JutuGo / Fair play

The arithmetic is public, and so are the guardrails

Every JutuGo tournament is decided by a formula, not a judgement call. This page sets out all three scoring methods, the anti-fraud system behind them, and what we do when someone tries to cheat.

The rule underneath everything

One verified account, one vote per tournament. Every rating and every vote is recorded against your account. You can change your mind once, but you cannot stack. Unverified accounts can browse and click, and their input is recorded, but it never reaches a score.

Everything else on this page exists to protect that one sentence, because a scoring formula is only as honest as the inputs feeding it.

Why the account is the unit. A rule like this only holds if accounts are expensive to create in bulk. Verification at signup, device caps and network caps all exist to make the second account harder than the first, and the tenth much harder than the second.

Star rating

Used for photo, craft and kids tournaments. Viewers rate an entry from one to five, and the highest average wins — but not a plain average, because a plain average is trivially beaten by three friends giving five stars.

Instead we use a weighted average that pulls every entry toward the field mean until it has collected enough ratings to have earned its position.

score = (sum_of_ratings + C × field_average) / (count + C)
where C = 25 and field_average is the mean across all entries
  • Thirty five-star ratings from friends moves you far less than the raw average suggests, because 25 imaginary ratings at the field average are sitting in the calculation with them.
  • Entries below 5 ratings are unranked and cannot take a prize position at all.
  • Ratings are shown to two decimals on the leaderboard alongside the raw average, so you can see exactly how much the weighting moved an entry.
  • Ties break on rating count first, then on earliest submission.

Up and down votes

Used where a straight verdict fits better than a scale. An entry collects upvotes and downvotes, and the score is not the raw difference between them — raw totals just reward whoever was seen by more people.

We use the lower bound of a Wilson confidence interval, which asks how good an entry's approval rate can be shown to be, given how many people actually voted.

score = wilson_lower_bound(upvotes, downvotes) × 100
90 up / 10 down → 82.2
900 up / 600 down → 57.5

An entry with ninety approvals out of a hundred beats one with nine hundred out of fifteen hundred, which is the correct answer and the opposite of what a raw net score produces.

The watch-time blend

Video tournaments combine three signals into one score out of 100. Each factor is scored against the field, then weighted.

40%

Total watch time

Every verified second spent on your video, added up.

35%

Completion rate

How far through the average viewer got before leaving.

25%

Votes

Wilson score on upvotes against downvotes.

watch = 100 × ln(1 + hours) / ln(1 + field_best_hours)
completion = average_seconds_watched / video_length × 100
votes = wilson_lower_bound(up, down) × 100
final = 0.40×watch + 0.35×completion + 0.25×votes

Two deliberate choices sit in there. Watch time is log-damped, so a single video going viral cannot absorb the whole 40 points and flatten everyone beneath it. And completion rate is the factor nobody can buy — you can purchase views, but you cannot purchase people choosing to stay.

Worked example. Entry A gets 8,000 views watched to 83%. Entry B gets 40,000 views watched to 15%, and more votes. Entry B has five times the reach and still loses, because two of the three factors do not care how many people arrived.

How watch time is counted

  • Verified accounts only. Anonymous plays appear in the public view count and contribute nothing to any score.
  • One session per viewer. We take your longest single session on a video, capped at the video's own length. Looping it overnight adds nothing.
  • Real playback only. Scrubbing ahead is not counted, playback must be at normal speed, and the tab must be in focus.

What we detect

Vote manipulation is not usually subtle. It leaves the same fingerprints every time, and most of them are visible in the logs without anyone needing to prove intent.

  • Device and network caps. Two accounts per device, twenty per network per hour. Generous enough for a shared office or a hostel, useless for bulk account farming.
  • Account age weighting. Accounts created after a tournament opened carry reduced weight in that tournament, which removes most of the value of creating accounts to order.
  • Single-target accounts. An account whose entire history is one vote on one entry is a strong signal, especially in clusters.
  • Timing bursts. Hundreds of votes on one entry arriving inside a few minutes from newly created accounts.
  • Self-voting. You cannot rate or vote on your own entry, and linked accounts voting for each other in a ring are detected retrospectively.
  • Playback anomalies. Impossible session patterns, headless clients and background tabs reporting watch time.
Shared connections are not punished. IP is treated as a signal to investigate, never as an automatic block. A college network producing forty honest voters is normal and gets treated as normal.

What happens to flagged activity

  1. Flagged, not blocked. Suspect votes stay in the log and continue to display during the round, so nobody can probe the system by watching their score move.
  2. Stripped at lock. During the 48-hour window after judging ends, flagged votes are removed from every affected entry and scores are recalculated.
  3. Published. The results page shows the number of votes removed, tournament-wide. Removal is never silent.
  4. Escalated where it was organised. An entrant who bought or farmed votes is disqualified and forfeits any prize position. The prize passes to the next qualifying entry.
  5. Account action. Coordinated manipulation costs the account, and the accounts created to support it.

Entrants are not penalised for votes cast on their behalf without their involvement. The votes come out; the entry stays in.

What hosts cannot do

Hosts fund the prize and set the rules before a tournament opens. After that their power over the result is close to zero, by design.

A host can

Write the rules, set the prize and split, choose the judging method before opening, and reject entries at review for breaking a stated rule — with a written reason.

A host cannot

Adjust a score, add or remove votes, change the judging method or weights once entries open, see standings before the public does, or reject an entry for a reason not stated in the rules.

The scoring runs on our infrastructure, the weights are published before the round opens, and the inputs behind every final score are visible on the results page. A host who wants a particular person to win has to convince the same voters as everyone else.

Disputes and appeals

The 48 hours between judging closing and standings locking exist for this. If you believe an entry was manipulated, or that your own entry was wrongly rejected or wrongly flagged, raise it in that window and it is reviewed before any money moves.

  • Rejected at review? The reason is written on your entry. Reply to it and a second reviewer who did not make the original decision looks at it.
  • Suspect another entry? Use the report link on that entry. Reports are confidential and the reporter is never revealed to the host or the entrant.
  • Disagree with the final standings? Every score's inputs are on the results page. If the arithmetic is wrong we will correct it and say so publicly.

Once standings lock and prizes transfer, results are final except where fraud is proven afterwards, in which case account action still follows even though the money has moved.