// per-tool methodology
pdf object explorer
drop a PDF · parse raw object tree · detect embedded JavaScript · /Launch actions · encrypted streams · /EmbeddedFile · suspicious patterns · export report · runs locally
public grade
B
solid triageraw 10/14 · raw 9–11 / 14
Bwhat this grade means
real parsing or rule engine · at least one structured export · honest about boundaries
capability class · heuristic screener
rule-based scoring · statistical anomaly detection · gap analysis — flags worth checking, not verdicts
max grade for this class: B (A when honesty + output + UI all maxed)
- heuristic screeners flag leads, not verdicts — every hit needs corroboration
- threshold tuning affects false-positive and false-negative rates; uncommon baselines skew scores
- capability class H is capped at B unless honesty, output usefulness, and UI all score 2 and raw ≥ 12
known limitations
- meets the public B minimum ship bar: raw ≥ 9/14, UI dimension = 2, IF/OU/DQ/RB/HN ≥ 1 each, no critical red flags
- expect rough edges on uncommon schema variants, oversized inputs, or partially corrupted artifacts
- treat flags as leads — corroborate with primary sources before drawing conclusions
- outputs require independent verification before any legal, financial, medical, safety, or evidentiary use
B minimum ship bar
- newly added forensics tools must clear the public B minimum before merging
- minimum: letter grade B or A · raw score ≥ 9/14 · UI dimension = 2 · IF/OU/DQ/RB/HN ≥ 1 each · no critical red flags (missing engine, placeholder logic, no exports)
- the ship bar is enforced by quality.audit.json sidecars and npm run tools:grade-forensics --check