fix: let a removal fixture actually build the thing it claims still builds

FE-GATE-020 proves a capability can be removed by rebuilding the whole project
without it. The fixture it built could not get that far, and the failures all
came from the fixture rather than from anything about removability.

It was not a repository. The supply-chain inventory is defined as the tracked
file set, so it asks `git ls-files` what the project contains; with no
repository to ask, generation failed and took every provider suite down with
it. It is now initialised on preparation and committed after the removal — not
before, or the index would still list the files the removal deleted.

It had no `.gitignore`, so once it did have a repository, every generated
artifact and every linked module landed in the index and the inventory refused
the fixture for tracked and generated paths colliding. It carries the ignore
rules now, and therefore records the same tracked set as the repository it was
copied from.

Each removal script kept its own copy-target list and they had drifted: the
reference-feature fixture omitted `playwright.capabilities.config.ts`, which the
inventory requires. There is one list now. It also gained the install and
workspace identity — `.npmrc`, the lockfile, the workspace file — without which
the fixture is a different project, and the release evidence a candidate is
assembled from, without which no candidate can be built at all.

A tracked root the removal deletes is no longer required of the result: the
optional-recipe fixture deletes `recipes/`, and the inventory policy demanded
it back. Roots that are gone are pruned from the fixture's policy.

Two smaller causes. A platform integration file asserted the reference feature's
own route ids, so removing the feature left it importing a deleted module —
typecheck, the test run, coverage and the residue scan all failed on that one
misplaced assertion, which now lives in the feature's test tree. And the
canonical exact-count authority was re-imposed on a contract the fixture
deliberately reduces, failing the fixture for the reduction it exists to prove;
`CI_CONTRACT_MODE` already marked those runs and is now honoured by default.

The reference-feature fixture goes from failing before its first assertion to
1,612 passing with one failure, and that one is the live process-tree
observation test already red on the main tree.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/operations/adapter-remediation-ledger.md

@@ -462,6 +462,56 @@ The full `tests/unit` + `tests/integration` run is **1,845 passed / 1,864**,
 cgroup, RLIMIT and `/tmp` permission behaviour already recorded above — the same
 file failed identically before this work. No adapter test fails.
 
+## Operational contract review (2026-08-15)
+
+A fourth review looked past the adapter layer at the operational contract:
+feature on/off, environment separation, folder boundaries, and which gates were
+actually green. It found five red gates and three structural gaps. Every row
+below names the defect, not the symptom.
+
+| id | area | disposition | what was actually wrong |
+| --- | --- | --- | --- |
+| `OPS-01` | release | `FIXED` | `public/` is copied verbatim into `dist/`, so every build — production included — shipped the local runtime document. Runtime config now comes from `config/runtime/<profile>.json`. |
+| `OPS-02` | release | `FIXED` | Release coherence proved the artifacts agreed with each other, never that they belonged in production. `FE-GATE-027` refuses an artifact whose `APP_ENV`, auth mode, endpoints or build identity do not match a declared `RELEASE_TARGET`, and refuses an undeclared target outright. |
+| `OPS-03` | runtime | `FIXED` | `REQUEST_TIMEOUT_MS` was validated and then never passed to the V3 executor; every operation ran on its contract's own deadline. It is now a ceiling that may tighten a contract, never loosen one. |
+| `OPS-04` | build | `FIXED` | `VITE_ROUTER_BASE_PATH` drove the router and the Service Worker scope but not Vite's asset `base`, so a sub-path deployment emitted root-absolute assets. One value now feeds all three. |
+| `OPS-05` | provider | `FIXED` | bubblewrap 0.9.0 drops whatever follows the option stream inside an `--args` file, so the sandboxed command was never executed: bwrap printed usage and exited 1. Options stay hidden; the command travels on real argv. |
+| `OPS-06` | provider | `FIXED` | The scope wrapper read its liveness pipe through `fs`, a blocking `read(2)` on a pipe the supervisor never closes. `process.exit` deadlocked joining that thread, so a completed provider was reported as a timeout kill. |
+| `OPS-07` | release | `FIXED` | `mkdir`/`open` modes were left to the ambient umask, so a hardened runner produced directories it could not enter and handed `tar` a file it could not re-open. |
+| `OPS-08` | release | `FIXED` | Promotion cleanup deleted this promotion's exact five through a pinned descriptor and only then noticed the leaf had been substituted, leaving a half-emptied directory a retry could not distinguish from a completed one. |
+| `OPS-09` | removability | `FIXED` | The removal fixture was not a repository, had no `.gitignore`, and each removal script kept its own copy-target list that had drifted. Supply-chain generation therefore failed inside every fixture and took the whole provider suite down with it. |
+| `OPS-10` | removability | `FIXED` | A platform integration file asserted the reference feature's route ids, so removing the feature left it importing a deleted module. The assertion moved to the feature's own test tree. |
+| `OPS-11` | removability | `FIXED` | A removal fixture runs against a deliberately reduced CI contract; the canonical exact-count tests re-imposed the full authority on it and failed the fixture for the reduction it exists to prove. |
+| `OPS-12` | browser | `FIXED` | Four browser-capability specs answered capability requests without the `protocol` field the hardened envelope requires, so every capability was refused and the download and part-upload paths asserted against an empty transcript. |
+| `OPS-13` | browser | `FIXED` | A refused capability document answered `recovery: NONE`, contradicting both the design record and the vault, which already answers `REISSUE_CAPABILITY`. |
+| `OPS-14` | performance | `FIXED` | Playwright matches accessible names by substring, so the navigation entry matched the home page's call to action too; the run died on a strict-mode violation before the first measurement and produced no evidence at all. |
+| `OPS-15` | visual | `FIXED` | The platform overview baseline predated the reference routes moving from `integration-defined` to `session-required`, so the only visual gate covering that page failed for its own staleness. |
+| `OPS-16` | architecture | `FIXED` | `src/contracts` imported `src/application` for the shared `Result` and the compatibility predicate; neither package owned the shared vocabulary. Both moved down to contracts. |
+| `OPS-17` | architecture | `FIXED` | The documented "no adapter depends on another concrete adapter" rule had no executable form, and `diagnostics` imported a guard out of `telemetry`. The guard moved to the adapter kernel and the rule is now enforced with a same-directory backreference. |
+| `OPS-18` | architecture | `PARTIAL` | Generic presentation still reads the installed-feature registries. The rule freezes the exact set of modules doing so today; a new edge fails. Lifting the assembly into `bootstrap` is not done. |
+| `OPS-19` | documentation | `FIXED` | README and the manual accessibility checklist both claimed six routes while ten were registered, leaving four screens outside the declared manual review scope. The list is now derived from the route registry by `verify:documentation`. |
+
+### Still red after this pass
+
+`tests/unit/ci-artifact-contract.test.ts` → *applies effective aggregate cgroup
+limits without exposing command or credentials*. It reads the live process tree
+and cgroup of a running sandbox, and the supervisor now completes a whole run in
+well under a second while `systemctl show` and `ps` each cost hundreds of
+milliseconds, so the observation loses the race. It was already red before this
+work and is not a product defect; the assertions it makes about cgroup limits
+and credential exposure are not currently proven by an automated run.
+
+Lab performance now produces evidence, and that evidence shows the
+named-interaction budget missed on this machine (367–724ms against 200ms). The
+metric measures a full lazy-route navigation while the budget is an
+INP-shaped 200ms, so the two do not describe the same thing. No budget was
+changed to make this green.
+
+WebKit remains unavailable in this environment (`libevent-2.1-7t64`,
+`libavif16` are not installed), so 14 browser-capability specs and the WebKit
+E2E project are unverified here. Chromium and Firefox are 28/28 and visual is
+5/5.
+
 ## Rules for updating this ledger
 
 - A row moves out of `NOT_STARTED` only with a linked red test, its green run, and the commit id.