CRITICAL
[INVESTIGATE] BF Type 1 combo scores measure held/sold binary not actual returns — all combo rankings are potentially co…
HIGH
[INVESTIGATE] Session-start digest block missing from hook injection
HIGH
[ASSESS] Reconcile checkpoint-date discrepancy on 2026-07-12 pipeline kill
HIGH
[BUILD] soldtrack.html long-term size management — prevent hitting Cloudflare 25 MiB hard limit
HIGH
[INVESTIGATE] Session digest token audit — index all session startup injection items and their token cost
HIGH
[GOVERNANCE] Add Skill() built-in command exclusion rule to ~/.claude/CLAUDE.md
HIGH
[GOVERNANCE] Auto-snapshot before every code edit — `.bak` rollback discipline must be machine-enforced, not remembered
HIGH
[GOVERNANCE] Measurement Gate (D3) — extend /seal-plan + close_audit.py to require measurement paste for quantitative cl…
## 2026-07-19 — Full pipeline run (3 new CSVs) via step-by-step foreground execution; mobile deployed; stale registry gaps found
**Session (3 new CSVs dropped → full `run_pipeline.py` refresh):**
- **New CSVs processed:** `Invest-Multi(4_30_23) - Stock score (holdings) (70).csv`, `New Research Tab 3.20.24 (14).csv`, `Secondary Metrics (69).csv`. Auto-imported to `data/raw/` via Step 0 during the first (killed) run attempt.
- **Background run killed mid-flight (unexplained):** first `run_pipeline.py` invocation, backgrounded per user request, was killed by the harness at Step 11.5 (Signal Intelligence Layer) with no diagnosable root cause on this end — no crash, no error, no resource signal visible from the session. Root-caused the standing rule (`~/.claude/projects/.../memory/feedback_long_running_pulls.md`) to explicitly cover `run_pipeline.py`, not just finviz/ratio pulls.
- **False alarm — thought-orphaned processes were actually a separate manual finviz/ratio pull** (`finviz_pull.py` + `ratio_pull.py --fresh`, PIDs 22428/5392/22764/22848) running concurrently in a different terminal — confirmed via `Get-CimInstance Win32_Process` command-line inspection before any kill action. Left untouched.
- **Stale `.pipeline.lock` cleared** (PID 14588, confirmed dead) — was blocking any new pipeline invocation.
- **Second run executed step-by-step in foreground** (~40 individual script invocations, no backgrounding except Step 5b Cross-Tier Formula Builder — the one step that structurally exceeds the 10-minute foreground tool cap at ~16 min typical runtime) — full run completed clean.
- **Found `references/PIPELINE_REGISTRY.md` is stale** — missing 3 real pipeline steps present in `run_pipeline.py` source: Step 8.58 (Gate History Dashboard), Step 11.7 (Session Digest Dashboard), Step 11.8 (Equity Comp refresh). All three were run directly from source inspection so nothing was actually skipped in this run, but the registry doc needs a regen (last regenerated 2026-04-27).
- **Session Digest payload regenerated** — resolves the open `[INVESTIGATE] Session-start digest block missing from hook injection` recurring gap (TASKS.md NEW tier) for this session at least; root cause (why it stopped firing across sessions) still not identified.
- **Control Group Lab verdict reconfirmed on fresh data:** FORMAL recs UNDERPERFORMING control by -13.2pp (p=0.034, n=43) — same finding surfaced pre-kill, now verified stable on the completed run.