feat(fleet): stranded-cargo fallback frees parked-full traders (v2.9.3)

[mabry1985]

Closes #55.

Problem

A trader parked full of a good the local market doesn't buy sits idle indefinitely — observed live 2026-07-05: PROTOTRADER-1 sat with 40 AMMONIA_ICE at a non-buying waypoint for ~1 hour, needing two manual A2A rescues. _trade_loop's no-route branch just sleep(15)s, and the engine only detected the strand as a sitrep hint (report()), never acted.

Fix

New _clear_stranded_hold, called only from _trade_loop's no-route branch (so it never pre-empts a real trade run): repositions the ship to the nearest in-system market that buys the stranded good and sells it, accepting sub-margin — freeing the ship beats hoarding dead capital.

Guards (each tested)

• Never mid-flight (nav.status != IN_TRANSIT).
• Contract cargo protected — a stuck contract lead falls back to _trade_loop still holding its deliver haul; /my/contracts is the authoritative signal, and an unreadable list protects everything (fail-safe).
• Buyer must be provably reachable — T._good_markets raise (unproven scan) or an empty buyer set leaves the cargo aboard. This sells stranded cargo; it never jettisons (proof-gated destruction stays the mining loop's job).
• One side-trip per window (cleared flag).

Tests

test_stranded_cargo.py (8): reposition+sell, contract-cargo protected, no-buyer holds, unproven-scan holds, in-transit no-op, unreadable-contracts fail-safe, empty-hold no-op, once-per-window wiring.

Full suite: 338 pass incl. the 8 new; ruff clean. (One unrelated pre-existing failure in test_idle_capital_banking_toward_the_gate_stays_quiet reproduces on clean main in my local env — host idle_capital/gate-buffer logic, not touched here.)

🤖 Generated with Claude Code

[protoquinn]

👀 Quinn is reviewing — verdict (PASS / WARN / FAIL) + findings to follow.

[protoquinn]

QA Audit — PR #56 | feat(fleet): stranded-cargo fallback frees parked-full traders (v2.9.3)

VERDICT: WARN (CI still settling — test: in_progress; approve-on-green policy promotes this to APPROVED automatically once every check passes terminal-green)

---

CI Status

• test: in_progress ⏳

Diff Review (3 files, +290/-1)

• fleet.py: New _clear_stranded_hold() called from _trade_loop's no-route branch — repositions a parked-full trader to the nearest in-system market that buys the stranded good, sells it sub-margin, then loops to re-check routes. One side-trip per window via cleared flag.
• protoagent.plugin.yaml: version bump 2.9.2 → 2.9.3.
• test_stranded_cargo.py: 8 new tests covering reposition+sell, contract-cargo protection, no-buyer hold, unproven-scan hold, in-transit no-op, unreadable-contracts fail-safe, empty-hold no-op, once-per-window wiring.

Observations

• CLAWPATCH/LOW: Test suite feature mapped as structurally sound — 8 features mapped, 1 finding (test-suite signal), no correctness or security flags raised.
• LOW: cleared = True is set unconditionally before _clear_stranded_hold() returns — a transient failure (e.g. contracts API blip) won't retry in the same window. This is the documented design ("one side-trip per window") and resets on the next _trade_loop entry; no action needed, just noting.
• The fail-safe on unreadable /my/contracts (return False → protect everything) plus the "never jettison" guard (empty buyer set leaves cargo aboard) are well-judged. Contract cargo protected via authoritative signal, not heuristic.

— Quinn, QA Engineer

[protoquinn]

Submitted COMMENT review on #56.