Mining recent connect reports + kb/log.md
Goal
cp-skill-connect reports accumulate Synthesis Opportunities, Maintenance Observations, Flags, Off-authorisation Candidates, and Rejected Candidates sections that are rarely read back across reports. kb/log.md already captures some cross-artifact patterns (ABSTRACTION/SYNTHESIS/DUPLICATION/CONTRADICTION/FIX) by hand. This workshop mines the 45 connect reports dated within the last two weeks of 2026-07-18 (kb/reports/connect/**/*.connect.md, mtime -14) together with the current kb/log.md, looking for recurring patterns across reports that no single connect run would surface on its own: repeated synthesis candidates, systemic collection/authorization gaps, recurring rejected-candidate themes, and anything that reads as a missing kb/log.md entry.
What closes it
Every candidate pattern found is triaged to one of: appended to kb/log.md as a new entry, promoted directly to a note/ADR/proposal if it's already clear enough, or explicitly discarded as noise. The batch raw-findings files are scratch — safe to delete once their content is merged or discarded. The workshop closes when the merged pattern list has been triaged and the scratch files are gone.
Evaluation boundary
Scope is exactly the connect reports captured by find kb/reports/connect -type f -mtime -14 -name "*.md" as of 2026-07-18 (45 files) plus the current kb/log.md. This workshop does not re-run cp-skill-connect on anything, and does not resolve the individual design/theory questions a pattern surfaces — those become their own note or workshop if they earn it. This pass is preliminary: a first read for recurring shape, not a final verdict on any candidate.
Bookkeeping
batch-N-raw-findings.md— scratch, one per subagent batch, safe to regenerate from the source connect reports at any time.preliminary-analysis.md— the actual deliverable: merged/deduped patterns with a triage recommendation each.
Complete file listing (generated at build time)