Which extra jobs can fit?
Keep every old promise, use qualified crews and stay above a £6,000 cash floor. The first change reduces two boats’ availability in week 5.
Foray 285 · Essai 01 · 7 November 2026
Can PDA spot the fleet’s problem?
An open-source project analytics platform meets Winter Fleet: 48 existing promises, three boats, qualified crews and seven new offers.
This page presents the recorded trial. The PDA MCP server ran locally; this public report does not run it.
A useful record keeper; an incomplete fleet analyst.
PDA retained the proposed jobs and a changed availability assumption. The decisive boat, crew and cash answers came from the separate reference calculation.
This result concerns one fictional case and the tested platform version. It establishes neither human time savings nor operational readiness.
Two weeks are overbooked
Taking every offer needs 5.5 Morrow days in weeks 5 and 8. Only 4 and 5 days are available. Its skipper is also over capacity.
Changes and assumption drift
Seven proposals remain unapproved. Morrow’s 5→4 days is retained as 20% drift. PDA does not produce the required capacity or cash answer.
Ordinary reference · independently written arithmetic
Walk through the same case
This changes the reading view only. It does not change source data, PDA records or any fleet commitment. All amounts below are conditional forecasts.
Morrow and its skipper at the two pressure points
| Week / resource | Demand | Available | Loading | Result |
|---|
Weekly cash, after costs and receipts
Read the cash values as a table
| Week | Receipts | Costs + overhead | Closing cash |
|---|
The original forecast matches all twelve cash rows and every positive resource-demand row in the retained case witness. This checks the fictional arithmetic, not real vessel operations.
Observed through the real local MCP connection
What PDA actually returned
| Job | Observed result | What it means |
|---|---|---|
| Load the task windows and resource plans | Native intake failed | Project import and resource-plan writes did not complete. The original failures remain in the evidence. |
| Detect an overloaded boat | 5.5 days returned; available days not compared | A separately disclosed store seed reached the reader. It aggregates effort; it did not give the 137.5% demand/availability answer. |
| Retain the capacity change | 5 → 4 days; 20% drift | The same assumption ID and changed value survived a server restart. The case supplied the observation. |
| Keep proposed work separate from approval | Seven submitted changes | The changes stayed unapproved after restart. PDA correctly summed the supplied £15,050 cost impacts. |
| Forecast cash | Unavailable in the exercised financial model | A cost estimate does not include opening cash, receipt timing or a cash floor. The £81,550 cost forecast was supplied by our reference calculation. |
| Repeat a ledger save | £6,375 became £12,750 | A separate hypothetical replay showed append behaviour. The same entry must not be retried as though it were an idempotent save. |
Interpretation limits and wording hazards
The board report omitted supplied change costs. An INITIAL baseline’s confirmation called it approved, despite the stored label remaining INITIAL. Seven unknown schedule impacts were totalled as zero. An empty benefits register returned a health score of 1.0 beside a no-benefits message. These outputs require careful interpretation.
The red-flag scanner did explicitly name missing data. No risk likelihood, benefit target, daily timetable or permission was invented to make the case fit the tools. No external AI provider was called, so provider narrative quality is untested.
The separate enquiry
Ends, ways and means
End
Find which delivery jobs an open-source PDA platform can improve reliably, and whether the improvement justifies the preparation and integration effort.
Ways
Use an existing case; keep its meaning; compare exact platform outputs with a simple reference; replay a change; retain failures, unknowns and restart evidence.
Means
Pinned PDA code, Winter Fleet’s fictional source, an isolated local MCP connection, transparent calculations and inspectable evidence.
Winter Fleet is the first application. It remains Foray 450’s case; Foray 285 tests the platform. Later trials may examine a conventional project schedule or risk register. Download the reproduction package. Wider usefulness and adoption remain open.
Read the full foray promptInspect rather than take it on trust
Evidence and reproduction
A passed evidence check confirms a recorded result, including a failure. It does not turn that behaviour into a successful PDA capability. One unknown-tool request was a caller mistake and is excluded from platform findings.
Full findingsProspective protocolMapping and omissionsReproduceOriginal dossierReference dataExact MCP records