Portal gallery

A walk through the live TFactory portal at tfactory.freundcloud.org.uk, captured against real verify runs. These are authenticated screenshots of the running system — not mock-ups. Where a tab is empty or a run did not reach a level, the UI says so plainly; we kept those shots rather than staging green ones.

The front door

TFactory login — API-token or SSO — The login screen. Authenticate with an API token (or SSO); the SPA validates the token against the API before rendering the app.

The pipeline board

The TFactory pipeline board — Plan, Generate, Execute, Report columns with live tasks — The four-stage pipeline (Plan, Generate, Execute, Report) above columns of real tasks — including a triaged Report-column task and a failed run. This is the live state, in-pod verify path.

A task in the planning stage with lane chips and planner warnings — A task open on its Status tab — lane chips (unit · browser · api · integration · mutation), subtask counts, and any planner warnings.

A finished task, tab by tab

Task detail — overview — Task detail opens on Status. The tab strip — Status · Lanes · Verdicts · Report · Acceptance · Logs · Evidence — is enabled per tab as the run produces the data behind it.

The Verdicts tab — real reject verdicts with the five-signal breakdown — Verdicts. Each generated test gets a verdict (accept / flag / reject) backed by five signals: coverage, stability, mutation, lint-promotion, and semantic relevance.

The Report tab — VAL ladder above the triage buckets — The Report tab leads with the Verification Assurance Level — exactly how far the run was verified — above the Triager's dedup / commit / flag / reject buckets.

The Acceptance tab — per-criterion fidelity, honest UNVERIFIED labels — Acceptance-criteria fidelity. Each criterion is graded against a test that actually ran; criteria with no passing test are labelled UNVERIFIED rather than rounded up.

The Logs tab — phase-by-phase agent logs — Logs. The phase-by-phase record of the five agents running the pipeline.

The Evidence tab — honest about an absent browser lane — Evidence. For browser-lane runs this is a gallery of screenshots and recordings; for a task with no browser lane it says so plainly.

A failed run, shown honestly

A failed task — replan budget exhausted, with planner warnings — A failed run on the in-pod path: the planner exhausted its replan budget and the run is marked `failed` rather than looping forever. The warnings explain why.

Inside a project

The MCP server configuration and agent roster — MCP. Which MCP servers are available to the agents, plus the agent roster (Spec Gatherer, Researcher, Writer, Critic, …) and their models.

The Files browser — Files. Browse the task workspace — generated tests, contract, findings.

Visual Reports — visual-inspection baselines and diffs.

The GitHub PRs view — GitHub PRs. The PR endgame, surfaced inside the project once a GitHub token and repo are configured.

Test Plans — GitHub-connected view — Test Plans. Connects to GitHub to organise plans against issues.

These screenshots were captured by scripts/capture-portal-gallery.ts driving the live portal with a real form login (Playwright + Chromium) and saved to docs/static/img/gallery/.