Back to BotLane home
Trust · Release Standard

The 12-Gate Release Standard

A working demo is not a production release. This standard sets out twelve gates, grouped in four phases, that a BotLane release has to clear. Each gate says what it checks, what evidence it expects, and when it rejects a release.

Last updated October 15, 2026

Evidence, not assertions

Every gate is judged on evidence tied to the exact repository, commit, artifact, environment and policy version being released.

Missing means blocked

Missing, stale or failed evidence never quietly becomes a pass. A gate that doesn't apply needs a reviewed equivalent or an approved exception.

People authorize production

Approving a release's quality and authorizing its deployment are separate human decisions. Irreversible steps always need a person's sign-off.

  1. 01 / 07

    A working demo isn't a release

    This update works in the demo. Before it can reach production, it has to clear twelve gates, in four phases, in order.

  2. 02 / 07

    Source: where it comes from

    Each gate asks for evidence, not assurances: a verified commit and dependencies, a build that installs the same way every time, a locked-down runtime.

  3. 03 / 07

    Security: what ships, what it sees

    The artifact is traced to its source and scanned, every setting is validated, and credentials are checked for exposure, without ever being revealed.

  4. 04 / 07

    Recovery: staying up, coming back

    Health checks must notice when something breaks. Backups are restored, not just taken. And there's a tested way back to the last good version.

  5. 05 / 07

    Missing means blocked

    The restore drill hasn't been run, so gate 08 blocks the release. Missing evidence never becomes a pass. It waits until the drill exists for this commit.

  6. 06 / 07

    Launch: hardened, accepted, owned

    Permissions and exposure are reviewed, tests prove no required sign-off can be bypassed, and runbooks name an owner who answers when something breaks.

  7. 07 / 07

    People authorize production

    Every gate holds evidence, and production still waits. Approving quality and authorizing deployment are two separate decisions, made by people.

Reference

Full gate reference

Open any gate for the evidence it expects and what gets a release rejected.

Phase I · Source
01Provenance

Confirm the source, ownership, version and dependencies of what is being released.

Evidence expected
  • The exact commit, with its signature verified
  • A pinned dependency lockfile and its hash
  • An archived software bill of materials (SBOM)
  • Who wrote and who reviewed the change
Rejected if

The source can't be verified, dependencies float outside the approved policy, or the SBOM is missing.

02Repeatable Installation

Show that it installs the same way every time, from documented steps.

Evidence expected
  • Logs from a cold build in a clean environment
  • Toolchain and environment versions
  • Build output checksums across runs
  • Install steps and their exit codes
Rejected if

A clean install fails or the build isn't reproducible. Any platform-specific difference is recorded and justified, never papered over.

03Containerization

Package the release into a controlled, portable runtime with minimal privileges.

Evidence expected
  • The image or bundle identifier
  • An inspection of what the runtime contains
  • Runtime security settings: unprivileged, read-only where possible
Rejected if

The runtime is unidentified, over-privileged, or not the one that was reviewed.

Platform note: Some services don't run in containers. For those, the gate maps to a reviewed equivalent for that runtime's isolation and bundling. It is never reported as a container check that didn't happen.
Phase II · Security
04OCI Artifacts

Produce versioned, traceable deployment artifacts that can be verified.

Evidence expected
  • The artifact's immutable digest or version id
  • A vulnerability scan with no open critical or high findings
  • A signature and provenance attestation that verify
Rejected if

The artifact can't be traced to its source, fails verification, or carries unresolved critical findings.

Platform note: The original gate is written for OCI container images. Where a service ships as a different kind of bundle, that bundle and its deployment version are verified instead.
05Configuration Validation

Validate every setting and reject unsafe or incomplete configuration before it runs.

Evidence expected
  • A schema validation report for each target environment, with private values redacted
  • The version of the schema used
  • Tests showing bad configuration is refused
Rejected if

Required settings are missing or malformed, unsafe defaults apply, or unknown settings skip validation.

06Secrets Handling

Store, access and rotate credentials without ever exposing them.

Evidence expected
  • A secrets scan of code, history and build output
  • A review of who and what can read each secret
  • How secrets are injected at runtime, redacted
  • Rotation records
Rejected if

A credential appears in plain text anywhere, access is broader than needed, or rotation is overdue.

Platform note: Evidence describes how secrets are handled. It never contains a secret value.
Phase III · Recovery
07Health Checks

Verify that the application and the services it depends on are healthy.

Evidence expected
  • Health signals suited to the runtime
  • Synthetic checks that exercise real paths
  • Error and latency alerts, and proof they fire
Rejected if

There is no reliable way to tell the service is unhealthy, or alerts aren't verified.

Platform note: Where the runtime has no liveness probes or restart policies, the gate uses synthetic checks and alerting designed for that platform.
08Backup & Restore

Prove the release can be recovered, not just that a backup exists.

Evidence expected
  • The backup policy and where backups go
  • Logs from a restore drill
  • Integrity checks on restored data
  • Measured recovery point and recovery time
Rejected if

No restore has been tested, or restored data fails integrity checks.

Platform note: What gets backed up depends on what holds state. For stateless services, the gate covers configuration, assets and deployment history.
09Upgrade & Rollback

Upgrade safely and return to a known-good version when something goes wrong.

Evidence expected
  • The last known-good version
  • A rollback rehearsal and its result
  • Monitoring during rollout
  • Compatibility results for any data migration
Rejected if

There is no tested way back, or a migration can't be reversed safely.

Platform note: Rollback uses the deployment and versioning tools of the platform the service runs on.
Phase IV · Launch
10Security Hardening

Review permissions, network exposure, dependencies and transport security.

Evidence expected
  • A least-privilege review of every permission
  • Dependency findings and how each was handled
  • TLS and security header configuration
  • Application security tests and an exposure review
Rejected if

Permissions exceed what the release needs, exposure is unreviewed, or known findings are unaddressed.

Platform note: Controls are chosen for the stack in use. A control is only listed when it is actually deployed.
11Acceptance Testing

Verify functional, integration, safety and human sign-off criteria.

Evidence expected
  • The test suite revision and its results
  • The agreed acceptance scenarios
  • Tests confirming that any step set to need human sign-off can't proceed without it
  • Results for failure and fallback paths
Rejected if

A required sign-off can be bypassed, or a critical end-to-end scenario fails.

12Documentation & Runbooks

Set out procedures, ownership, monitoring, incident response and recovery.

Evidence expected
  • Reviewed runbooks
  • A named owner and on-call escalation path
  • A recovery drill record
  • Audit trail retention settings
Rejected if

There is no accountable owner, no recovery process, or the documents contradict how the system actually behaves.

Gate outcomes

Every gate records one of six outcomes.

Outcomes belong to one release: one commit, one artifact, one environment. A result from an earlier version never carries over.

Pending
Evidence hasn't been collected yet.
Passed
Current evidence for this exact version meets the gate.
Failed
Evidence shows the gate isn't met.
Blocked
Evidence is missing or stale. This never becomes a pass.
Equivalent verified
The gate was met through a documented, reviewed equivalent for this runtime.
Exception approved
A time-limited, reviewed exception. It can't waive a required human sign-off.

What this page is. It describes the standard and the evidence each gate expects. It is not a report on any particular release, and nothing on it should be read as a pass result. Any illustration elsewhere on this site that shows gate results is labelled as illustrative.