METHODOLOGY

Methodology & trust

This page explains where evidence comes from, what counts, how duplicates are removed, how scores are calculated and how dossier status changes. You do not need to trust Timory. You need to be able to audit it.

§ 01

Seven public communities, read through eight collection feeds

Sources are not treated as interchangeable. A concrete feature request in a software issue and a founder's promotional launch post carry different evidentiary weight.

SourcePrimary roleWhat can count
RedditFirst-hand pain and spendingSpecific experiences, constraints and stated payments
Hacker NewsTechnical and operating contextConcrete problems and constraints, not broad opinions
GitHub IssuesFeature gapsMissing workflows; routine bugs are excluded
YouTubePractitioner comments under developer videosFirst-person problems and stated spending; praise and promotion are excluded
Software RecommendationsExplicit search demandRequests with clear requirements
Indie HackersFounder contextAttributable customer evidence; launch copy is excluded
MicroConfLong-form contextAttributable customer statements, not general advice

SOURCE POLICY

Source access and public quotation policy

Every public evidence card keeps a direct source link, while excerpt limits vary by source. This applies to the public site so traceability does not become republication.

TierSourcesPublic treatment
GreenHacker News / Stack Exchange / GitHub Issues / YouTubeOpen API or openly licensed sources; excerpts default to 200 characters with a source link.
AmberIndie Hackers / MicroConfConservative observation; excerpts default to 100 characters with a source link.
RedRedditPost title, evidence metadata, source link and at most a 50-character excerpt.

Source content and platform rules remain with their respective rights holders. Public availability is not treated as ownership, and no platform endorsement is implied.

§ 02

Provenance before volume

Every excerpt is classified by where the claim comes from. First-hand pain, current spend, explicit willingness to pay and concrete feature requests can affect confidence. Second-hand reports, expert opinion and promotional content remain contextual but do not raise the score.

current spend / explicit willingness to pay
first-hand pain / feature request
second-hand / expert opinion / promotional / unknown

§ 03

Recurrence means independent recurrence

Several excerpts from one post still count as one content item. Repeated posts by one author are discounted. Closely related signals are merged only when they describe the same underlying problem for the same kind of user. This prevents one loud voice from looking like a market.

The brand mark diagrams this rule: three matching cards overlap with increasing opacity, showing that the same thing must be stated independently more than once before it becomes evidence.

§ 04

A deterministic confidence score

confidence = strength × evidence × diversity × payment × competition × trend × 10

Each factor is calculated from stored evidence. The same evidence produces the same score. The number is a confidence measure for the underlying problem, not a forecast of revenue, market size or founder fit.

§ 05

Evidence level and recency are two separate labels

Evidence level says how far a need has been corroborated. Signal records an early pattern from a small number of independent items. Corroborated requires independent recurrence: ≥3 distinct content items across ≥2 source families or ≥3 distinct authors. Payment-backed additionally requires first-hand payment evidence — someone stating their own current spend or a specific amount they would pay.

Fading is not a fourth level. It is a recency tag shown alongside any level when no new evidence has appeared for ≥21 days; the evidence already gathered does not become weaker, it just stops being restated. Archived (≥45 days) preserves history and is not listed publicly. Routine active-tier recalculation is upgrade-only; a human audit correction can lower a level when supporting evidence is withdrawn. Payment-backed means payment evidence exists in the record — not that the business is worth building.

§ 06

Extraction and judgement are separated

The system first extracts short evidence excerpts, evidence type and source context. Code then deduplicates counts, calculates factors and applies status thresholds. The extraction layer does not decide which business is worth building, and the scoring layer does not invent evidence.

§ 07

What the evidence cannot tell you

IMPORTANT LIMIT

Payment-backed does not mean worth building

The dossier can support claims that a problem is real, recurring and connected to spending. It cannot establish total market size, customer accessibility, acquisition cost, implementation difficulty or your personal fit. Those decisions remain yours.

PRIVACY

Privacy

We collect an email address only when you subscribe to a brief, unlock full evidence chains or create a market watch. Page views are counted on our own server — no analytics or advertising cookies — and the site keeps short-lived access logs that include the raw IP address, so we do not describe this site as anonymous. Retention periods, the one functional cookie, and how to have your data deleted are all set out in full on the privacy policy. You can unsubscribe through the link in every email.

TERMS

Terms of use

Timory provides research evidence and monitoring, not investment, legal or business advice. Source content remains the property of its authors and platforms. Public excerpts are intentionally short and linked back to their source. Availability, scoring rules and early-access plans may change as the product develops.