AtlatestRenderedmarkdown

folio / tree / docs / f0evidence-summary.md

Read the source

Readme

F0 evidence summary

Date: 2026-08-09

Normative commit: 699b7ea

Green gate

python3 tools/f0_contract_gate.py
python3 tools/f0_inventory.py /home/daviwil/Ops/folio \
  --positive-control --no-op-copy /tmp/folio-f0-copy.eiOW5Z --rounds 2 \
  --output docs/f0/evidence-green.json

Result: PASS. All 2,643 files were classified; 1,474 product-visible tasks were counted, matching the deployed folio/status control. The estate contains 1,448 task ID occurrences / 1,447 unique visible IDs, one duplicate ID (t-7681), 26 visible tasks without IDs, and three independently found unreachable IDs. The positive note, task, raw-ID, and link controls all fired. A disposable 2,641-file post-product-probe copy survived byte-exact no-op rewriting. The two census rounds took 2.09s and 2.92s.

The contract gate froze 30 MCP tools and 35 library mutation/query/index entry doors.

Product-path baseline

python3 tools/f0_product_baseline.py \
  --binary /home/daviwil/Projects/Code/sigil/folio/build/release/bin/folio \
  --root /tmp/folio-f0-copy.eiOW5Z \
  --output docs/f0/evidence-product-baseline.json

On the disposable copy: cold start plus initialize 43,510ms; common status query 3,720ms; full 2,588-note index rebuild 23,997ms; controlled inbox mutation 268ms; warm status query after mutation 4,033ms; peak child RSS 198,016 KiB. Every MCP control returned a success response and the process exited zero. No running service or canonical file was involved.

Accepted sabotage

python3 tools/f0_inventory.py /home/daviwil/Ops/folio \
  --omit-prefix notes/topics --rounds 1 \
  --output docs/f0/evidence-sabotage.json

Result: expected exit 1. It named omitted files and reported the exact classification denominator mismatch: only 1,541 of 2,643 files scanned. Restoring the scan produced the green result above.

Boundary observation

A source-level product harness is frozen in tools/f0_product_baseline.sgl. The worktree's dependency rebuild did not complete within a five-minute diagnostic bound, and the pre-existing July development library stack-overflowed on this 2,588-note estate. Neither was used as positive evidence. The deployed August release binary was therefore used for the current product baseline, and the independent Python census was checked against its 1,474-task result.

BLOCKING REVIEW PENDING. F1 has not started.