Files and folders
Everything a run writes, and where.
One folder per site, one folder per run inside it. The site level keeps a copy of the newest run's reports.
The runs folder
The layout below is what a site looks like after a few runs.
runs/<site>/
SITE.md what the page sells, to whom, its walls
MAP.md, map.json every page the crawler found
FLOW.md drafted checkpoints (--flow)
personas/ the prospects built for this site
flows/ flow files for this site only
analytics.json optional: real-visitor weighting
VERDICTS.md mark each wall real: or false:
COMPARE.md A/B verdict (--compare)
AGGREGATE.md … copy of the newest run's reports
<date>/<time>/ one run (<time>--<variant> for A/B)
RUN.md seat, model, sessions, exits, tokens, minutes
AGGREGATE.md the one-page report
DETAIL.md every table behind it
VERIFIED.md the verifier's panels (--ladder)
REPORT.md the writer's report (--ladder)
wide/ verify/ deep/
<model>/<session>/Per session
Each session folder holds the evidence for one prospect.
| File | Contains |
|---|---|
session.jsonl | Every step, thought and typed string. |
meta.json | URL, brain, model, effort, version, variant, how it ended, flow score, assertions, token usage, steps and duration. |
verifications.jsonl | One line per completion claim: the page that was judged, the verdict, and any --expect results. |
shots/ | Retina screenshot per step, and every email received. |
video.mp4 | The recording (video.webm without ffmpeg). |
filmstrip.html | Every step with its thought. |
report.md | The session's story, with Assertions when --expect was used. |
FIXES.md | The expert panel. |
Files you add
Four files change how a run behaves, and none of them is required.
| File | Purpose | Guide |
|---|---|---|
.env | Mail and order settings | Environment |
personas/*.yaml | Your own prospects | Personas |
flows/*.yaml | Journeys to score | Flows |
runs/<site>/analytics.json | Real-visitor weighting | Personas |