[AZ-421] Batch 82 housekeeping

Archive AZ-421 to done/ and advance autodev state to await batch 83.

Co-authored-by: Cursor <cursoragent@cursor.com>

_docs/02_tasks/done/AZ-421_ft_p_15_16_18_cache_offline.md

@@ -0,0 +1,79 @@
+# FT-P-15 + FT-P-16 + FT-P-18 — Tile cache schema + offline-only + no-raw-frame retention
+
+**Task**: AZ-421_ft_p_15_16_18_cache_offline
+**Name**: Tile cache manifest schema, resolution floor, offline-only operation, no raw nav/AI-cam retention (AC-8.1, AC-8.3, AC-8.5, RESTRICT-SAT-1)
+**Description**: Combined coverage for three storage / cache compliance scenarios that share the same `tile-cache-fixture` and FDR-archive observation surface.
+**Complexity**: 3 points
+**Dependencies**: AZ-406, AZ-407
+**Component**: Blackbox Tests / Positive / Storage policy (epic AZ-262)
+**Tracker**: AZ-421
+**Epic**: AZ-262 (E-BBT)
+
+## Problem
+
+Three contract-level checks on the SUT's storage behavior — manifest schema (FT-P-15), offline-only operation (FT-P-16), and no-raw-frame retention (FT-P-18) — share the same fixture mount and observation surface. Combining them into one task avoids three near-identical pytest harnesses.
+
+## Outcome
+
+- pytest scenarios at `e2e/tests/positive/test_ft_p_15_cache_schema.py`, `test_ft_p_16_offline_only.py`, `test_ft_p_18_no_raw_retention.py` (three test files; one task).
+- FT-P-15: SUT's startup self-check observed via FDR; assert each tile manifest entry has CRS, tile matrix, dimension, lat-adjusted m/px, capture date, source, compression; m/px ≥ 0.5.
+- FT-P-16: 60 s Derkachi replay with `e2e-net.internal: true`; assert all egress to non-`e2e-net` destinations is 0 (paired with NFT-SEC-02).
+- FT-P-18: walk FDR + tile cache for any file matching nav-camera raw-frame pattern; assert only the failed-tile-gen thumbnail log files present (≤0.1 Hz cadence); total thumbnail log < 1 GB over 8 h (NFT-LIM-02 cross-check).
+
+## Scope
+
+### Included
+- FDR-side startup self-check observation (FT-P-15).
+- Network-egress counter read from Docker network stats (FT-P-16; the actual block is enforced by `internal: true` from AZ-406).
+- Filesystem walk of FDR + tile cache (FT-P-18).
+- Aggregate assertion per scenario.
+
+### Excluded
+- Defense-in-depth (DNS blackhole) variant — owned by NFT-SEC-05 inside AZ-437.
+- Cache-poisoning safety — owned by NFT-SEC-01 (AZ-436).
+- Stale-tile rejection — owned by FT-N-05 (AZ-427).
+
+## Acceptance Criteria
+
+**AC-1: FT-P-15 — manifest schema completeness**
+Given the SUT's startup self-check observable via FDR
+Then each tile manifest entry has all of: CRS, tile-matrix identifier, pixel dimension, lat-adjusted m/px, capture_date, source, compression. Missing fields fail the AC for that entry.
+
+**AC-2: FT-P-15 — resolution floor**
+Given the manifest entries
+Then m/px ≥ 0.5 for every entry; entries below the floor are rejected at load (verified by FDR `tile-load-rejected` events).
+
+**AC-3: FT-P-16 — offline operation**
+Given the SUT runs against the local cache only
+When 60 s of Derkachi replays
+Then Docker network stats show 0 packets to non-`e2e-net` destinations from the SUT container.
+
+**AC-4: FT-P-18 — no raw-frame retention**
+Given a completed Derkachi replay
+When the test walks `fdr-output` + tile-cache volumes
+Then no file matches the nav-camera raw-frame pattern (JPEG/RAW with original 5472×3648 dimensions, OR 880×720 H.264-decoded full-resolution frames). Only the ≤0.1 Hz failed-tile-gen thumbnail log files are present.
+
+**AC-5: FT-P-18 — thumbnail log budget**
+Given the thumbnail log
+Then `total_thumbnail_log_size < 1 GB` extrapolated over an 8 h flight (NFT-LIM-02 cross-check).
+
+**AC-6: parameterization**
+Given conftest parameterization
+Then all three methods run per `(fc_adapter, vio_strategy)`.
+
+## System Under Test Boundary
+
+End-to-end through public boundaries.
+
+- **Allowed**: FDR archive read, filesystem walk of mounted volumes, Docker network stats (a public Docker API).
+- **Forbidden**: importing the SUT's cache module, monkeypatching the manifest loader.
+
+## Constraints
+
+- "Raw-frame pattern" detection uses file-format heuristics (magic bytes for JPEG/RAW) AND dimension probing. False positives are inspected manually; false negatives that miss real raw frames are a real defect.
+- The 1 GB / 8 h thumbnail budget is extrapolated from a shorter run by `(observed_size / observed_duration_h) × 8`.
+
+## Document Dependencies
+
+- `_docs/02_document/tests/blackbox-tests.md` § FT-P-15, § FT-P-16, § FT-P-18
+- `_docs/00_problem/restrictions.md` § Satellite Imagery (manifest schema)