Methodology

Founder's AI Blueprint publishes original research on how AI-native founders actually build. This page is the publication-wide methodology — the editorial principles, source typology, and verification disciplines that every issue is built against. Per-issue notes (sample frames, specific exclusions, data-quality flags) live as appendices attached to each issue.

---

Editorial principles

Sample sizes are disclosed inline. Every quantitative claim in an article cites the n behind it. A number with n=8 reads differently than the same number with n=200, and both deserve to be visible at the point of claim, not buried.

Methodology is published per issue. Every issue ships with a notes appendix covering its sample frame, classification rules, exclusion lists, and any data-quality flags worth surfacing. The appendix is linked from the article itself.

Findings are descriptive of the universe we measured at the time we measured it. We do not make causal claims from cross-sectional data. We do not generalize beyond the segments observed. We disclose when a finding is directional versus when sample size and verification support a stronger claim.

Corrections are public. When a source flags an error or when a number moves materially, we update the issue and note the change. The original is preserved in the repo for diff.

---

Source typology

Every issue draws on a combination of the following:

Public datasets. PitchBook, Crunchbase, Carta state-of-private-markets reports, and equivalent industry data. Used for funding distributions, deal counts, valuation benchmarks, and investor activity at the aggregate level.

Founder surveys. Our own structured surveys, fielded to AI-native founders. Used for choices the public datasets cannot measure: model selection, eval discipline, pricing models, hiring decisions, infrastructure tradeoffs. Survey methodology (response counts, segmentation, anonymization) is disclosed per issue.

Engineering blogs and stack disclosures. First-party posts from companies whose stack decisions are themselves the data point. Cited inline; treated as illustrative rather than statistical.

Live pricing and product documentation. API pricing pages, model documentation, infrastructure pricing — fetched at publish time and re-verified before each major edit.

Strongest published work in the field. Synthesis pieces, YC playbooks, Stratechery, Lenny's Newsletter, Epoch AI data insights, and similar. Cited inline, never used as a stand-in for primary data.

---

Verification disciplines

Cross-source verification. When a named claim (investor lead, founder background, deal terms) matters, we verify against at least two independent sources — typically the primary dataset plus a public announcement, press article, or company disclosure. Discrepancies are resolved against the source closer to the original event.

Exclusion is disclosed. When companies or rounds are excluded from a named analysis (data conflicts, ambiguous classification, etc.), the issue lists them by name with the reason. Inclusion criteria are written before the analysis runs.

Numbers are reproducible. The merged datasets, classification rules, manual override lists, and chart-stats outputs for every issue are reproducible from the documented exclusion lists and analysis scripts. Random seeds for bootstrap confidence intervals are fixed.

Publish-time re-verification. Before each major content change, the publish-time verification checklist runs on every cited number that could have moved (model names, API prices, headcount, revenue per employee, etc.).

---

What this work does NOT support

The disciplines above bound what we can claim:

• No causal claims from cross-sectional data. Companies with higher valuations may have stronger fundamentals or better fundraising operators; the data we publish cannot distinguish.
• No prospective claims beyond the measured window. An issue describing 2023–Q1 2026 funding does not predict Q2 2026 or later.
• No global generalization when the universe is curated or geographically weighted. Our universes are documented and disclosed; findings describe the segments we observe.
• No per-founder identification in pattern findings unless the individuals have made the information public themselves. Aggregate patterns yes; individual attribution no.

---

Issue 001 — analytical notes

These two notes carry the reasoning behind claims in Part 1 of the Issue 001 brief. They live here, not in the article, because the article is written to be acted on; the article states each conclusion in one line and links to the note below it.

Why the AI stage ladder is this clean

The 4-5× step-up pattern looks unremarkable until you compare it against historical software stage-ladder data, where the spread is choppier and the medians less predictable round-over-round. A regular geometric step-up signals that the market is in equilibrium for AI rounds — not panic, not bubble, just consistent investor expectations about what each stage justifies. For founders planning a raise, this means the round-size question has an unusually defensible answer: budget for the next stage's median plus your sector-specific premium, and the data will back you up. For founders who haven't started, it means the ratchet from $3M to $15M to $64M is structural; the path is the path, and the gaps between stages are too large to skip routinely.

On the Carta Series A comparison

Comparing our AI pre-money against Carta's combined-market post-money would be partially self-referential, because the combined market already includes AI. The cleaner comparison — and the one the article uses — is against Carta's published AI segmentation directly. Carta also reports that the contemporary software seed-to-A 24-month graduation rate has dropped from ~30% (Q1 2018 cohort) to ~15% (Q1 2022 cohort), an industry-wide phenomenon not unique to AI, and the reason the article recommends planning 18-24 months of runway.

---

Per-issue methodology

The Issue 001 methodology appendix — sample frame, AI-native classification, sub-sector tagging decisions, investor lead verification, data quality flags, and the full source list — is at /methodology/issue-001, and linked from the bottom of Issue 001.

Questions about a specific number, suggested corrections, or notable discrepancies between our findings and your own data: reach out via the survey link in the article closing.