Releases & reconciliation

The core of Reloqui. Guide: Create & reconcile a release.

Capabilities

• 11-stage reconciliation engine — scope resolution, evidence collection, clock-skew normalization, deterministic attribution, heuristic attribution (rule-tagged), classification, backport detection, force-push handling, confidence scoring, policy evaluation, snapshot write.
• Release state machine — draft → reconciling → ready | ready_with_exceptions | blocked_by_policy → approved → published, plus approval_suspended, superseded, rolled_back.
• Evidence classification — included, included_heuristic, excluded, uncertain, orphaned, each with a reason code.
• Confidence score & tier — weighted 0–1 score, high/medium/low tier, component breakdown, coverage-gap handling.
• Notable changes — breaking / security / deprecation / feature / fix highlighting.
• Backport & force-push handling — carried-forward changes tagged; rewritten commits invalidated and penalized.
• Manual curation — link/unlink evidence with required reasons; survives recompute.
• Baselines & compare — baseline selection (previous release / timestamp) and release-to-release diff with confidence delta.
• Dependency-change diffing — manifest diffs (package.json, requirements.txt, go.mod, pyproject.toml) between releases.
• Recompute modes — full reconcile, scoring-only recompute, rescore action.
• Release trains — coordinate multi-service delivery (guide).
• Rollback — record rollbacks with reason + confidence impact (guide).
• Release 360 — the full evidence graph, confidence, exceptions, incidents, dependency changes, revisions.
• Export package — evidence + notes + approvals + snapshot as JSON.