Match on facts.
Read the reasons.
Goldstone crawls funding from official government and program sources, then scores each opportunity against your Vault. Hard disqualifiers run first. Every remaining match carries a per-component rationale you can open and inspect.
Deterministic. Explainable. Re-runnable.
Eligibility before fit.
Most platforms show you a long list and hope you sort it out. The Goldstone match engine applies hard disqualifiers first — if a program excludes your region, sector, stage, or check-size, it never enters your feed. Time spent on misses is time you don't get back.
- Region exclusions checked against your Vault
- Sector eligibility against your industry classification
- Stage gates against your maturity field
- Check-size range against your project use-of-funds
Components that sum to 1.0.
Surviving opportunities are scored on a small number of weighted components, each contributing a real fraction of the total score. Stage-adjacency logic catches near-fits cleanly. Re-running the engine gives the same answer — there is no stochastic step.
- Region · industry · stage · check-size each weighted
- Stage-adjacency for near-fits, never silent fudging
- Per-component rationale stored with every score
- Deterministic — same Vault, same answer
Official channels only.
The funding index is built from official government and program sources — Grants.gov, SBIR, NIH, NSF, CORDIS, and regional agency portals — through a normalization chain that deduplicates listings, extracts amount and deadline by regex (not estimation), and filters by keyword.
- Government APIs and program portals
- Normalization & dedup before anything reaches your feed
- Amount and deadline parsed, not guessed
- Source health monitored continuously
Click any score. See the math.
Every match opens to a breakdown — which components scored, what they were scored against in your Vault, and which disqualifiers were checked and cleared. There is no black box. If a result surprises you, you can see exactly why.
- Per-component fit shown numerically
- Vault fields that contributed are listed
- Disqualifiers passed are recorded
- Re-score on Vault change is logged
What the engine is not.
Matching is heuristic and explainable by design. There are no embeddings, no neural network, no learned model in the scoring path. That is a deliberate choice — trust comes from being able to read the math. AI is used elsewhere (for optional document polish), not here.