Release Gates

A report that is allowed to say no

Release Board → Release Gates (there is no separate menu item)

A release gate is only useful if it can fail. LocalizeShip evaluates four checks against the workspace and returns one of three verdicts, then writes evidence you can attach to a build.

Release gates

The four checks

1. Translations Approval Matrix

Walks every source revision × target locale:

Condition	Counter	Violation text
No approval or `Missing`	`missingCount`	“Entry ‘…’ missing translation for [DE].”
`Stale`	`staleCount`	“Entry ‘…’ in [DE] is Stale (source revision updated).”
Any other non-approved state	`unapprovedCount`	“Entry ‘…’ in [DE] is in state NeedsReview.”

Pass condition:

(policy.RequireAllLocalesApproved ? unapprovedCount == 0 : true) &&
(policy.BlockOnMissingTranslations ? missingCount == 0 : true) &&
(policy.BlockOnStaleTranslations ? staleCount == 0 : true)

Details line: total cells, unapproved, stale, missing.

2. Visual QA Matrix & Geometry

Walks every capture tile:

Coverage == Unknown or Registered → unvisited;
any unsuppressed finding with severity Error or Fatal → error finding.

Pass condition:

(policy.BlockOnUnvisitedCaptures ? unvisitedTiles == 0 : true) &&
(policy.BlockOnVisualErrors ? errorFindings == 0 : true)

A tile with Coverage == Failed is not counted as unvisited and may have no findings, so it can effectively pass this check. Track failed captures separately.

3. Voice Coverage & Revisions

Counts voice entries where IsStale is true (stale source or stale translation revision). Pass condition is BlockOnStaleVoiceClips ? staleVoices == 0 : true.

It does not count Missing or DurationMismatch.

4. Subtitle Timing & Reading Speed

For each cue:

HasInvalidTiming (start < 0 or end <= start) → violation;
otherwise CharactersPerSecond > workspace.Subtitles.MaxCharactersPerSecond → violation.

Pass condition is BlockOnSubtitleTimingErrors ? timingViolations == 0 : true.

It does not enforce min/max duration or characters per line.

Overall verdict

bool allPassed = approvalCheck.Passed && visualCheck.Passed &&
                 voiceCheck.Passed && subCheck.Passed;

int defectFailures = errorFindings + staleCount + staleVoices + timingViolations;
int incompleteFailures = unvisitedTiles + missingCount;

if (allPassed) report.OverallStatus = GateStatus.Passed;
else if (defectFailures > 0) report.OverallStatus = GateStatus.Failed;
else if (incompleteFailures > 0 || unapprovedCount > 0)
    report.OverallStatus = GateStatus.IncompleteCoverage;
else report.OverallStatus = GateStatus.Failed;

public enum GateStatus { Passed, Failed, IncompleteCoverage }

Verdict	Practical meaning
Passed	All four checks passed under the active policy
Failed	At least one defect exists (visual error, stale translation, stale voice, subtitle timing/CPS)
IncompleteCoverage	No defects, but coverage is incomplete (unvisited tiles or missing translations), or there are unapproved cells

A gate that says IncompleteCoverage is not a soft pass. It means “we have not looked at everything.”

GatePolicy

public class GatePolicy
{
    public bool RequireAllLocalesApproved = true;
    public bool BlockOnStaleTranslations = true;
    public bool BlockOnMissingTranslations = true;
    public bool BlockOnUnvisitedCaptures = true;
    public bool BlockOnVisualErrors = true;
    public bool BlockOnStaleVoiceClips = true;
    public bool BlockOnSubtitleTimingErrors = true;
    public bool BlockOnGlossaryViolations = true;
}

All default to true.

The Release Gate window calls Evaluate(workspace) without passing a policy, so every run uses the defaults. The booleans are configurable through the API, not through the window. BlockOnGlossaryViolations is not read by any check in 1.0.0.

Example of a custom policy from code:

var policy = new GatePolicy
{
    RequireAllLocalesApproved = true,
    BlockOnMissingTranslations = true,
    BlockOnStaleTranslations = true,
    BlockOnUnvisitedCaptures = true,
    BlockOnVisualErrors = true,
    BlockOnStaleVoiceClips = false,
    BlockOnSubtitleTimingErrors = true
};

var report = new ReleaseGateEvaluator().Evaluate(workspace, policy);

Reports

HTML

GenerateHtmlReport produces a self-contained dark-themed page with:

release version and UTC generation timestamp;
source locale;
overall verdict and passed/total check count;
one card per check with PASS/FAIL, details and an escaped violation list.

Violation text is escaped with SecurityElement.Escape. The HTML contains no external assets.

JSON

GenerateJsonReport calls UnityEngine.JsonUtility.ToJson(report, true). It serializes the ReleaseGateReport:

{
  "ReportId": "...",
  "ReleaseVersion": "1.0.0",
  "OverallStatus": 0,
  "TimestampTicks": 638965000000000000,
  "Checks": [
    {
      "CheckName": "Translations Approval Matrix",
      "Passed": false,
      "Details": "...",
      "Violations": ["..."]
    }
  ]
}

OverallStatus is serialized as an integer (Passed = 0, Failed = 1, IncompleteCoverage = 2). Parse it with that in mind.

File locations

Both export buttons write next to the project folder:

LocalizeShip_Report_v<ReleaseVersion>.html
LocalizeShip_Report_v<ReleaseVersion>.json

They reveal the file in the OS browser after writing.

What the gate does not check

placeholder or rich-text validity;
glossary violations;
merge conflicts or unreviewed imports;
voice duration mismatch or missing audio;
subtitle min/max duration or characters per line;
visual warnings (only errors and fatals);
player-build rendering;
linguistic or RTL correctness;
whether a provider table was actually written after approval.

If you need these as blockers, extend ReleaseGateEvaluator or run your own checks alongside it. The JSON report is easy to post-process.

CI reality

There is no shipped headless gate CLI. The evaluator is a plain C# class in the editor assembly; GenerateHtmlReport and GenerateJsonReport are public methods on ReleaseGateEvaluator, not private window helpers. You can:

run the 12 EditMode acceptance tests in batch mode with your own -runTests invocation;
call ReleaseGateEvaluator from a custom editor script executed with -executeMethod;
archive the JSON/HTML produced by that script.

Do not describe the included gate window as a CI tool. See CI Release Gates for an honest recipe.

Reading a report in a release meeting

Read the verdict.
Read PassedChecks / TotalChecks.
Read the Details line of each failed check before the violation list; it tells you the scale.
Separate defects (Failed) from coverage gaps (IncompleteCoverage).
Assign owners by category: translations, visual, voice, subtitles.
Re-run after fixes and archive both reports if the verdict changed.