Abstract

A method is disclosed for scoring funding opportunities against an organisation's capability registry with a deterministic, network-free, versioned engine, and for letting a language model lower a score only by a fixed amount and only on quoted evidence. Six weighted components (inverse-document-frequency-weighted keyword coverage with a concave recall lift, maturity-weighted depth, graded eligibility, award fit, deadline runway and application burden) are summed under exactly one hard gate, expiry, which zeroes the score; every other unknown is scored, not rejected, and marked estimated: unknown eligibility, award or deadline at a fixed middling value, a missing synopsis by capping coverage and skipping burden deductions, an unknown cost share with no deduction. A confidence value computed purely from the presence of four fields is never blended into the score and, inside the engine, only arms a ceiling. An ordered set of named ceilings (low confidence, partner-only eligibility, absent or thin coverage, aspiration-only maturity) then lowers the running score, each recorded only when it binds. A bulk path ranks sparse search hits with the same engine, no model and no persistence. An optional per-opportunity screen asks a model for a rationale and an optional quoted disqualifier, admitted only when the quote meets a minimum length and its leading segment occurs, case-insensitively, in the full stored source text; it then subtracts a fixed amount, floored at zero, recorded as a zero-weight component. Model failure leaves the deterministic score untouched, and stored scores carry the engine version in an append-only history.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS