HomeGuides › Sourcing

Where citable market claims come from

Sourcing · ~8 min read · public-methodology

The single fastest way to lose a room is a number nobody can trace. "The market is $9 billion" invites exactly one question — according to whom? — and if the answer is a chatbot or a broker's press release, the whole analysis is done. Every defensible figure comes from a public source you can name, open, and date. Here's the hierarchy, ranked by how much trust a figure inherits from its source.

The source hierarchy, by trust

Sources are graded 1–5. Higher-trust sources are primary issuers of fact; lower ones need corroboration before they anchor a headline number.

Source classTrustUse it for
Official statistics
Census/BLS, Eurostat, OECD, KOSIS, UN Comtrade, trade associations
5Unit counts (facilities, firms, vehicles), spend baselines, penetration rates
Regulatory filings
SEC/EDGAR 10-K/S-1, Korea DART, central-bank data
5Revenue, segment sizes, pricing signals, competitor scale of listed players
Standards & regulators
ISO/IEC, IEEE, SAE, 3GPP, NHTSA, FDA, EU regulations, MOTIE
4Adoption mandates, technical inflection points, regulatory tailwinds/restraints
Freely published research
NVCA public releases, a16z/Bessemer notes, government white papers
3Framing, growth narratives, directional CAGR — corroborate, never cite alone for a headline
Company disclosures
Public pricing pages, investor decks, earnings-call transcripts
3Price-per-unit assumptions, funding, product capabilities
Named-byline press
Reuters, Bloomberg, FT, TechCrunch (named reporter + date)
2Deal facts, timeline events — corroborate against a primary source where possible

What's deliberately absent: paid, licensed research (the big syndicated-report firms). It's copyrighted, its methodology is opaque, and reproducing its numbers isn't defensible — so it's never cited or reproduced. Everything above is public and license-free.

The evidence score

A report's sourcing isn't just pass/fail — the mix of source tiers rolls up into a single evidence score, so a report leaning on trust-2 press reads differently from one built on trust-5 filings.

evidence score = mean(trust of filled slots) ÷ 5 × 100 example: slots sourced at trust 5, 5, 4, 4 → mean 4.5 ÷ 5 × 100 = 90 / 100

The gate: no source, no ship

Discipline only counts if it's enforced. Every report has a fixed set of required citation slots — the parent market, the unit count, the unit price, the value pool, the CAGR, and a regulatory driver per region. Each must carry a real public source, or the report cannot pass QC.

required slots filled: 6 / 6 → policy PASSED required slots filled: 5 / 6 → BLOCKED — "unit price" uncited

An open slot is surfaced to the analyst, not silently dropped. A claim without a public source doesn't ship — that's the product's spine, and it's the difference between a report and an opinion.

Primary over secondary, always. Cite the original issuer, not a blog re-quoting it. A trade body's own statistic outranks a news article about that statistic — and when you can only find the secondary source, say so, and go one more click to the primary.
Honest caveat. A source being public doesn't make it current. Every figure is re-derived to the report date, with units and currency explicit — two figures are never silently mixed. A dated 2022 statistic carried into a 2026 report unchanged is a sourcing failure even if the citation is real.

Every figure in your report, sourced

Commissioned reports ship with a sourcing appendix: every required figure, its public source, and the evidence score.

See the sourcing ledger → Commission a cited report