Triage: decide which findings matter
A discovery run leaves you with a raw list of findings. Triage turns that list into decisions: which findings are real, how exploitable they are, and which are worth proving and ultimately fixing. Each verdict comes from reading the code within a fixed time box. Proving a finding with a test or a working reproduction comes later, in a validation loop.
Two records
Section titled “Two records”This section keeps two records. The rubric describes one row per finding; run observations describe one row per run.
| Triage rubric | Run observations | |
|---|---|---|
| One row per | Finding | Run: one tool, harness and model on one repository |
| Answers | Is this finding real, and how exploitable is it? | What did the run produce, and what did it cost? |
| You record | Verdict, exploitability, what decided it, minutes | Findings reported and triaged, verdict split, cost, refusals, blind spots |
| When | While you triage | After you finish triaging a run |
The run row summarises the finding rows. Its triaged count, total minutes and verdict split are totals over the rubric’s records for that run.
Reading order
Section titled “Reading order”- Read the triage rubric before you start: the verdicts, the evidence rules and the time box.
- Triage the run’s findings and record one rubric row for each.
- Fill in one run observations row for the run. Its last section explains which numbers a small sample can’t support.
- For a second run, repeat steps 2 and 3 and compare the two rows.
The Validate a finding exercise starts with steps 1 and 2 on your pilot repository.