How Dossiary knows
what it knows.

last updated 2026-09-06 · this page changes when the pipeline changes

Dossiary is a record of creators, built from evidence. Every stored number answers three questions: where it came from, when it was observed, and how sure we are. This page explains how, in plain sentences you can quote.

Where every number comes from

Every important fact in the record answers three questions. The source names where the value was read — a platform API, a licensed vendor, or a collector we run. The date we checked says when it was read; nothing pretends to be current. The confidence says how much you should trust it — high, medium, or low. A fact missing any of the three does not enter the record.

Where the data comes from

  • Telegrama collector we host ourselves, on Telegram's own APIs.
  • YouTubethe official YouTube Data API.
  • Instagram and TikToklicensed commercial data vendors, swappable per record. The vendor is stamped into each reading's source.
  • X, Facebook, Snapchatlinked handles and basic stats only, for now.

The record holds public information: public profiles and their public counts. Classification — a creator's category, or how strongly an account is affiliated with Nigeria — is machine-inferred and always labeled with its own, usually lower, confidence.

What the confidence levels mean

  • highRead directly from a platform API or a collector we run, parsed without ambiguity.
  • mediumFrom a licensed vendor's read of a public page, parsed cleanly.
  • lowInferred — a category guess, an affiliation score — or aged past its freshness window.

How often we re-check

Each profile shows when its numbers were observed. Signed-in profile views can request a refresh when those numbers pass the workspace plan's freshness window. Published profiles also receive routine checks. Anonymous views do not trigger collection. Requests wait for available sources and collection capacity; the displayed date changes only when a new observation arrives.

Basic and full profiles

A creator we've just found starts as a basic profile: handle, followers, what they post about, and when we last checked — shown with honest, low confidence. A creator we go deeper on becomes a full profile: what they charge, how to reach them, who they've worked with, with identity, contact details, rates and audience each checked separately and each carrying its own age. Search covers both, and which one you're looking at is always visible.

Where humans stay in the loop

Two things never happen automatically. Merging two creator records into one waits for human approval, and both histories are preserved. Attaching a regulated category — sports betting, for example — also waits for human approval. Machine suggestions queue for review; they do not publish themselves.

How rankings work

Top creators lists rank one niche on one platform. Each creator scores out of 100, and the score is built from three things we can see: reach (40% of the score), engagement (40%) and how often they post (20%). The weights below are read straight from the code that ranks, so this page cannot describe a formula the product is not using.

  • 25% · FollowersHow many people follow the account, on a log scale so the biggest accounts cannot flatten everyone else.
  • 15% · Average viewsAverage views per post in the period — what the account actually reaches, not what it could reach.
  • 25% · Engagement rateAverage likes, comments and shares per post divided by followers. 10% or better scores full marks.
  • 15% · Average engagementAverage likes, comments and shares per post — the size of the response, not just its rate.
  • 20% · Posts per monthPosts in the period scaled to 30 days. Posting more than once a day earns nothing extra.

Two rules bind the score. A number we have never observed is left out of it — it is not read as zero, and its absence never costs a creator a place; only the numbers we actually hold are averaged. And older observations count for less: the score keeps half of what freshness is worth every 60 days and never falls below 50% of the merit behind it, because a stale fact is still a fact.

A creator only joins a list once we have seen what their posts did in the period. A follower count on its own earns no place — high or low — because it says who might see them, never what happened when they posted. Being left out is not a low score; it is us saying we cannot compare them yet.

Nothing else enters a position. Not payment, not a claim, not a membership — the top of every list is public and free to read, and the rest of the list is the only thing a membership buys.

Audience signals

A profile's Audience signals section says what can honestly be said about an audience from numbers the record holds for the period. Each signal is placed in a band — typical, above, below, or unknown — against a reference range printed below, with the numbers it used, how many posts it read, and one sentence saying how. Dossiary does not estimate fake followers or a credibility score from public counts and posts. Audience demographics are separate supplied reports, labeled by collection method with their own source and date. An absent report is unknown. There is no overall score, and nothing here enters a ranking.

  • Engagement for the sizeAverage likes, comments and shares per post divided by followers, placed against the reference band for this platform and follower size.
  • Followers to followingFollowers divided by the accounts this one follows, from the latest observation; under 1 is the unusual shape for a creator.
  • Comments per likeComments divided by likes, summed across the posts in the period where both were observed.
  • Short-video reachAverage views on short videos in the period divided by followers, against the reference band for this platform.
  • Posting rhythmThe spread (standard deviation) of the gaps between consecutive posts in the period, in days; the band compares that spread with the average gap.
  • Posts with numbersThe share of posts in the period on which at least one engagement number was observed — how much of the record the rows above are reading.

A band is a comparison, not a grade. An unknown band means an input was never observed — followers, the following count, or fewer than 5 posts with engagement numbers in the period — and it is never read as below. The ranges below are read straight from the code that reads a profile, so this page cannot describe a band the product is not using.

Engagement rate that is typical for the size

  • Instagram · under 10K2.0% – 8.0%
  • Instagram · 10K–100K1.0% – 5.0%
  • Instagram · 100K–1M0.70% – 3.0%
  • Instagram · 1M and over0.50% – 2.0%
  • TikTok · under 10K6.0% – 18%
  • TikTok · 10K–100K4.0% – 12%
  • TikTok · 100K–1M3.0% – 9.0%
  • TikTok · 1M and over2.0% – 6.0%
  • YouTube · under 100K2.0% – 8.0%
  • YouTube · 100K and over1.0% – 5.0%
  • Facebook · any size0.30% – 2.0%
  • X · any size0.20% – 2.0%

Short-video reach that is typical

  • TikTok10% – 200% of followers
  • Instagram10% – 100% of followers
  • YouTube10% – 200% of followers

The other ranges

  • Followers to followingtypical from 1× upwards; under that, the account follows more than follow it
  • Comments per liketypical between 0.50% and 10%
  • Posting rhythmtypical when the spread of the gaps is between 0.5× and 1.5× the average gap
  • Posts with numbers80% of posts or more; below that the other rows say they read a thin sample
  • Minimum posts5 posts with engagement numbers in the period before a post-based signal reads a band

Where the ranges come from: they are starting conventions, set by hand from the ranges industry benchmark reports published for 2024–2025 by platform and account size. They are not a measurement Dossiary made, and they will be replaced by the record's own quartiles once it holds enough accounts per platform and size. A platform with no range here — Telegram, for one — gets no comparison at all, never a guessed one.

What we refuse

Top spots are never for sale; paid visibility is labelled and never moves a position. Unknown values are never rendered as zero and never penalized. And we publish no invented trust signals — no fabricated testimonials, no padded counts, no fake-follower percentage, no credibility score, no demographics inferred from a creator’s own profile. Where this page names a number, that number has a source.

Questions we get

Where does Dossiary's data come from?
From platform APIs where they exist, from licensed commercial data vendors where they don't, and from collectors we run ourselves. Every reading is stamped with its named source. The record holds public information: public profiles and their public counts.
What do the confidence levels mean?
High means the value was read directly from a platform API or a collector we run, and parsed without ambiguity. Medium means it came from a licensed vendor's read of a public page, parsed cleanly. Low means it is inferred — a category guess, an affiliation score — or the value has aged past its freshness window.
How fresh is the data?
Each profile shows when its numbers were observed. Signed-in profile views can request a refresh when those numbers pass the workspace plan's freshness window. Published profiles also receive routine checks. Anonymous views do not trigger collection. Requests wait for available sources and collection capacity; the displayed date changes only when a new observation arrives.
What happens when Dossiary doesn't know something?
It renders as unknown. Unknown is never displayed as zero, and it is never treated as a penalty when creators are compared.
Can a creator pay to rank higher?
No. Top creators are ranked on merit and never for sale. Any paid visibility must be clearly labelled placement next to results and cannot change a ranking position.
How is the Top creators score worked out?
Each creator scores out of 100 on three things: reach — followers and average views — is 40% of the score, engagement — the rate and the size of the response — is 40%, and how often they post is 20%. A number we have never observed is left out of the score entirely; it is not read as zero and it never costs a creator a place. Numbers checked longer ago count for less, halving every 60 days down to a floor of 50%. A creator only joins a list once we have seen what their posts did in the period: a follower count on its own earns no place, high or low.
Where do audience demographics come from?
Audience reports can contain countries, cities, age groups, gender, languages and interests supplied by platform analytics, a creator or a data provider. Each report states the audience it describes, its source, date, collection method and confidence; provider estimates are labeled as estimates. Dossiary does not infer these details from a creator's location, language or content. Missing dimensions stay unknown, and an incomplete country list is never scaled up to 100%. Interests can overlap. The report date is independent of the selected post period. There is no fake-follower percentage or overall credibility score.
Which platforms are covered?
The record contains Instagram, TikTok and Telegram observations. Coverage and available metrics vary by account and source. YouTube is a supported collection lane; linked handles on other platforms do not imply complete analytics. The profile shows the observations actually available.
Can I correct a number about me?
Yes. Use 'Is this you?' on your profile to begin an ownership claim, or write to hello@dossiary.app with the correction and its evidence. Claims require proof of control. Identity merges and regulated category changes are reviewed by a person.