note(2pc): realign REVIEW-2pc to narrowed scope — registry pull-through cache DROPPED per operator; 2pc is now prune-policy only (PC1 surgical prune + teardown must NOT remove images, PC2 confirm PAT-auth+local-store retention, PC3 deploy/teardown/redeploy reuses local layers). Break-it checklist updated.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

machine-docs/REVIEW-2pc.md

@@ -1,38 +1,53 @@
-# REVIEW-2pc — Adversary verdicts for Phase 2pc (image pull-through cache + sane prune)
+# REVIEW-2pc — Adversary verdicts for Phase 2pc (sane image-prune policy)
 
 SSOT: `/srv/cc-ci/cc-ci-plan/plan-phase2pc-image-cache.md`. DoD = PC1 + PC2 + PC3,
 each Adversary cold-verified here before Builder may write `## DONE` to STATUS-2pc.md.
 
+**SCOPE CORRECTION (operator, 2026-05-29):** the registry pull-through cache (old PC2)
+is **DROPPED / deferred to IDEAS** — single authenticated non-pruning host ⇒ Docker's own
+local image store already IS the cache. Phase 2pc is now **prune-policy only**.
+
 ## Status: AWAITING CLAIM
-Phase 2pc opened 2026-05-29 (operator interjection into paused Phase 2). As of this
+Builder has not yet bootstrapped 2pc (no STATUS-2pc.md, no `claim(2pc…)`). No gate
-file's creation the Builder has **not** bootstrapped 2pc (no STATUS-2pc.md, no `claim(2pc…)`).
+claimed → no verdict yet. Watching origin/main; cold-verify on first claim.
-No gate is claimed → no verdict yet. Watching origin/main; will cold-verify on first claim.
+
+## DoD (narrowed scope)
+- **PC1 — Conservative prune policy.** No reflexive `docker image prune -af`. NEVER prune
+  during a deploy/test run. Keep base/in-use images. Prune only dangling + age-gated old
+  layers, only under genuine disk pressure. Per-run teardown still removes the run's
+  **volumes/secrets/services** (sacred) but **must NOT remove images.**
+- **PC2 — Local cache retained + authenticated (confirm).** Daemon stays PAT-authenticated
+  for `docker.io`; local image store retained across runs, teardowns, reboots → repeat
+  deploy reuses local layers (no re-download), at most an authenticated manifest check.
+- **PC3 — Verified + documented.** Adversary proof: deploy → teardown → redeploy does NOT
+  re-download layers (via `docker` events/pull output / measured pull-time drop); normal run
+  doesn't evict cached base images; disk bounded WITHOUT `-af`. docs/ notes policy;
+  deviations in DECISIONS.md.
 
 ## Pre-claim baseline recon (read-only; NOT a verdict — just what "before" looks like)
-- **PC1 / prune.** Current prune policy lives in `nix/modules/swarm.nix:15-19`:
+- **autoPrune** (`nix/modules/swarm.nix:15-19`): `flags = ["--all" "--filter" "until=24h"]`,
-  `autoPrune = { enable=true; dates="daily"; flags=["--all" "--filter" "until=24h"]; }`.
+  no `--volumes`. `--all` evicts *any* image unused for 24h → would drop warm base images
-  - `--all` removes *any* image not used by a container in 24h → would evict warm base
+  between runs (exactly PC1's complaint). The destructive `docker image prune -af` cited in
-    images between runs (PC1's complaint). No `--volumes` (correct — warm volumes survive).
+  JOURNAL-2 (507, 690-693) was a **manual** operator action mid-deploy, NOT this systemd unit.
-  - The destructive `docker image prune -af` churn cited in JOURNAL-2 was a **manual**
+  → PC1 must (a) tighten autoPrune off `--all` toward dangling-only/age-gated, AND (b) ensure
-    operator action mid-deploy (JOURNAL-2:507,690-693), not this systemd unit.
+  no `-af` exists in any harness/janitor/teardown code path.
-  - PC1 acceptance to run cold: confirm (a) no reflexive `-af` remains in harness/janitor
+- **Teardown image-removal grep target:** DECISIONS.md:708 documents a manual cleanup recipe
-    code paths; (b) prune never fires during an active deploy/run; (c) a normal run does
+  ending `docker image prune -f`. Must confirm the *automated* per-run teardown
-    NOT evict cached base images; (d) disk stays bounded without `-af`.
+  (run_recipe_ci.py / harness) does NOT `docker rmi` / `image prune` the run's images.
-- **PC2 / pull-through cache.** Does NOT exist yet — no `registry:2`, `registry-mirrors`,
+- **No registry cache** exists (confirmed) and per scope correction none should be built.
-  `registry-1.docker.io`, or pull-through config anywhere in repo (`nix/`, `runner/`).
-  Expect a new `nix/modules/registry-cache.nix` (Nix-reconciled service) + daemon
-  `virtualisation.docker.daemon.settings.registry-mirrors` + sops PAT (nptest2) for upstream auth.
-  - PC2 acceptance to run cold: (a) 2nd deploy of an image pulls from cache not Docker Hub
-    (cache logs / measured pull-time drop); (b) survives a prune (re-pull is local, not a
-    Hub hit); (c) measured cold-vs-warm deploy speedup; (d) cache-miss pulls authenticate.
-- **PC3 / bounded+documented.** Cache must have disk cap / own GC (LRU/old eviction);
-  scope = docker.io only; docs/ notes cache+prune policy; deviations in DECISIONS.md.
 
-## Break-it probes to run once PC2 lands (anti-anchoring checklist)
+## Break-it probes to run once PC1 claimed (anti-anchoring checklist)
-- Cache must NOT mask a genuinely-broken image pull (cardinal rule — don't weaken a test).
+1. **Teardown must NOT remove images.** Deploy a recipe, capture `docker images` digest set,
-  Probe: request a nonexistent/garbage tag through the mirror → must still FAIL, not serve stale.
+   run the real teardown, re-check: the recipe's image layers must STILL be present locally.
-- registry-mirrors must be transparent to abra/swarm — verify a real abra deploy's pulls
+2. **Redeploy reuses local layers (PC3 core).** After teardown, redeploy the SAME recipe and
-  traverse the cache with NO command change / no pull special-casing.
+   confirm via `docker events` / pull output there is NO layer download (only a manifest
-- Cache survives a D8-style rebuild as a service (Nix-reconciled), but its contents are NOT
+   check, or fully local). Measure the pull-time delta vs a genuine cold pull.
-  in the git closure (re-warmed by pulls) — verify both halves.
+3. **No mid-run prune.** Grep all code paths; confirm nothing prunes images while a
-- PAT secret must not leak into published logs / dashboard / world-readable registry config.
+   deploy/test is active (the JOURNAL-2 landmine). autoPrune is daily/off-run only.
+4. **Cache must NOT mask a broken image (cardinal rule).** A pinned version still resolves to
+   the correct digest; a genuinely-new/changed digest still triggers a real pull — the
+   retained store must not serve a stale image for a recipe that actually changed.
+5. **Disk stays bounded without `-af`.** Confirm the surgical policy + disk-pressure trigger
+   actually reclaims under pressure (don't trade rate-limit churn for a full disk).
+6. **PAT auth intact + not leaked.** Daemon still authenticated to docker.io (under 200/6h);
+   PAT not exposed in published logs / dashboard / world-readable config.