v1.0.1

A maintenance release on top of v1.0.0: four bug fixes from a full-repo audit, with no change to how a healthy run behaves. Everything is backwards-compatible; the engine tiers, the cost ledger, and the VERIFY gate are untouched.

Fixes

• Resumed reports no longer misattribute citations (#3, #7). A run resumed after SYNTHESIZE rebuilt its citation numbering from the full evidence set rather than the corpus-budget subset synthesis actually numbered, so once the budget trimmed anything the resumed report's [n] markers could resolve to the wrong sources. Resume now rebuilds from the same selected subset; a regression test crashes mid-VERIFY under a trimming budget and asserts the resumed references match an uninterrupted run exactly.
• Setup gaps fail fast with remediation instead of a traceback (#1, #6). quarry research and resume now run the same preflight quarry verify does and stop with the exact fix when a local model, Docker, or an API key for the paid engines is missing, instead of crashing several frames deep in the arbiter. Bootstrap also falls back to a WMI adapter check when nvidia-smi is not on PATH, so a real NVIDIA card is reported honestly rather than silently routed to a CPU-only sync.
• Preflight resolves the bundled config from the repo, not the cwd (#4, #10). The SearXNG config checks looked for docker/compose.yaml relative to the current directory, so the installed quarry run from anywhere but the repo root falsely reported the config missing and refused to start. It now anchors on the same location the quarry searxng launcher uses.
• Docker is optional when SearXNG is remote (#5, #9). Preflight treated a missing Docker binary as a blocking failure even when search.searxng_url pointed at an already-running instance. The Docker check is now a skip, not a failure, for a non-local SearXNG URL.
• Documentation was updated to match: the CLAUDE.md module map, Troubleshooting, and the Run Guide (#11).

License: MIT.