Foray 285 — Testing an open-source project analytics platform

Can an open-source PDA platform make project-delivery analysis more useful and dependable than a straightforward table or calculation?

PW-285 · FORAY-285-PDA-PLATFORM · Established 7 November 2026.

The first essai applies antnewman/pda-platform to the existing fictional Winter Fleet case. This is a separate enquiry from Foray 450, which created and used the case to explore organisational change. No change to that enquiry or its retained results is implied.

Ends

Establish which delivery jobs PDA can perform accurately, which it makes easier, and where its data model or implementation prevents a useful answer. The desired result is a practical basis for adopting selected capabilities, improving a bounded integration, or declining to use the platform. Tool counts, attractive reports and technical execution alone do not establish that end.

The wider enquiry includes reliable arithmetic, preservation of project meaning, visible unknowns, correction and repeat use, and the amount of preparation needed. Practical benefit and human time savings require a later observed user trial; this essai does not invent those measurements.

Ways

1. Start with an existing management problem and frozen source data. State the questions and comparison before the PDA run.
2. Pin the open-source version and use its real MCP interface locally. Preserve native failures. Any diagnostic adapter must be identified and must not repair the analytical answer invisibly.
3. Translate only source-supported facts. Record what the translation preserves, loses and leaves unsupported.
4. Compare returned results with transparent reference calculations and existing case checks. Credit calculations to the component that actually made them.
5. Introduce an existing case change; inspect whether the changed answer remains faithful. Check save, retrieve, correction and restart where the chosen MCP tools support them.
6. Leave a readable worked result, exact calls, source hashes and reproduction instructions. The broader foray stays open beyond one case.

Means

• PDA Platform, MIT-licensed upstream, commit defd859819680f4bbc6341ae01a2b6e131a02d5d (live main verified 7 November 2026).
• Winter Fleet source commit 23e133b42ceaa60e7268a3eec96c704598ad7b3e: twelve weeks, three vessels, twelve people, 48 existing commitments and the first released event packet.
• An isolated local database, a genuine MCP client/server connection, Python reference arithmetic, retained machine-readable evidence and a local HTML reading entrance.
• Fictional source only. The original MCP run used no hosted service, external model-provider call or live client information.

First essai and current result

Winter Fleet through PDA is complete. The platform retained change and assumption records; native task/resource intake failed, and the exercised analyses did not supply the capacity/cash decision. A separate calculation found a feasible five-job candidate. No additional decision insight or human time saving from PDA was demonstrated.

The broader foray stays open. Later trials can examine different delivery questions, better-fitting source models or explicit repairs. A first result is not evidence that every PDA module succeeds or fails.

Download the reproduction package to inspect the fictional inputs, exact outputs and comparator. This public report contains no live PDA service and makes no changes to a fleet or project.