A fit score that shows its working
Every notice is scored against each of your product lines. The score comes with the facts that produced it, the facts the notice did not give us, and anything that rules you out whatever the score. It is labelled fit, not win probability, because that is what it is.
In one paragraph
[ SUMMARY ]Most tender tools give a relevance percentage and hope you trust it. Defense Explorer separates three questions that a percentage blurs: how well does this fit, what do we not know, and is there anything that stops us bidding.
- 01Fit per product line, with the best line and the runner-up reported.
- 02Missing notice data scores zero and is listed as an unknown, never counted as a perfect match.
- 03Blockers such as a stated clearance you lack, an explicit NATO-only restriction or a passed deadline shown even at a score of 95.
- 04Deadline and status kept as separate fields, never folded into the score.
- 05The same scoring service behind the dashboard, digest, API and MCP, so every surface agrees.
How the score is built
[ MECHANICS ]Your company profile is made of product lines. Each line has keywords, CPV codes, domains and markets, and the profile can carry exclusions, delivery countries and adjacent categories. You can paste a website or capability statement and get proposed product lines, which are shown for correction and saved only when you accept them.
Five components
Fit is a weighted average of five components: keywords (30), CPV similarity (25), full-text relevance (15), domain overlap (15) and market (15). CPV similarity rewards an exact code most, then the same category, group and division. A keyword in the title counts for more than one in the body. A match on an exclusion term cuts the score sharply.
Unknown is its own state
A component is only assessed when your side of it exists: if you gave no CPV codes, CPV leaves the calculation rather than scoring zero. If the notice is missing the data, the component stays in and scores zero, and the gap is listed as an unknown. Unknowns include missing CPV codes, missing deadline, unknown status, unstructured eligibility text and inferred clearance. An inferred clearance is always an unknown, never a requirement.
Blockers
Blockers are facts that stand in the way regardless of fit: a clearance stated in the source when the profile holds no facility clearance; an explicit NATO-only or EU-only restriction your country fails (a NATO buyer name or an EU funding mention does not count); a passed deadline; or a notice that is closed, withdrawn or awarded.
Reasons that cite facts
Reasons name the product line, the matched keywords, the matched CPV pair and the buyer country that admitted the notice, so a colleague can check them in a minute.
Feedback that tunes, never hides
Mark a result relevant, not relevant or unsure. When at least three different notices marked not relevant share a keyword or CPV cue, and none sharing it was marked relevant, that cue is down-weighted for your company. It is never zeroed, and bid decisions in the capture pipeline never count as relevance feedback.
What a match looks like
[ API ]The same object the dashboard renders. See the [API reference](/docs/api/).
curl "https://api.defenseexplorer.com/api/v1/matches?minScore=70&limit=25" \
-H "Authorization: Bearer dx_api_…"
{
"match": {
"score": 92,
"level": "strong",
"reasons": ["capability fit", "clearance fit"]
}
}Where the score appears
[ 05 SURFACES ]- 01App
Match dashboard
Fit, blockers, unknowns, timing, per-line scores and relevance buttons on every row.
- 02Email
Daily digest
Top new fits since your last delivery, with blockers, and no win-probability claims.
- 03Server
Saved searches
Evaluated each morning with the same ranking, with new-since counts.
- 04Team
REST API
Matches with reasons for your CRM or reporting.
Read - 05Team
MCP
match_opportunities for Claude and other assistants.
Read
What it does not do
[ LIMITS ]- +Rank the whole candidate set before paginating
- +Keep unknowns and blockers separate from fit
- +Explain every score with the facts on both sides
- +Learn from repeated not-relevant feedback, per company
- –Predict whether you will win
- –Use semantic search yet: ranking is lexical (full text, CPV, domain, market)
- –Decide eligibility from a buyer’s name or funding source
- –Score beyond the first 3,000 prefiltered candidates per query
Questions
[ 04 ANSWERS ]01Is a score of 90 a strong reason to bid?
It is a strong reason to read the notice. Check the blockers and unknowns first, then the cited brief and requirements. Fit measures match on the available facts, not your chance of winning.
02Why did a notice with little information score low?
Because missing notice data scores zero and is listed as an unknown. Earlier versions of many matching tools, including ours, gave sparse notices full marks by default. We removed that.
03Can I have separate scores for different products?
Yes. Each product line is scored separately. The best line determines the headline score, the runner-up is reported, and per-line scores are visible on the dashboard.
04Does marking something not relevant hide similar notices?
No. Only a pattern across at least three notices, with no contrary relevant mark, down-weights a cue, and only to half weight. Nothing is suppressed on one click.
See which defence opportunities fit your company.
Set up a company profile in minutes and get a ranked, explained shortlist today.