Methodology
Scores are computed by rules. Research is web-grounded with mandatory citations. AI drafts prose only after the facts exist — and a mechanical QC gate keeps unsourced numbers from ever reaching you.
25 diligence items across seven categories — Corporate & Legal, Financials, Traction & Metrics, Market & Competition, Team, Product & IP, Fundraise Materials — each weighted 1–3 and mapped to the stages where investors expect it. Weight-3 items are deal-breakers: the ones that stall diligence outright (cap table surprises, missing IP assignments, deck ≠ model inconsistencies are the most-cited round-killers in published VC diligence guidance).
How many deal-breakers apply grows with the stage, because investors expect more evidence as you raise. The exact weight-3 count, straight from the rubric:
| Stage | Rubric items expected | Deal-breakers (weight-3) |
|---|---|---|
| Pre-seed | 13 | 5 |
| Seed | 23 | 8 |
| Series A | 24 | 9 |
| Series B | 22 | 9 |
The full weight-3 list for your stage renders live on the homepage checker and the stage playbook — both read from the same rubric this page describes, so the count above and those lists can't disagree.
What this rubric covers — and what it deliberately doesn't. These 25 items are classic operating diligence: the corporate, financial, traction, market, team, product/IP and fundraise-materials checks a generalist fund runs on almost every deal. It is intentionally not a substitute for deal-specific legal screens that a subset of companies need — AI training-data provenance and model-licensing, foreign-investment / CFIUS review on a cross-border round, sector regulatory approvals. Those are counsel's call, they turn on facts a checkbox can't capture, and they don't map to a stage weight. Where your company plausibly triggers one — AI-native, cross-border, regulated — the report flags it in the risks section and tells you to take it to specialist counsel, rather than pretending a generic rubric has cleared it.
Your answers are scored against the items expected at your stage — a pre-seed company isn't penalized for missing audited cohort tables. Score = weighted share of expected items you have. Letter grades (A+–F) exist because they communicate; the number rides along.
Why weight-3 is reserved for a short list: a deal-breaker here means an item that stalls diligence outright at that stage — a cap-table surprise, an unsigned IP assignment, a deck that contradicts the model. Items like a driver-based projections model or SOC 2 / security posture matter and can absolutely lose a specific deal, but they cost you speed and price rather than halting the process at the first meeting, so they carry weight 2 (and security only appears at Series A/B, where the enterprise motion makes it sharp). The intent is calibration to how a data room actually gets triaged, not a ranking of importance — the full item-by-item weights and stages are visible in the homepage checker and the report's rubric appendix, so you can audit every call.
Why the Team section is deliberately light — and it's not an oversight. At pre-seed and seed the team is the primary thesis, and a partner's real team diligence is judgment work: back-channel reference calls, reading co-founder dynamics, weighing a prior-founder track record, checking equity split and vesting for a future-blowup risk. None of that is checkbox-able — it turns on conversations and pattern-matching, not on whether a document exists — so scoring it 🟢/🔴 would be false precision. The rubric therefore scores only the team artifacts that are verifiable (founder-market-fit written down with references on hand, a hiring plan tied to the model, a documented option pool) and hands the actual team judgment to where it belongs: the IC-memo thesis section (which states the founder-market-fit case and flags where the reference evidence is thin) and your readout call. So a light Team count in the scored rubric is a deliberate line between what a rules engine can grade and what only a human should — not a claim that team matters less.
Your metrics are scored against stage- and model-specific target ranges an IC memo cites verbatim — ARR, YoY growth, net dollar retention, gross margin, burn multiple, CAC payback, magic number, Rule of 40 (plus GMV/take-rate for marketplaces). Ranges are institutional consensus for 2024–2026 from the sources below. For AI-native companies we shift the bar the way the same benchmark providers now split their own data — the AI-native cohort in these reports runs faster growth and thinner gross margin than classic SaaS, so we score against the AI cut rather than the blended SaaS cut (Bessemer's State of the Cloud and ICONIQ both now break AI-native out separately; that split, not a separate proprietary dataset, is what we apply). We show ranges, not point estimates, because 2025–26 dispersion — especially AI vs. non-AI — is the whole story; verify against the source reports before quoting.
In your report, each benchmark row names the specific report edition and carries a working link, so you can open the source and check the bar before you quote it back to a partner — the same discipline the market-sizing section uses. See it live in the sample report's benchmark scorecard.
Three research passes run for your company with a live, citation-returning research engine: (a) bottom-up market inputs — how many buyers exist, what they pay, published market figures with the source named; (b) competitive landscape including incumbents and the do-nothing option; (c) comparable financings 2024–2026 at your stage. Every figure carries a numbered source reference.
A driver-based model skeleton computed from your inputs (ARR, growth, burn, cash, raise): bear/base/bull ARR scenarios compounded quarterly against the next-stage bar, runway today and post-raise (assuming burn ramps to deploy the round over ~30 months), the trailing burn multiple, a stage-normal use-of-funds allocation with a ramped hiring estimate, and the median-dilution ownership waterfall. Every assumption is printed next to the number it drives so you can replace it with actuals — the model's job is to show the structure investors stress-test, not to predict your future.
An AI analyst drafts the prose sections — IC-memo thesis, risks & mitigants, diligence Q&A, process plan, market, competitors, comps, deck specifics, work order — from the scored gaps and sourced research, instructed to keep every number attached to its citation and to never invent figures. Where your case is weak, it is instructed to say what evidence would strengthen it rather than assert past it.
Before delivery, the report is mechanically checked: all sixteen sections present and substantive (nine drafted, seven computed), and the market and competitor sections must contain source references — a report without citations cannot ship. Failures route to a human (who also fields your readout call), and your 24-hour clock stays our problem.