Our method
How SeedBasis researches an idea, grades its sources, states its confidence, and reports what it could not find.
Last updated 19 October 2026 · Markdown version
Every report follows the same rules, whatever the idea. They are written down here so you can judge the research, not just read it.
Every claim is a finding
A finding is one specific claim with the evidence behind it: the exact quote, the source it came from, how reliable that kind of source is, and how confident we are. Charts and tables are built from findings, so any number you see links back to where it came from.
How confident we are
| Confidence | What it means |
|---|---|
| Verified, 90 to 100% | Stated directly by a primary source: named documentation, published terms, a filed record, or a plain statement by the party involved. |
| Strong evidence, 70 to 89% | Strong indirect evidence: several independent signals agree, and nothing contradicting was found. |
| Moderate evidence, 50 to 69% | Reasoned from partial evidence. Other explanations are possible and were not ruled out. |
| Weak evidence, 30 to 49% | Thin evidence, flagged wherever it appears: a lead to check before relying on it. |
| Unknown, under 30% | Below the bar for stating anything: the report says it is unknown and what would settle it. |
Confidence describes the evidence behind a finding, not a promise that it is right. A finding with high confidence can still be wrong.
Not all sources are equal
| Tier | What it is | Examples | Highest confidence |
|---|---|---|---|
| T1 | A primary statement by the responsible party | Official API documentation, regulation text, company filings | Up to 100%, Verified |
| T2 | A structured record of facts | Package registries, trademark and court databases, archived web snapshots, a company's own published prices and terms | Up to 85%, Strong evidence |
| T3 | A public claim someone makes about themselves | Marketing pages, testimonial counts, vendor comparison pages | Up to 60%, Moderate evidence |
| T4 | Community signal | Forum threads, reviews, social posts | Up to 50%, Moderate evidence |
A claim is only as strong as the weakest source it rests on. A number that comes from a company’s own marketing page can never read above 60%, Moderate evidence, however confident the wording. The only way to raise it is to find a better source. What a company publishes about what it charges is read as its record (T2), since it is the one place that states its prices; what it says its product achieves stays its own claim (T3).
Estimates show their working
- Every estimate is labelled as one and shows its arithmetic or reasoning where the number appears.
- An estimate is never presented as a sourced fact.
- Guessing an audience from how a website looks is a hypothesis, and is labelled as one.
- Finding nothing is reported as finding nothing. “No complaints found” does not mean nobody has the problem.
When sources disagree
We never pick one silently. Both claims are shown with their sources and tiers, the stronger source is named, and confidence goes down, because disagreement is information about uncertainty.
What we could not reach
Every area of a report shows how many sources it reached out of how many it tried, and which ones failed. A claim whose source could not be reached is left out, not weakened, so a thinner answer is never passed off as a complete one.
Can you actually build it?
Feasibility is judged against the access you actually have, not what a well-funded company could get. The idea is broken into the capabilities it depends on, and each one is classified by the access it needs.
| Access | What it means |
|---|---|
| Public | Documented and available to anyone. |
| Public, with limits | Available, but rate limits, costs or terms materially shape the product. |
| Gated | A third party decides whether you get it, such as a partner program, licence or certification. |
| Private | It exists, but is not exposed to outside developers. |
| Nonexistent | No interface exists at all. |
| Unsolved | Nobody can reliably do it yet. |
| Unknown | We could not determine it, and say what to check first. |
Gated is hard. Private and nonexistent are unavailable: no amount of effort or money changes them. The overall verdict is worked out from the capabilities by a fixed rule, so a product whose core depends on something unavailable is never softened into “challenging”.
Which way we lean
A wrong “no” can kill a good idea, and nobody ever finds out. A wrong “yes” can cost months. We lean toward flagging uncertainty rather than resolving it either way, so “unknown, check this first” is an answer you will see.
Nothing generic
Every step in a plan must cite the findings it came from and name something specific from them. A step that could have been written without researching your idea is removed automatically. If a plan comes out with four steps, it ships with four steps.
Nothing about you is assumed
Any claim about you must rest on something you told us during onboarding. If you did not tell us, the report does not say it.
What this does not replace
Reports are research, not professional advice. The legal and capital sections show what to check and who to ask; they are not legal, financial, tax or investment advice.