How the awards are computed

This page is published before the season starts and does not change once ratings open. When winners are announced, this is the rule book they were computed with. If you think a result is wrong, the grievance officer page tells you how to raise it.

1. Who can rate

• Only a phone-verified PUJO360 account can rate, and only after a verified visit to that pandal.
• A verified visit is either a scan of the pandal’s official QR poster, or a confirmed geofence visit: at least 90 seconds within 50 metres of the pandal with location accuracy of 50 metres or better.
• A manual “I’m here” check-in counts as a visit for your own tracker but does not unlock rating.
• One rating per account per pandal. It can be edited for 24 hours after it is first given; after that it is fixed.

2. What a rating is

A rating is one overall score from 1 to 5 stars, plus up to three “what was best here” category picks. The overall stars feed every category the pandal is nominated in; the category picks add a bonus signal to that category only. Text comments are stored but never scored.

3. Trust weighting

Every rating carries a weight between 0 and 1. An account that was created five minutes ago at a pandal gate and has one visit carries less weight than one with many visits across several days. The weight rises with account age, number of distinct pandals visited, number of distinct days active, and a consistent device. Weights are computed by a fixed formula, not by hand.

4. Removing outliers and brigading

• Ratings from accounts that share a device, IP address or network cluster and rate the same pandal within a short window are grouped and their combined weight is capped as if they were one account.
• Bursts of ratings far outside a pandal’s normal pattern (for example, hundreds of 5-star ratings in ten minutes from new accounts) are held back from scoring and shadow-flagged for review.
• Mock-location and failed device-integrity checks invalidate the visit, and therefore the rating.
• Removed ratings are never silently deleted; they are excluded from the score and kept in the audit trail.

5. The score: a Bayesian average

The published score for a pandal in a category is a weighted Bayesian average that pulls small samples toward the global mean:

score = (m × C + Σ weightᵢ × starsᵢ) ÷ (m + Σ weightᵢ), where C is the global mean of all verified ratings across all pandals in that category and m = 20 is the prior weight.

• m = 20 means a pandal needs roughly 20 ratings’ worth of weight before its own average dominates the global mean.
• A pandal appears on a leaderboard only when it has at least 20 verified-visit ratings in that category (counted after trust weighting and outlier removal). Below that, its score is not shown anywhere.
• Ties are broken by the higher number of verified ratings, then by the earlier time the threshold was reached.

6. Refresh and lock

• Leaderboards are recomputed every 15 minutes. The “computed at” time is shown on the page.
• All ratings hard-lock at 23:59 IST on Vijaya Dashami (21 October 2026). Nothing given, edited or removed after that affects the results.
• Final results are published within 48 hours of the lock, after the review in section 7.

7. Jury override — logged and disclosed

A superadmin may override a rank only to correct verified manipulation or a data error (for example, a pandal listed twice). Every override records who did it, when, the previous rank, the new rank and a written reason in the audit log, and is disclosed on the final results page next to the affected entry. Overrides are never used to reward or penalise a pandal on taste.

8. What is not part of the score

• Money. Committees cannot pay for a rank, and timed-entry pass sales do not affect any score.
• App installs, shares, photo counts or crowd levels.
• Ratings from PUJO360 staff or their households, which are excluded by account.