How FedGrade Forecast works

FedGrade Forecast predicts which agencies are likely to release solicitations soon, with a confidence score grounded in observed pre-RFP signals. It is the platform's predictive layer — distinct from the Brief, which is descriptive. Forecast lives under Radar at /radar/forecast and can be viewed on every tier.

What it predicts

A Forecast prediction is built from related signals pointing at a probable upcoming solicitation. Each prediction carries:

• the target agency
• a predicted title
• an estimated solicitation window (a start and end date)
• a confidence score from 0 to 1
• a rationale that cites the signal ids behind it

It does not predict the dollar value, vehicle or exact scope of the solicitation — only that one is probable within the window.

Signal types

The forecaster understands these signal types: pre-RFP briefings, sources-sought notices, industry days, NDAA markups, strategic plans and strategic-plan initiatives, budget justifications, recompete clocks, OMB memo mandates, Inspector General recommendations, and shifts in Senior Executive Service priorities.

Today the signals actually flowing in are sources-sought notices and pre-RFP briefings; the other types light up as their sources are connected.

How confidence is scored

Each signal is scored on named factors — agency velocity (how quickly that agency usually moves from signal to solicitation), specificity, fiscal-year alignment and vendor engagement. The baseline scoring is rule-based and deterministic. When a language model is configured, it refines the factor scores and rationale and the result is blended with the rule-based score (60% model, 40% rules). A model rationale that does not cite its signal id is rejected and the rule-based version is used instead.

Materialization

A prediction materializes when a real SAM.gov solicitation matching the predicted agency and signal type is posted within the predicted window. When only one of those matches it is recorded as partial; when the window closes with no match it is recorded as missed. The Materializations tab on /radar/forecast shows these outcomes, and the public calibration ledger scores them — if the forecaster is wrong, the misses are visible.

When it runs

Predictions are recomputed once a day, around 12:00 UTC. Outcomes are evaluated once a day as well. There is no manual recompile control.

The tabs

/radar/forecast has four tabs: Pipeline (the ranked opportunity pipeline — Team and above, see Opportunity ranking), Predictions, Signals and Materializations.

How it's used

1. Pursue early. Open a pursuit at /my-pursuits for a promising prediction and start collecting evidence before the solicitation drops.
2. Watch-driven alerting. Watches alert you when a new prediction matches your agencies, keywords or vendors. On Pro and above you can also + Track an individual prediction.
3. Digest. Team and above receive the twice-monthly Forecast digest.