GOOD MEASURE LABS · OCTOBER 9, 2026
A puzzle, and a question.
Will people enjoy coming back to a short lost-and-found relationship puzzle? We do not know yet. RecallMeter tests that one question with a small, original experience.
What the puzzle measures
Each UTC date deterministically arranges six original illustrated objects and six text tags in a two-by-three board. After inspection, eight choices ask for the objects’ six locations, one immediate neighbor and one tag association. Every correct choice counts once; there is no speed bonus or penalty. The first daily result successfully saved by cooperating tabs is retained using browser-local locking when available. Each save re-reads current history rather than overwriting from a stale tab. Without locking, sequential saves still reconcile, but simultaneous saves are best effort; localStorage itself is not transactional. Practice is untimed and separate. A suggested two-minute visit is not a time limit.
The result describes this puzzle only. It is not a standardized assessment, IQ estimate, brain age, diagnosis, health advice or evidence of cognitive improvement. No training effect is claimed.
Why try this format?
LinkedIn’s June 8, 2026 Wend announcement reports 86% of its game players returning the next day and 82% after seven days. These are incumbent-reported figures for an established portfolio and network, not expected RecallMeter outcomes. The announcement does not establish comparable cohort definitions or independent demand for this format.
Puzzmo’s Memoku page combines matching with Sudoku constraints and offers archive access through Puzzmo Plus. It demonstrates an existing paid archive model, not willingness to pay here. RecallMeter has no paid features.
Pixel Recall asks people to memorize pixel art and redraw it. Puzzle Theory’s Recall asks people to repeat a growing tile sequence and offers daily and practice modes. Daily memory puzzles already have competition. Our six-object location, neighbor and tag format is a design distinction whose appeal remains unproven. There is no empty-market or growth claim.
Sources checked October 9, 2026. All illustrations and puzzle code are original; no competitor assets or puzzles are reused.
Measurement, with consent
The allowlisted events are daily_start, daily_complete, daily_return and result_share. The tag is off until a dedicated reviewed production stream is configured and a visitor opts in for the current session. Practice never emits these events. Completion and return events require a successful coordinated first save; stale duplicate tabs, unsupported locking and storage failures do not emit them. This introduces browser/storage selection bias in addition to consent bias. The return event means a daily completion in a browser with an earlier saved daily completion; it does not prove a unique person’s retention. Share means the native share operation resolved or copy succeeded, not that another person received or opened it.
No answers, scores, history, custom identities, arbitrary event parameters, query strings or referrers are transmitted. Google Analytics uses standard browser identifiers; consent, storage deletion and multiple devices limit inference. Results and up to 30 daily dates and scores stay in this browser. The game has no accounts, backend, payments, public leaderboard, uploads or model API calls.
When results would be worth interpreting
First verify production consent behavior, delivery to the intended stream and actual receipt. Label every synthetic test separately, leave synthetic launch data out of evaluation, and begin a clean real-visitor window after testing ends. Do not call a successful build, accepted HTTP request or empty report evidence of engagement.
Wait for at least 200 consenting real first-completers in a defined cohort and seven complete follow-up days before an exploratory repeat-completion estimate. This is an initial reporting floor, not a power calculation or proof of product success. Use matched native analytics cohorts for next-day and seven-day completion, state denominators and uncertainty intervals, and separate those from local-history return-event counts. If cohorts cannot be matched with approved aggregate reporting, report that limitation rather than a retention rate. Count daily completions, not just visits; disclose consent and storage selection bias. Smaller samples remain inconclusive. No user-level exports or new identities are required.
Primary question: repeat daily completion. Guardrails: accessible completion, minimal page weight, no unwanted tracking, and no health claims. Any future iteration should have a dated hypothesis, source commit, real cohort definition and comparable window. Organic traction and monetization are currently unknown.