# Hatchways Limited Pilot Setup

Define the smallest information Hatchways would need to run one internal GitHub-style assessment through HotTea without waiting on a GitHub App or procurement rollout.

Boundary: One internal or synthetic Hatchways-style repo assessment, one reviewer, one candidate session, GitHub Actions fallback only, no customer production rollout.

## Hatchways Inputs

### repo_url

Needed: A source or synthetic candidate repo URL with a practical bug-fix, feature, or code-review task.

Why: HotTea packages repo, SHA, PR, diff, tests, and session exports into the reviewer packet.

### task_brief

Needed: Candidate-facing instructions, expected files, public tests, hidden-test intent, and finish criteria.

Why: The same brief appears in the candidate page, VM ae-task command, and reviewer packet.

### public_test_command

Needed: One command a candidate can run locally or in the VM.

Why: Public checks anchor the candidate's verification behavior and reviewer timeline.

### hidden_result_summary

Needed: Optional pass/fail summary or rubric label from Hatchways hidden grading.

Why: The reviewer packet can display hidden grading without exposing hidden tests to candidates.

### reviewer_email_or_role

Needed: One Hatchways evaluator role for packet review.

Why: The pilot only needs one reviewer to decide whether the evidence adds signal beside current grading.

## Adapter Steps

1. Create or choose a single candidate session in HotTea.
2. Add /github-action.yml to the candidate repo or run the same POST shape from the existing Hatchways marking workflow.
3. Set AGENTIC_EVIDENCE_SESSION_ID and AGENTIC_EVIDENCE_TOKEN as repository or workflow secrets for the pilot repo only.
4. Let the candidate use the signed workspace URL and provisioned VM, or attach outside AI transcript evidence if the VM is not used.
5. Open the generated reviewer report and compare it beside Hatchways hidden grading and human review notes.

## Acceptance Evidence

- Reviewer can identify what changed, which tests ran, and whether hidden grading passed.
- Reviewer can inspect substantive AI collaboration separately from procedural prompts.
- Engineering can see how repo/PR/test/diff/session-export evidence enters the packet without a GitHub App.
- Product can decide whether an Agentic Mode add-on is worth account-level GitHub App work.

## Current Blockers

- The real GitHub App still requires account-level creation and GITHUB_APP_INSTALL_URL deployment configuration.
- Until that app exists, GitHub Actions remains the verified fallback adapter.

## Not Claimed

- No official Hatchways partnership.
- No live GitHub App before GITHUB_APP_INSTALL_URL is configured.
- No enterprise procurement readiness.
- No customer production rollout in the pilot boundary.
- No perfect outside-AI prevention or proctoring claim.

## Proof URLs

- Pilot setup JSON: https://hottea.ai/hatchways/pilot-setup.json
- GitHub Action adapter: https://hottea.ai/github-action.yml
- Integration guide: https://hottea.ai/hatchways/integration.md
- Trial acceptance checklist: https://hottea.ai/hatchways/trial-acceptance.md
- Demo handoff: https://hottea.ai/hatchways/demo-handoff.md
- Sample reviewer packet: https://hottea.ai/sample-report