Confidence & evidence
Every reconciled release carries a confidence score (0–1) and a tier. This guide explains exactly how it’s computed and how to raise it.
The tiers
Section titled “The tiers”| Tier | Score | Reading |
|---|---|---|
| High | ≥ 0.80 | Well-supported — deterministic evidence, artifacts and deployments present, no red flags. |
| Medium | ≥ 0.60 | Reasonable, but something is missing or heuristic. |
| Low | < 0.60 | Thin or contested evidence — review before trusting. |
The score and tier appear on the release, the Release 360, and as a colored chip in lists.
How the score is built
Section titled “How the score is built”The score is a sum of weighted positive components minus deductions, clamped to 0–1:
Positive components
| Component | Weight | Earned when |
|---|---|---|
tag_anchor |
+0.20 | The release has a git tag or tag evidence. |
all_prs_linked |
+0.15 | Every included PR carries an issue key. |
artifact_present |
+0.20 | At least one artifact was published. |
deployment_present |
+0.20 | At least one deployment occurred. |
no_orphans |
+0.10 | No orphaned changes. |
no_force_push |
+0.10 | No force-push invalidation. |
builds_green |
+0.05 | All builds passed. |
Deductions
| Component | Weight | Applied when |
|---|---|---|
missing_artifact |
−0.20 | No artifact, and a delivery connector exists. |
missing_deployment |
−0.15 | No deployment, and a delivery connector exists. |
orphans_exist |
−0.10 | Orphaned changes present. |
force_push_detected |
−0.10 | A force-push invalidated commits. |
heuristic_ratio_high |
−0.15 | > 30% of included items were heuristic. |
rollback_detected |
−0.20 | A rollback occurred in-window. |
The full breakdown is on the Release 360 (confidence.components).
Coverage gap — no false penalties
Section titled “Coverage gap — no false penalties”If your org has no connector capable of producing artifact or deployment evidence, Reloqui does not deduct for their absence. Instead it flags a coverage gap (counts.coverageGap = true) with zero-weight components — so a team that only wires up SCM isn’t punished for evidence it never had a way to produce.
Reading exceptions
Section titled “Reading exceptions”Confidence and exceptions go together. counts.exceptions sums orphans + uncertain items + force-push + rollbacks. Policy can gate a publish on these — see Governance.
How to raise confidence
Section titled “How to raise confidence”- Connect a delivery source — artifacts + deployments are worth up to +0.40 combined.
- Link PRs to issues — earns
all_prs_linked. - Tag your releases — earns
tag_anchor. - Resolve orphans — attribute them (repo/branch mappings) or curate manually.
- Reduce heuristic reliance — precise repo URLs and issue keys keep attribution deterministic.
Calibration & thresholds
Section titled “Calibration & thresholds”The scoring weights are a calibrated, versioned model (configVersion on each score). Your policy thresholds (min confidence to pass, the approval floor) are configurable per project. Combine a high floor with required approvals for a strict gate.