, operator/contributor docs, roadmap, changelog, and generated assets

Evidence

• compileall: passed
• unit suite: 242 tests passed
• model properties: 7/7 passed
• provider fault cases: 7/7 passed; live transport false
• default 500-contract generated performance gate: passed; live capacity false
• model evidence and 12-case offline Databento certification: passed
• sdist/wheel build and Twine checks: passed
• fresh Python 3.14 installed-wheel smoke outside checkout: passed for all new workflows
• experiment reproduction: semantic match
• corpus verification: passed

Evidence ceiling

This PR establishes offline software and research-process behavior only. It does not certify authentication, entitlements, live chain coverage, real reconnect/latency behavior, dealer inventory, predictive validity, execution quality, or profitability. Predictive validity remains unmeasured.

Recovery

Revert the merge commit. Generated corpus and experiment outputs remain local and operator-owned. No committed migration mutates user data.

Not included

No credentialed live run, Git tag, GitHub Release, PyPI publication, REST/MCP service, or production deployment.