Opportunity ranking and the pipeline

The Opportunity Finder is the unifying capability. Forecast tells you what might be coming and when; pWin tells you how likely you are to win a given pursuit. Ranking ties them into the one number a BD lead acts on: which pre-RFP opportunities, in what order, deserve scarce capture dollars now. It is a Team+ capability.

How the rank score is built

The rank score is a transparent weighted product of four named components — never a single opaque number:

• pWin — the win-probability from the pWin engine, consumed read-only. The engine scores six weighted factors (incumbent strength, agency buying-pattern fit, vehicle/NAICS fit, size & set-aside fit, teaming / past-performance gap, competitive density). Competitive density currently carries zero weight: it is anchored to won edges in the entity graph, that graph has none yet, and we would rather show you a zero-weighted factor with its reason than quietly dilute the other five.
• value — the estimated contract-value band.
• proximity — how close the predicted solicitation window is.
• fit — how well the opportunity matches your workspace's profile (target agencies, NAICS, vehicle, set-aside).

Every component is inspectable and carries its own sub-rationale. Every ranked item shows its inputs with citations: the pWin factors that moved the win-probability, the forecast signal ids that set the proximity window, and the profile dimensions that matched the fit. A rank with no citation behind a component is a thin rank, and FedGrade says so.

Honest, not falsely precise

When pWin is insufficient-evidence or the value is unknown, FedGrade does not emit a confident rank. The item is graded insufficient-evidence and surfaced as "not yet rankable" — never buried in a falsely-precise ordinal. Underwriting a pursuit you can't score is the exact failure this discipline exists to prevent.

The ranking math is pure and deterministic, so the ordering is reproducible. No LLM is involved in the ordering itself. Forecast confidence is used only as a tie-break and a damper on proximity precision — never as a stand-in for pWin.

The pipeline

The ranked list is the pipeline. It glues three already-built surfaces together: the forecasted opportunities (Forecast), the latest pWin per opportunity (read-only), and a workspace profile derived pragmatically from your Cases' target agencies. An empty profile is handled honestly — fit reads as a gap rather than a fabricated match.

Export and reach

The pipeline exports as an executive board report (DOCX read-ahead memo or PPTX briefing deck) — see capture deliverables. It is also available over the API at GET /api/v1/opportunities (scope opportunities:read) and as the rank_opportunities MCP tool.

Where the engine score appears — and where it does not

The engine score appears in the capture assessment panel (the drawer on a library opportunity, and the panel on a Case), in the ranked pipeline on /forecast, over the API at GET /api/v1/opportunities, and on a Case after you run "↻ Recompute pWin".

It does not drive the pWin column on the /library/opportunity-finder table. That column is a lightweight demo heuristic over the notice's own text and vehicle — it is labelled "pWin (heuristic)" for exactly that reason, and it is no longer written onto a Case when you click Track. If you want a real score for an opportunity, open the row and run the capture assessment.

A pWin shown on a Case carries its provenance next to it: engine (the six factors above), imported (a value from your own upload), manual (typed by a teammate), or unknown source (recorded before FedGrade tracked provenance). Only engine is a model output.

Honest limits

Ranking is only as deep as its inputs. Early in a research library, more opportunities land "not yet rankable" because pWin or value is thin — the pipeline grows sharper as award and forecast data ingests. Treat the ordering as a prioritization aid, not a guarantee.