JUDGING PROTOCOL // FULL DISCLOSURE
Judged by AI. Audited by a human.
Every PeakShot entry is scored by Claude, Anthropic’s AI model, against the photographic brief of the zone it was shot in. The rubric is public, the settings are deterministic, every entrant gets feedback, and no podium is announced until a human has checked it. Here is exactly how it works.
Rubric criteria
5
Weighted, published
Total score
100
Per entry, per zone
Feedback
2–3 sentences
For every entrant
Podiums audited
All
By a human, pre-announcement
THE PIPELINE
From summit to score
What happens to your photograph between tapping submit on the fell and seeing your name on a leaderboard.
Shoot and sync
Your best frame syncs from the hill with its capture data and a weather stamp, so judging sees the conditions you actually shot in — a moody, clag-bound frame is not marked against someone else’s bluebird day.
Scored by Claude
Claude (the AI model built by Anthropic) reads your photograph against the zone’s stored photographic brief — what the view is and what makes it work — and scores it on the public rubric below.
Feedback within minutes
You get a provisional score and two to three sentences of constructive feedback shortly after syncing. Every entrant, every entry — something no human-judged competition can offer at any scale.
Final judging at close
When the competition closes, every entry is re-scored in a single batch on the season’s fixed rubric version, so the final standings are judged under identical conditions.
Human podium audit
Before any winner is announced, a human — the founder — audits every podium: the photos, the scores and the evidence behind them. No result is published on AI say-so alone.
SCORED OUT OF 100
The rubric, in full
This is the exact weighting Claude scores against. It is the same for every zone, every fell and every member — and when it changes, the version number changes with it.
Composition
30Framing, balance, foreground interest and how you used the view the zone points you at.
Light & atmosphere
25Quality of light, weather and mood — what the mountain gave you, and what you did with it.
Technical quality
20Focus, exposure and sharpness, judged fairly for a phone-first, in-app camera.
Fit to the zone brief
15How well the frame answers the zone’s photographic brief. The brief describes the view — never the location.
Distinctiveness
10Whether your frame stands apart from every other entry shot from the same ten-metre square.
A note on the brief-fit criterion: every Photo Zone carries a written brief from route scouting — the view it frames and what makes it work photographically. Your entry is judged against that brief. The briefs, like the zones themselves, stay secret until you are standing in one.
SAME JUDGE, EVERY TIME
How we keep the judging consistent
A competition is only fair if entry number one and entry number nine hundred meet the same judge. Three measures make sure they do.
Fixed rubric version
The judging prompt and rubric are versioned. A season is judged start to finish on one version — the rules cannot quietly change under your feet mid-competition.
Deterministic settings
Judging runs at deterministic model settings, so the same photograph against the same brief produces the same score, not a roll of the dice.
Calibration against drift
A fixed calibration set of reference entries is periodically re-scored. If the scores move, we know the judge has drifted before it touches a leaderboard — and we investigate.
INTEGRITY SCREEN
The judge is also a gatekeeper
Alongside scoring, the AI screens every entry for things that do not belong in the competition — and flags them for a person to decide.
What gets flagged
- Suspected screen photos — a photograph of a screen or a print rather than the view in front of you.
- AI-generated imagery — frames that appear synthesised rather than captured through the in-app camera.
- Off-brief entries — submissions that do not plausibly answer the zone’s brief at all.
What happens next
A flag is not a verdict. Flagged entries go to a human review queue — the AI never disqualifies anyone on its own. And the screen is one layer of a wider fair-play system: capture-time hashing, GPS plausibility checks and in-app-camera-only entry mean a suspect photo usually contradicts its own evidence trail before a reviewer ever opens it.
Finally, every podium is audited by the founder before winners are announced. If something got through, it stops there.
COMMUNITY VOTE
People’s Choice — the award the AI doesn’t touch
The AI picks the podium. The community picks its own winner, and the two awards sit side by side.
Blind pairwise voting
You are shown two anonymised photographs from the same competition and asked one question: which is the stronger frame? No names, no scores, no follower counts — just the pictures. Each choice feeds an Elo-style rating, and the highest-rated entry at season close takes People’s Choice.
Built to stay honest
- Entries stay anonymised until the season closes.
- You are never shown your own entries to vote on.
- A daily voting quota and velocity limits blunt any attempt to brigade the result.
- Public galleries use EXIF-stripped renditions — no location data ever leaves the originals.
HONEST LIMITS
A judgement, not a measurement
We will not pretend an AI score is an objective truth about your photograph. It isn’t — and no judge’s score ever has been.
What we claim — and what we don’t
AI judging is a judgement, not a measurement. Photography is not a lab result, and a score of 82 is an informed opinion rendered consistently — not a fact about your frame. What we can promise is that the opinion is the same for everyone: the same rubric, the same settings, the same brief, published in full on this page. That is precisely why the rubric is public and why every podium is audited by a human before it stands.
Disagree with a score? Appeal it.
If you believe the judge has misread your photograph, you can request a re-judge from your entry page. Your entry is scored again against the same rubric version, and appeals are visible to the human audit. We would rather re-run a judgement than have you wonder whether it was fair.
HIKE · REACH · SHOOT · JUDGE