Past assessments

Use published GOV.UK service assessment reports as precedent when the project has a local library. An MCP server named gds-assessment-reports searches that library. Do not load the whole library, and do not fetch GOV.UK during the assessment.

The published plugin does not contain the reports. The project builds the library with scrape_gds_reports.py, which sits next to the gds-assessment-how-to skill. Run it from the project root. It writes docs/past-assessments/. In this plugin checkout, the same script writes context/ when that folder is already there.

If docs/past-assessments/catalogue does not exist, and this checkout has no context/past-assessments, assess without published precedent. Say that the library has not been generated. Do not invent citations.

Before you score

1. Call search_past_assessments for this phase (discovery, alpha, beta, or live).
   • Pass point for each Service Standard point you own.
   • When the evidence looks thin, pass rating amber, red, or not-met.
   • When the evidence looks strong, also search green or met for the same point.
   • Keep at most 8 reports in total. Include at least one that failed a point you own and one that passed it, when both exist.
2. Call get_past_assessment with the file value for each report you will cite. Read that markdown. Stop at 8 full reports.
3. The search already drops the old 18-point standard and the four curated exemplars. Those exemplars (discovery-lead-stop.md, alpha-tech-spikes.md, beta-ur-fail-assistive-tech.md, beta-tech-pass.md) show how this plugin writes a verdict. They are not published panels.
4. If the tools are missing, read docs/past-assessments/catalogue/<phase>.json and then the matching file under docs/past-assessments/reports/, using the same cap of 8. In this checkout, the same files are context/past-assessments/<phase>.json and context/few-shot-examples/. Say that the MCP server was unavailable. If neither library exists, assess without precedent.

How to use what you read

Cite the published report when it bears on your decision: service title, phase, overall result, the point rating, and the source URL. Name the similarity in a sentence (same phase, same point, same kind of gap).

The evidence for this service decides the point. A published green rating does not clear a red line in context/rubrics/. In particular, red line B1 still marks Beta point 5 Not Met when no participant used assistive technology, whatever past reports did.

How published ratings translate

Write your own decision as Met or Not Met. Do not write green, amber, or red in the report.