Engineering and verification

The repository tracks subsystem tasks, named gates, generated catalogs and registered automated checks.

The gate suite

One runner executes the checks in order and prints a single pass and fail summary. Each check is tagged: offline for deterministic runs with no network, network for checks that need the model proxy and are reported as skipped rather than failed when it is unreachable, and gated for paths that are intentionally blocked until a sanctioned route is enabled.

The exit code is non-zero only when a runnable check actually fails. A skipped network check is reported as skipped, which is a different sentence from a passing one.

What a gate proves

A gate asserts one property of the code, on the host that ran it, at the moment it ran. The seat contract check runs each measured seat's own binary and reads its top-level help and its subcommand help, because a probe that reads only the top level cries wolf and misses flags that exist one level down. It never runs a seat's real work and never spends a token.

The website content gate reads the documentation itself: it asserts that every topic names a route that exists, that every cited source path exists on disk, and that no section is empty. Documentation that cites a file which has moved fails the gate.

A passing gate is not a working feature

A gate exercises the code it names, with the stubs, fixtures and fakes it names. It does not exercise a real provider, a real device, a real account or a real deployment.

The distinction is the reason this site separates declared from live. Two council seats carried evidence that was never true, and both said measured; only a gate that spawned the real binary caught it, because the other gates injected a stub runner and proved the plumbing rather than the argument.

So a green suite is evidence about the code on this host at this time. It is not a certificate of production connectivity, and a dated report does not certify a later change.

Generated artefacts and drift

The subsystem directory lists actual tasks, their tiers, verification kinds and recorded evidence. A done flag or an imported historical claim is not a current health certificate.

Regenerate the HUD catalog with python scripts/build_hud_catalog.py and this website's inventory with python scripts/build_website_inventory.py. Both accept --check, which compares the generated file against the repository and fails on drift without writing anything.

Implementation sources

Repository paths below describe the implementation. They do not certify a running deployment or a configured external account.

  • docs/systems-plan.json
  • bench/run_all_tests.py
  • bench/test_website_content.py
  • bench/test_council_seats.py
  • tools/probe_seat_contracts.py
  • scripts/build_hud_catalog.py
  • scripts/build_website_inventory.py

Subsystem directory · HUD guide · Verification and known limits