How we verify Perceptle answers

perceptle.com does not publish a written answer key. A number on this site is either read off the game after a replay, or reconstructed from the public client and marked that way. I would rather leave a day blank than fill it. The archive can list a puzzle number without a solution. That is a choice, not a missing file.

Replay

On 22 October 2026 I played puzzle #97 on perceptle.com and used the result line as the source. Entering 1 on Count produced “off by 48 dots,” which fixes the total at 49. Clicking Track at x = 491.81 produced “0 px from the landing spot.” Those two are marked replay. Angle, Tempo, and Split on that page did not get the same HUD check, so they stay marked reconstructed even though they sit inside the client’s generator ranges.

Reconstruction

The daily targets are produced by the site’s own script from the date. A reconstruction means we followed that public logic, or a page that documents the same seed, and then checked the result against the ranges: dot counts, landing bands, angle spans, BPM, and fractions. It does not mean the game printed the number. Friday #98 is entirely in this bucket until someone replays it and reads the HUD.

Illustration of many small dots used in the Count test — Count groups dots that flash for about two seconds. Positions here are illustrative, not a copied puzzle.

Illustration of a ball arc and a landing mark — Track asks where a hidden ball will land. The arc is a schematic, not a recorded flight.

What else is checked

Puzzle numbers: #1 is 18 July 2026, then one per day. Confirmed against the in-game archive.
Friday flag: Friday and puzzle number at least 14. #7 is the counterexample.
Scoring curve and σ values: read from the public script, not from a guide’s paraphrase alone.
Averages: perceptle.com/api/stats, with the snapshot time written next to the table.
Other domains: opened in a browser on 22 October 2026 and described from that visit.

The daily script only refreshes the stats snapshot. It does not promote a reconstruction into a replay, and it does not create an answer page for a day with no targets. A correction email to [email protected] that includes the date, the test, and what the HUD said will be rechecked before the page changes. The editorial policy is the process around that email.

Why not publish all 97 solutions?

We do not have a replay or a reviewed reconstruction for each day. Publishing a pattern that “looks right” would be an invented answer. The calendar is complete. The solutions are not.

What does reconstructed mean?

The number came from the public puzzle logic, not from the game’s result line on that day. Treat it as a strong lead, and replay the archive inside the game if you need certainty.