Two signals, two different questions
A ranking should help people find consistently strong tattoo work. To do that, the system needs to distinguish execution from personal taste, and evidence from mere exposure. The product uses two separate judgment formats.
Head-to-head
“Which tattoo is better executed?” Two pieces are compared directly. The result tells us how one piece performed against a specific opponent.
Public approval
“Do you like this tattoo?” One piece is shown. Right swipes and left swipes describe approval among the people who actually judged it.
A portrait and a tiny fine line flower should not be forced into a comparison simply because both are available. The intended matching foundation is style and size, with subject, color, healing appearance and photographic presentation helping refine otherwise similar matches.
AI describes and classifies the images to help organize these comparisons. It does not vote, award artistic points or decide the winners. Skin tone is never a ranking reward or penalty.
What an Elo-style rating actually means
Elo is one way to turn head-to-head results into a relative performance estimate. Its useful idea is expectation: beating an opponent you were unlikely to beat provides different evidence from winning a comparison you were strongly expected to win.
A conventional illustrative model estimates the probability that A beats B as E(A) = 1 / (1 + 10^((R(B) − R(A)) / 400)). After the comparison, it updates A by R′(A) = R(A) + K × (S(A) − E(A)). S is 1 for a win and 0 for a loss; K controls how quickly the rating moves. B receives the opposite change in this simple example.
Try a head-to-head example
Educational demonstration only. These are not the production scoring rules.
Example K = 24 and scale = 400. These constants illustrate conventional Elo behavior, not an approved Tattoo Awards formula.With equal ratings, either piece has a 50% expected chance in this demonstration. A win with K = 24 produces a 12-point increase, with a corresponding 12-point loss for the opponent. An upset produces a larger adjustment. Neither outcome proves permanent superiority from a single vote.
Raw Elo alone does not solve sparse data, uncertain ratings, disconnected styles, unfair matchups or biased judging. Confidence-aware alternatives and calibration are still being evaluated. A starting number must never be presented as an earned official standing. Our public example ratings are labeled Elo for that reason.
Why 100% approval can still be uncertain
Public approval is the proportion of valid positive reactions among valid likes and dislikes: likes / (likes + dislikes). A skipped tattoo is not automatically a dislike. Merely seeing an image is not a vote.
Four likes from four people is a perfect observed percentage, but it tells us much less than a large, independently collected sample. A confidence-aware estimate helps prevent tiny samples from appearing decisively superior.
Compare approval and evidence
This Wilson lower-bound example illustrates uncertainty. It is not the live scoring formula.
Illustration uses a 95% Wilson interval under independent observations. Real audiences may be biased or correlated; the interval does not correct those problems.Audience context matters too. People browsing floral ideas have already selected a topic they may like. A generic discovery feed has a different audience. We retain voting mode and context so popularity within a particular gallery is not silently treated as universal artistic superiority.
From individual tattoos to an artist ranking
Validate the judgments
Keep raw voting events, mode and context. Exclude self-votes, duplicates, invalid activity and synthetic test evidence from real quality calculations.
Evaluate each tattoo within comparable work
Use head-to-head outcomes for relative execution evidence and public responses for approval evidence. Track the uncertainty around each source separately.
Calibrate the sources
The two signals live on different scales. A percentage cannot simply be added to an Elo number. Their calibration and weighting must be validated before combining them.
Assess the portfolio
An artist’s standing should reflect consistent quality across distinct tattoos, not the sum of uploads or a single lucky image. The final aggregation rule and minimum coverage remain undecided.
Publish only eligible standings
Official ranks need enough independent evidence and the required participation eligibility. Keep insufficiently evaluated artists Unranked rather than presenting an initial model estimate as a finished result.
Keep quality and visibility separate. Voting participation can unlock features or help exposure. Paid features can change commercial visibility. Neither should be interpreted as proof that the tattoo was better executed.
Awards recognize a particular category and period. Their verified performance evidence can inform the broader record, but the treatment of an award itself is not yet finalized. See awards and rankings.
How fair judging is protected
One piece, one exposure
The intended no-repeat rule applies to the tattoo, not its image file. Another angle is the same piece. Skipping still consumes an exposure; preloading an offscreen image does not.
No self-voting
An artist must not evaluate their own tattoos. Current authenticated artist judging also excludes work from the judge’s primary city. Coordinated vote trading is not independent evaluation.
A considered pace
There is no blanket per-minute voting cap. Normal 1.5–2 second judgments and occasional faster choices are allowed. Five consecutive sub-second judgments currently trigger a 10-second breather.
Separate uncertainty
Few judges, narrow portfolio coverage and uneven opponent pools can all weaken a claim. More votes from one connected group do not automatically create more independent evidence.
Judge the tattoo, not the photography. Good presentation helps people see the work; expensive lighting does not itself establish artistic quality. The intended matching system uses photographic presentation as a secondary fairness consideration, not a direct quality bonus.
Current authenticated comparisons use the existing classifier’s shared-style, color, complexity and photographic-quality compatibility checks. The newer size-and-style-first pipeline is being evaluated in the Image Processing Test Center. It has not silently replaced the live matcher.
Why local discovery can differ from quality rank
A city leaderboard answers a different question from “Who is nearest to me?” Quality evidence belongs to the artist’s work. Distance and location help decide which eligible artists are relevant to a particular visitor.
The intended discovery layer can take proximity into account without changing the underlying artistic quality estimate. A move between cities should not erase the artist’s history, and a regional search should not masquerade as a global competition.
National and global comparisons need broader coverage and calibration between otherwise disconnected groups. We will not label a small local test pool as a representative world ranking.
Updates, history and current implementation
The target is to record accepted votes immediately and publish ranking snapshots approximately every five minutes when there are changes. That is the intended ranking architecture, not a claim that official rankings are already being computed.
Monthly award results will freeze at the end of the city’s local calendar month and undergo integrity and ownership checks. Annual results will use their published annual rules. New standing changes should not rewrite a legitimate historical award.
| Capability | Status |
|---|---|
| Authenticated artist head-to-head events and exposure records | Implemented |
| Single-image reaction event support | Implemented for the account flow; public preview swipes remain local demonstrations |
| Public quality ranks, final Elo/approval combination | Not published; formula under evaluation |
| Monthly and annual awards | Planned; no confirmed competitions or winners yet |
| Independent award ownership verification | Required before confirmation; review workflow pending |
| National and global awards | Future expansion |
If you see a fictional artist or a Elo label in the public design previews, those samples are there to demonstrate the interface. They are not evidence about a real artist.