Skip to content

Production readiness scorecard

Generated: 2026-08-29T13:38:03Z

Commit: c1b59bbaaa78f7fe9b4470b3bd5bb60f36b0d62d

How to read this

Every status below is derived, never asserted:

StatusMeaning
metEvery evidence path exists and the associated static gate passes.
gapEvidence is missing or the gate fails.
unverifiableThe criterion needs a deployed environment, real credentials, or a human judgement. CI cannot prove it either way, so it is excluded from the score and listed explicitly rather than counted as met.

Score is met / (met + gap). Overall: 100% (37 met, 0 gap, 4 unverifiable).

Summary

DimensionScoreMetGapUnverifiable
Availability100%500
Latency & performance100%501
Security & supply chain100%1000
Accessibility100%202
Recovery & resilience100%901
Cost & resource control100%600

Availability

Can we tell when the platform is down, and does a bad deploy get caught before users do?

CriterionStatusVerificationNotes
Every user-facing endpoint has a declarative SLO.
avail-slo-catalog
metgo run ./cmd/slo-coverage-audit
SLOs are alerted on with multi-window multi-burn-rate rules, not single-window thresholds.
avail-burn-alerts
methelm template test helm/teslasync
An authenticated smoke gate runs against the deployment after every release.
avail-post-deploy-smoke
metgo run ./cmd/ops-gate -check smokegate "smoke" passes
Liveness, readiness, and preStop drain endpoints exist and behave correctly under shutdown.
avail-probe-contract
metgo test ./internal/ops/...
Rollout is staged with canary controls and an explicit pause switch.
avail-staged-rollout
metgo run ./cmd/ops-gate -check rolloutgate "rollout" passes

Latency & performance

Do we have measured budgets for backend and frontend, and do regressions block a release?

CriterionStatusVerificationNotes
API latency SLOs exist with explicit percentile budgets.
lat-api-budgets
metgo run ./cmd/slo-coverage-audit
Core Web Vitals (LCP/INP/CLS/FCP/TTFB) are collected from real users and have SLOs.
lat-web-vitals
metgo run ./cmd/slo-coverage-audit
Frontend performance regressions are caught in CI before merge.
lat-regression-gate
metgh workflow view perf.yml
Latency and Web Vitals thresholds are wired into the automated rollback policy.
lat-rollback-thresholds
metgo run ./cmd/ops-gate -check rollbackgate "rollback" passes
Repeatable capacity profiles exist for the load shapes that break this system.
lat-capacity-profiles
metgo run ./cmd/ops-gate -check capacitygate "capacity" passes
Each capacity profile has been executed against a representative environment and its results recorded.
lat-capacity-executed
unverifiablegh workflow run capacity-test.yml -f profile=<id> -f confirm=RUNneeds a deployed environment or real credentials; CI cannot prove this either way

Security & supply chain

Can a consumer prove what is in a release, and can an attacker move a tag under us?

CriterionStatusVerificationNotes
SAST, container scanning, secret scanning, and Go vulnerability scanning run in CI.
sec-scanning
metgh workflow view security.yml
Security workflow actions and scanner images are SHA/digest pinned.
sec-immutable-actions
metgo run ./scripts/check-security-workflow-pins.go
Release workflow actions and images are SHA/digest pinned too.
sec-release-immutability
metgo run ./cmd/ops-gate -check supply-chaingate "supply-chain" passes
Every published image ships a signature, an SBOM attestation, and SLSA build provenance.
sec-sbom-signing-provenance
metgo run ./cmd/ops-gate -check supply-chaingate "supply-chain" passes
Each release publishes its vulnerability status, and a fixable CRITICAL blocks every public release side effect rather than only the notes.
sec-vuln-status
metgo run ./cmd/ops-gate -check workflowsgate "workflows" passes
No credential values live in the repository; ops manifests reference env var names only.
sec-no-repo-secrets
metgo run ./cmd/ops-gate -check smokegate "smoke" passes
No workflow interpolates an untrusted input into a shell script that holds secrets.
sec-workflow-injection
metgo run ./cmd/ops-gate -check workflowsgate "workflows" passes
The pod-fatal preStop drain endpoint is not reachable through any Service or Ingress.
sec-drain-plane-isolated
methelm template test helm/teslasync | go run ./cmd/ops-gate -verify-helm-render -gate "rollout" passes
The Helm chart ships no static database or Grafana password and rejects known weak overrides.
sec-generated-chart-credentials
metgo run ./cmd/ops-gate -check helm-secretsgate "helm-secrets" passes
Workloads can consume one Secret materialized from Vault or a cloud secret manager without credentials in Helm values.
sec-external-secret-sources
metgo run ./cmd/ops-gate -check helm-secretsgate "helm-secrets" passes

Accessibility

Is the UI usable with a keyboard and a screen reader, and is that enforced?

CriterionStatusVerificationNotes
Documented accessibility guidelines exist for contributors.
a11y-guidelines
unverifiablemanual reviewthe artifact exists, but the assessment is a human judgement; CI cannot score it
Automated axe + keyboard checks run in CI on every PR touching the frontend.
a11y-automated-gate
metgh workflow view frontend-quality.yml
Data visualisations carry accessible descriptions rather than being image-only.
a11y-chart-annotations
metnode web/scripts/audit-chart-a11y.mjs
A recorded accessibility audit exists with tracked findings.
a11y-audit-record
unverifiablemanual reviewthe artifact exists, but the assessment is a human judgement; CI cannot score it

Recovery & resilience

When something breaks, do we know what to do, and have we proved the backups work?

CriterionStatusVerificationNotes
Every external dependency has a degraded-mode runbook.
rec-degraded-runbooks
metgo run ./cmd/ops-gate -check runbooksgate "runbooks" passes
Every new migration is reviewed for forward compatibility, rollback, duration, and lock risk.
rec-migration-review
metgo run ./cmd/ops-gate -check migrationsgate "migrations" passes
A measurable rollback policy exists with an ordered, executable remediation plan.
rec-rollback-policy
metgo run ./cmd/ops-gate -check rollbackgate "rollback" passes
Backup artifacts are verified automatically, not assumed.
rec-backup-verification
metgo test ./cmd/backup-verify/...
Every SQL fixture the drill depends on matches the live schema and is executed in CI, not merely checked for existence.
rec-fixtures-executable
metgo run ./cmd/ops-gate -check fixturesgate "fixtures" passes
A scheduled restore drill is defined and wired to a workflow.
rec-restore-drill-defined
metgo run ./cmd/ops-gate -check restoregate "restore" passes
Recovery has finite RTO/RPO targets, evidence semantics, and explicitly assigned incident roles.
rec-objectives-and-ownership
metgo run ./cmd/ops-gate -check restoregate "restore" passes
A production-artifact restore drill has succeeded and its RTO is recorded.
rec-restore-drill-executed
unverifiablegh workflow run backup-restore-drill.yml -f mode=production-artifactneeds a deployed environment or real credentials; CI cannot prove this either way
Shutdown drains in-flight work instead of dropping it, the grace period can hold the whole budget, and both are covered by tests.
rec-graceful-shutdown
metgo test ./internal/ops/... ./internal/app/...
Config cannot drift between Go, Compose, and Helm.
rec-config-parity
metgo run ./cmd/ops-gate -check config-paritygate "config-parity" passes

Cost & resource control

Can this deployment run away with someone's money or disk?

CriterionStatusVerificationNotes
Every workload declares CPU/memory requests and limits.
cost-resource-limits
methelm template test helm/teslasync
Autoscaling has explicit min/max bounds rather than unbounded growth.
cost-autoscaling-bounds
methelm template test helm/teslasync
The default signal_log retention is finite and wired to an executable cleanup schedule.
cost-data-retention
metgo run ./cmd/ops-gate -check retentiongate "retention" passes
Fleet API calls reserve against a shared daily spend ceiling while preserving command capacity.
cost-fleet-api-budget
metgo run ./cmd/ops-gate -check fleet-api-budgetgate "fleet-api-budget" passes
AI provider spend is rate-limited and budgeted per-account.
cost-ai-spend-limits
metgo test ./internal/ai/limit/...
Capacity tests have hard duration ceilings and cannot be pointed at production.
cost-capacity-ceilings
metgo run ./cmd/ops-gate -check capacitygate "capacity" passes

Open gaps

None.

Not machine-verifiable

These are not claimed as done by CI. Each needs a real environment, a real drill, or a human assessment:

  • lat-capacity-executed (Latency & performance) — Each capacity profile has been executed against a representative environment and its results recorded. — run: gh workflow run capacity-test.yml -f profile=<id> -f confirm=RUN
  • a11y-guidelines (Accessibility) — Documented accessibility guidelines exist for contributors. — run: manual review
  • a11y-audit-record (Accessibility) — A recorded accessibility audit exists with tracked findings. — run: manual review
  • rec-restore-drill-executed (Recovery & resilience) — A production-artifact restore drill has succeeded and its RTO is recorded. — run: gh workflow run backup-restore-drill.yml -f mode=production-artifact

MIT License · Self-hosted Tesla intelligence