feat(intelligence): 🧾 prove Empathy publication boundaries - #95
Conversation
Refs #94 Refs egohygiene/relay#33 Roadmap-Step: EMP-Q05
Preserve the producer-owned snapshot while validating the consumer upgrade. Refs #94 Refs egohygiene/relay#33
Exercise the existing Mindgarden integration assertions with every consumer contract run. Refs #94 Refs egohygiene/relay#33
Apply the canonical MegaLinter Python policy in focused consumer checks, retain explicit unittest and public-safe CLI boundaries, and reject encoded DTD declarations in historical rollback verification. Refs #94 Refs egohygiene/relay#33
|
Validation completed against exact head
Fresh exact-head provider evidence (all attempt 1):
Existing baseline defects remain explicit: EgoLint is 30/31 because its catalog expects MegaLinter Identity output was generated only in an isolated test checkout and kept out of Repository Intelligence composition inputs. These checks do not claim post-merge deployment, live-route, intentional provider failure, rollback deployment, or restored-current acceptance. Local log SHA-256 digests:
Independent source review found no blocker in the publication trust boundaries. Only Empathy's separate deploy job receives Pages/OIDC write permissions; the build, receipt, live verification, standalone, and safety-fixture jobs use read permissions. PR lanes retain review evidence and cannot enter the snapshot publishers or privileged producer-refresh path. Producer guards require an allowlisted workflow/event, both repository identities, and the default branch while deliberately preserving useful non-success conclusions. Remote actions remain commit-pinned, the Quartz pin and reviewed-public Mindgarden projection remain unchanged, and Relay is constrained to the intelligence subtree. Receipt validation, tamper rejection, failed-deployment semantics, exact live-byte mismatch detection, safe diagnostics, duplicate-key rejection, and encoded XML declaration rejection are exercised by fixtures. The provider scheduling fixture exercises a real skipped ancestor and observes the copied publication guard outcomes. Cancellation and fork-origin cases are predicate/fixture coverage; no actual cancelled deployment or fork dispatch experiment is claimed. Operational deployment receipts and live publication proof remain post-merge acceptance. |
|
Reviewed branch GitHub associates every run below with that exact PR head. The generated artifacts represent synthetic PR merge
For both real Pages build lanes, Configure Pages and Upload Pages were skipped. Deploy job 107147582896, receipt finalization 107147583487, and live verification 107147584127 were skipped. The run has six review/provenance/report artifacts and no Eight Pages/standalone ZIP archives were downloaded and independently rehashed here. The safety reviewer independently downloaded and rehashed the two small checks/scheduler archives. All ten match GitHub's advertised digests. Extraction rejected traversal, absolute paths, duplicate entries, and symlinks. The complete current public artifact exactly matches the site retained with provenance. The standalone's 19 public files (18 payload files plus manifest) exactly match the composed site's RI subtree. The pinned Relay implementation independently revalidated the manifest, payload inventory, routes, source identity, and full composition. All 63 captured Quartz files remain byte-identical after RI composition; every baseline file hash was checked. Run evidence and receipts stay outside the public tree; no raw activity workspace or invented
The exact source Git blobs independently match the profile/configuration and normalized-summary digests recorded in Rollback independently validates old source
The rebuilt composition preserves all 81 historical paths. Only synthetic tag dates vary in the two normalized XML feeds; the other 79 files match historical bytes. The report records the new composed digest separately and makes no false whole-site equality claim. Old public provenance remains old; no current-generator manifest is invented. Current, rollback, and standalone diagnostic reports each remain below 8 KiB, contain only allowlisted identities/outcomes, and record all required stages as successful. They contain no private filesystem roots, raw producer payloads, job outputs, or credential-like values. PR reports explicitly record
|
Pull Request Summary by devActivityMetricsAchievements
|
Refs #94
Refs egohygiene/relay#33
Empathy needs reviewable evidence connecting its Repository Intelligence build to the complete Mindgarden/Quartz composition and eventual Pages deployment. This PR upgrades both current integrations to reviewed Relay
9a6315978766c336566b9fa7139b800fa8789ba5and adds the consumer publication checks while preserving Empathy as the sole Pages authority.075afd3f712da0088a07f5284a7b3aba37dd61b6, and existing normalized-report producers. Trusted default-branch producer refreshes retain useful non-success conclusions; PR origins cannot enter that privileged path.rollback-v1.4compositions on PRs and retain 30-day review artifacts. PRs have no Pages upload, deployment, receipt finalization, snapshot publication, or secrets path.agent/prove-repository-intelligence-949916165cd6e20ce1186a62b7d2f25333e33be1b5792b4a86e83a7cafb99342a4489285a3d7fe89973e4e05412ef8311fb0b0d61380255536a818186b09cefdb4a8c808d2dcceddb54ab6a27004eded2eThe represented merge and reviewed branch have the same tree. A normal merge incorporated the producer-owned report refresh that advanced main during implementation.
Final provider/artifact evidence records exact run attempts, all ten independently verified archive hashes, manifest/bundle/composition digests, report input identities, and rollback proof. Current and rollback Pages builds, standalone RI, 81 consumer tests, scheduler checks, MegaLinter, CodeQL, commit validation, dependency review, and lint-architecture generation passed on this head. Both PR Pages uploads and all deployment jobs were skipped.
Complete validation and trust-boundary review records 194 local root tests without skips; the full garden suite; architecture, lifecycle, continuity, roadmap/journal, and automation validators; canonical lint, syntax, duplicate-key, and diff checks. All 63 Quartz files remain unchanged in the 82-file current composition; standalone output matches its intelligence subtree.
Repository CI is not wholly green: automation reaches the pre-existing EgoLint catalog/workflow pin mismatch, and OSV finds the same 85 blocking findings in unchanged dependency inputs. The local complete-suite attempt also records existing stale task-catalog entries and the missing Cargo workspace fixture. Those unrelated defects remain outside this checkpoint.
Actual default-branch/manual/producer deployment, deployment receipts and live routes, dedicated RIW-002 failure retention, operational rollback, and immediate restored-current proof require merge and consumer-side execution. The documented sequence remains acceptance work on #94. Relay #33 and #106 remain open. Observatory-dependent views intentionally remain unavailable without a commit-matched read model. Checkout-directory portability is separately tracked by Relay #109; repeatability here is scoped to the declared CI context.
This PR is ready for review and has not been merged.