Our record
We tell you how many bidders a contract will draw. This page shows how often that has been true, measured on real awards, misses included. Nobody else in this market publishes this.
How it's measured. Every prediction is archived before the solicitation closes. When the award later appears in the federal record, we look up how many bidders it really drew and write down what happened. What we said cannot be edited afterwards; the database refuses the change.
In terms of raw profit, what would winning an additional 6 to 20 contracts per year do for your bottom line?
1. Ten proposals
Write ten proposals picked off SAM.gov on your own and you might nail 2 or 3. Write ten from the ones we send you and you'll likely nail 6.
This is just to show you that your odds are far higher when we do the research for you.
And of the ones we tell you to leave alone, 9 in ten draw the crowd we said they would.
"Nail" means the notice drew 2 bidders or fewer, where a good bid has a real chance. The numbers behind that: of all posted notices, 25% actually drew 2 bidders or fewer; of the ones we flagged, 62%; of the ones we waved off, 93% drew a crowd. Measured in the federal award record on 1,416 notices we predicted before they closed (46 days on record): 324 we flagged, 769 we waved off. That is 1,416 of the 20,623 predictions archived so far; the rest stay unresolved until their award posts, and are counted neither way.
We can't guess at how many you'll win. But the same well-written, well-researched bid wins more often in a field of two than in a field of twenty. Finding the fields of two in your line, and putting them in your inbox before you write a word, is the work we do for you.
2. When we say a number, this is what happens
The odds printed on a row are not a model's opinion. They are what rows like it actually did, re-measured against this archive every night. The test of an honest number is that the two right-hand columns agree.
| Odds we print | Notices | We said | It happened |
|---|---|---|---|
| 0% to 20% | 769 | 8% | 7% |
| 20% to 35% | 230 | 26% | 24% |
| 35% to 50% | 93 | 41% | 44% |
| 50% to 65% | 191 | 59% | 59% |
| 65% to 100% | 133 | 68% | 67% |
Current fit: bootstrap, on 1,416 resolved outcomes (28 October 2026). A bootstrap fit scores the archive with the current model's history, which includes awards that posted after those notices closed, so it runs a little generous; the live fit replaces it once enough of this model's own predictions have resolved.
3. What we printed before 21 October 2026
Until then the odds on a row were the raw share of a buyer's past awards that drew two bidders or fewer. Checked here, that number ran about twice the truth: our confident calls (HIGH confidence, printed at 80% or more) came in quiet 48.8% of the time. The ranking was right; the number was not. The archive is append-only, so those rows keep what they said, and the table below counts them as printed at the time, with every miss.
| Confidence we gave | Checked | Right | Rate |
|---|---|---|---|
| HIGH | 750 | 394 | 52.5% |
| LOW | 239 | 117 | 49.0% |
| VERY LOW | 218 | 106 | 48.6% |
| MEDIUM | 210 | 69 | 32.9% |
"Right" here counts a call as quiet when the printed odds were 50% or better, and compares it with what happened.
4. The lab number, and why it is not the headline
Before the live archive existed, the only test was against history: train on past fiscal years, predict a year never seen, check. It counts a confident call right whether it said quiet or crowded, which on a crowded market makes it read high. It is a check that the method works, not a promise about a row.
Trained on FY2022-FY2027, on awards whose solicitation was publicly posted. Correct on confident calls, by what we had to go on: same buyer, category and industry code 88.6%; same buyer and category 77.3%; same category and industry code 71.2%; same buyer only 81.7%.
What changed on 21 October 2026, and why
The model learns from past awards. Until then it learned from every award with an offer count, most of which never had a public notice: sole-source orders and catalogue buys, two-thirds of them "quiet". That history pulled the odds on open competitions up. We measured both recipes against the live archive (1,416 resolved notices) before changing anything:
| Measured on the live archive | Every award | Posted solicitations only |
|---|---|---|
| Ranking quality (AUC, 0.5 is a coin flip) | 0.707 | 0.735 |
| Top tenth of our ranking that drew 2 bidders or fewer | 59% | 61% |
| Top fifth | 48% | 53% |
| Bottom half | 14% | 13% |
The posted recipe ranks better, so it is what runs now. And the printed odds are no longer the raw rate from either recipe: they are the rate at which rows ranked like this one actually came in quiet, refitted nightly on the archive above. The earlier experiment on this page (five years of history against eleven, September 2026) still stands: more history did not help.
One limit worth stating: the federal record carries a solicitation number on roughly 58% of definitive contracts, so a share of our predictions will never be matchable to an outcome. Those stay unresolved rather than being counted either way.
What we'll do for you, and how to get your first short list by reply.