Content strategy

The governing question is: what earns this exact candidate the right to move forward, and what observations justify that decision? Navigation follows problem → mental model → workflow → failure/recovery → evidence → source.

Readers and routes

Technical reviewers begin with the overview and system tour, then follow one release and read the recorded case. Learners read dependency boundaries and health behavior before the Redis, PostgreSQL and signer workflows. Operators begin with credential-free checks, then local development, lab prerequisites and troubleshooting. All three paths converge on source and evidence rather than tool catalogues.

Page patterns

Explanations establish the consequence of a problem, the system boundary and chosen mechanism before describing normal behavior. They then examine failure, useful observations, recovery and trade-offs. Flagships use continuous prose supported by diagrams and focused rule tables. References preserve exact endpoint names, units, thresholds, schema fields and default values. Operator pages place prerequisites, costs and mutation scope before commands, followed by expected observations, stop conditions and cleanup. Evidence cases preserve candidate, time window, observation, interpretation and public/private boundary.

The site avoids one-paragraph topic stubs, repeated tool marketing and unsubstantiated production claims. Related reading is part of the article footer and source-relative links in the canonical prose. A title is generated from the registry; an initial source H1 is removed with its anchor preserved for Markdown.

Claim vocabulary

Implemented describes current source behavior. Tested describes the scenario asserted by a named test. Observed describes a named retained run. Historical describes an earlier candidate/window. Illustrative describes a diagram or explanation rather than measured behavior. Not-collected describes absence. Separately, outcomes may be pass, fail, blocked or unavailable. A screenshot receives no scorecard verdict inferred from its appearance.

Keep application source, chart source, rendered revision, image digest, Freight, signer identity, runner identity and run identity separate. Always say production-like testnet when introducing prod. Manual Freight approval is an override, not an upstream verification receipt. Publication-time Cosign checks do not establish cluster admission enforcement.

Citation and maintenance

Canonical Markdown uses logical repository-relative links. Website prose and diagrams are maintained in the private devSatym/resilience-gate-docs repository; implementation and reviewed evidence come from the ignored public upstream checkout pinned to 2420ff4c2a88b1fa4a19e413cb0752e271480081. Preparation converts mapped document links to site routes and unmapped implementation links to reviewed upstream commit permalinks. Authoring links identify the documentation repository separately and require access to its private source.

The publication inventory records documentation HEAD and dirty state, upstream repository/revision, website or upstream input origin, and a digest of selected public inputs. Repository transfer and website publication are separate events. Their recorded states must not be inferred from a configured URL, a local build, or an earlier monorepo report.

Hosting guidance follows the pinned handbook’s Git-integrated Cloudflare Pages Free procedure at the root base. Native Git builds and previews change delivery of the static website, while platform claims and historical evidence retain their identities. The runbook distinguishes assigned Pages origin, local validation, Git/API grants, native deployment, the owner CNAME handoff, and custom-origin HTTPS verification. Earlier monorepo, repository-separation, and Workers-campaign receipts remain historical; new Pages outcomes use docs/website/audit/cloudflare-pages/ and their actual counts.

A source change should trigger review of the page’s mechanism, examples, diagrams and limitation statements. Historical records keep their original dates. A website rebuild never upgrades a historical observation into current verification. Review source tests and actual code where earlier prose differs; record the discrepancy without changing platform behavior.

Editorial decisions

Source review found broader prose claims around absent metrics, facilitator malformed responses, settlement-duration metrics and cleanup absence checks. The relevant pages document exact implementation and test coverage. Rationale inferred from current source is labeled analysis; the website does not fabricate original ADRs or incident history.

See information architecture and coverage.