Documentation

The stuff that didn't fit in the main README without turning it into a novel. Jump straight to the page that matches whatever's ruining your day.

Ops & infrastructure

Configuration: every env var used for configuration.
Authentication: login, roles, users, invites, API keys, SSO.
Storage: fs, s3, azure. Litestream replication.
Deployment: Docker basics.
Notifications: ship report outcomes and summaries to Slack or any webhook. Event rules on upload, schedule rules on cron, delivery log.
UI white-label: title, logo, favicon, header links. Make it look like your team's tool rather than someone else's.

Uploading reports: every upload path. The reporter plugin, sharded runs with auto-merge, HTML upload via CLI, GitHub Sync, manual report generation, and the direct-to-S3 presigned uploads.
Data migration: one-time import of an existing instance's reports and results, served in place from your storage backend.
GitHub Sync: scrape Playwright artifacts out of GitHub Actions on a cron so you don't have to touch any workflow files or just download the latest reports to your local instance.

Overview dashboard: a configurable home page - per-project letter grades and OK/NOT OK verdicts rolled into one signal, pinned to /.
Analytics dashboard: trends, slow tests, failure clustering.
Regression tracking: telling a genuine break (was green, now broken) from a flake - opened, tracked with its breaking/last-green commits, resolved on recovery.
Test management & quarantine: how flakiness is scored, when auto-quarantine kicks in, and how the reporter actually skips quarantined tests at runtime (spoiler: a file on disk, no magic, unfortunately).
Report export (PDF): one self-contained PDF per report, generated server-side for audits and review attachments.

Analysis: how failure analysis is done and why - providers, the model registry, screenshots, the background queue.
Routing: multi-model strategies (fusion, council, cascade, refine) per task - what each costs, when it helps, and how to experiment.
Selection: how to choose a model - general recommendations for local (LM Studio / oMLX / Ollama) and remote (OpenRouter).
Code assistant: hand your coding partner (pwrs-cli + Claude Code skill) access to the failure data so it can stop guessing what broke and pull real run history instead of poking around with gh, grep, and a web browser.