Methodology

Last updated October 1, 2026

GovInbox answers questions about federal contract opportunities from its own database. This page says where that database comes from, which parts of an answer are recorded facts, which parts are model estimates, and what the data cannot prove. GovInbox is not affiliated with the U.S. government, GSA, or SAM.gov.

Data provenance

GovInbox maintains a database derived from cited public federal procurement records. It is not GovInbox's own history. The underlying records are public U.S. government data, ingested under the sources' public-use terms:

• SAM.gov Contract Opportunities API (U.S. General Services Administration): notices, refreshed every 4 hours.
• USAspending award records: prime contract awards, used for award history and comparables.
• USAspending subaward reports: first-tier subcontract reports, used for the prime-to-subcontractor relationship records.
• SAM.gov exclusions list, SBA size standards (13 CFR 121.201), GSA CALC labor rates, Simpler.Grants.gov postings, DOGE-reported terminations: used where noted in specific answers.

Every answer carries as-of dates. Corrections to source records are picked up on the next ingest cycle; records that have gone stale past the freshness window are suppressed or flagged instead of shown as current. GovInbox does not call any third-party API at request time; every tool answers from the ingested database.

Facts vs. estimates

Every structured answer separates two sections:

• source_fact: something a cited record states (a notice id, a deadline as published, an award value). Each fact ships with its provenance: source system, official record id, record version, and retrieval time.
• govinbox_model_estimate: something GovInbox computed (a recompete forecast, an estimated solicitation window). Each estimate ships with its basis and a plain statement of what would prove it wrong.

Nothing in any answer is a recommendation. GovInbox does not produce bid/no-bid verdicts, prime/sub/pass recommendations, winnability scores, win probabilities, or eligibility determinations.

Recompete forecasts

Predicted event: a federal contract whose period of performance is ending will be re-solicited (recompeted) within the forecast window, and the named prior award is the contract being replaced.

How it is computed: a nightly job matches recent notices to past set-aside awards with the same NAICS from USAspending and scores each candidate 0-100 on: NAICS match (15), PSC match (15), agency match (20), set-aside match (10), timing of the award end date relative to the notice (20), place of performance (5), title overlap (10), and recompete keywords in the title (5). An award that started after the notice was posted is never a candidate. Confidence tiers: high (score 70+), medium (55-69), low (20-54). "Likely recompete" means high or medium.

Validation status (honest): no out-of-sample backtest has been published yet. The score weights and confidence tiers are rule-based heuristics, not calibrated probabilities. A forecast is GovInbox's estimate, not an agency announcement; every forecast row is labeled "no solicitation observed." Publishing backtest precision/recall is planned before the forecast tool is submitted for review.

Aggregates

Every aggregate in an answer ships with its basis:

• basis_n: how many records the number was computed from.
• Observation period: the date range of the underlying records.
• Field definitions: exactly which fields were counted and how.
• Exclusions: which records were left out (e.g. records with missing dates).
• Missingness: how much of the underlying data was absent.

This applies to buyer-behavior figures (office cadence, median response window, amendment/cancel rate), value benchmarks (p25/p50/p75), and competition signals (median offers, single-offer share). An aggregate with too few records to be meaningful is reported as "not observed" instead of a number.

Subaward relationship records

Prime-to-subcontractor records come from USAspending first-tier subaward reports. The honest verb is "reported first-tier subcontract award amounts": these are subcontract award amounts as reported on the filing, not proof of cash paid. The reporting threshold follows the current FAR framework ($40,000 per FAR 4.1403); reports below the threshold are not required, so coverage is incomplete by nature.

Two caveats are stated on every answer: the NAICS code on a subaward report classifies the prime contract, not the subcontract work; and absent a report is recorded as "not observed", never as proof of no relationship. Evidence tiers describe the record only (subaward_reported, naics_overlap_only, sources_sought_text_present) and are never merged.

Size standards

Size-standard values come from a curated subset of the SBA Table of Size Standards (13 CFR 121.201). GovInbox records the standard that applies to a notice's NAICS as a fact and records a company profile's size fields exactly as supplied. It does not determine whether a firm meets a size standard; that question is listed under facts requiring authoritative verification.

What the data cannot prove

• Whether a forecasted recompete will actually be solicited, or when.
• Whether a company is eligible for a set-aside (certification status, size, and ownership are determined by SBA and the contracting officer, not by GovInbox).
• Whether an incumbent will bid again, or whether a prime needs a subcontractor.
• Anything about state, local, or grant opportunities beyond the federal contract sources listed above.

Open questions are listed on every brief instead of being filled in with guesses.

Changes

If this page changes in a way that matters, the date at the top changes with it.