Methodology
What the scores mean, how they are computed, and — just as importantly — what they do not claim.
What this is not
Capital Candor does not predict whether an investor will fund you. Nothing here is calibrated against fundraising outcomes, because we do not have outcome data, and a product that dressed up heuristics as prediction would be doing the thing this one exists to avoid. What the engine does is narrower and checkable: it compares your round against researched, sourced facts about each firm, and tells you how strong the overlap is and how much the underlying evidence can bear.
Fit and Confidence are two numbers, never one
Fit answers “how well does this firm match this round?”Confidence answers “how much do we actually know about this firm?” They are reported side by side and are never multiplied together, because collapsing them hides which one is weak. A well-evidenced mediocre match and a thinly-evidenced strong one are different problems and deserve different responses.
Confidence can only ever lower a recommendation, never raise one. A high Fit score with low Confidence does not reach the top tier — the evidence has to support the recommendation, not just the number.
The seven dimensions
Fit is a weighted average of seven dimensions. The weights are a documented judgment call, versioned as fit-weights-v6 so any change is traceable and old scores stay comparable to the rule that produced them.
| Dimension | Weight |
|---|---|
| Sector / thesis overlap | 22% |
| Funding stage | 17% |
| Check size | 17% |
| Recent investment activity | 17% |
| Portfolio similarity | 11% |
| Lead vs. follow | 11% |
| Geography | 5% |
Missing evidence is never scored as a negative
If we do not know a firm's typical check size, that dimension is marked unknown and excluded from the average — it is not scored zero. Not knowing something about an investor is a fact about our research, not a fact about the investor. When more than half of the total weight is unknown, no score is produced at all and the firm is reported as insufficient evidence rather than given a number that would look like knowledge.
The same principle governs freshness: evidence has a verification date, and stale evidence caps how much confidence it can carry.
Tiers
Scores are grouped into bands so the output is a decision rather than a leaderboard. The thresholds are 80 for the top tier, 65 for Worth Pursuing, 45 for Research Further, and 25 below which a firm is flagged as a likely mismatch. Confidence is then applied on top and can move a firm down a tier, never up.
Where the data comes from
Every firm in the dataset was researched from public sources — firm sites, fund-close announcements, regulatory filings, and press — with each material claim recorded against the document it came from and graded by that document's reliability. A firm's own statement of its thesis and its observed investing behavior are stored separately, because they frequently disagree, and when sources conflict we record the conflict rather than quietly picking a winner.
The public directory shows firms with at least 2 distinct sources and 3 portfolio companies on record. Firms below that bar are still scored for signed-in founders, but are not published as browsable research, because a page carrying a name and little else is not research.
Rankings are never sold. No firm can pay to appear, to rank higher, or to have a concern removed.
What we are still missing
The weights have not been validated against how experienced investors and founders would actually rank these firms. Until that validation exists, the weights stay where they are rather than being tuned on intuition — tuning a scoring model by feel is how it starts describing its author instead of the market.