Buy-Again Score
Would they buy it again?
That one question decides everything. The Buy-Again Score reads what owners already wrote about a product and deduces whether they would buy it again, stay again, or play again. No survey, no star-rating average, no marketing copy: just the language people used when nobody was asking them to rate anything.
The two ledgers
Two numbers, never blended silently.
Explicit
Owners who say it outright: “I’d buy this again,” “never again,” “would recommend.” This ledger is precise but small and self-selecting, the people who bother to state intent are not a random sample of everyone who bought the thing.
Inferred
Owners whose language implies it without saying so: complaints about a broken promise, delight described in passing, the small tells that separate a keeper from a regret. This ledger covers the whole corpus, not just the vocal minority.
Both numbers are reported side by side, every time. The explicit rate is never presented as if it were the population rate, and the inferred rate is never presented as if every owner had spoken up. Counting them separately is the whole point.
What makes it trustworthy
The epistemics, published in the open.
The exact scoring mechanics stay internal, the way any calibrated instrument keeps its working parts out of reach of gaming. What is public is how the score earns trust.
Minimum-evidence gates
A product with too little review evidence does not get a score. No evidence, no number, rather than a confident-looking guess.
Thin evidence, thin scores
A handful of reviews cannot swing a score to the extreme ends of the scale. Confidence is earned with volume, the same way it should be.
Two ledgers only, for now
The instrument is built to separate a broken promise (the product failing to deliver what it claimed) from a simple mismatch (too hard for a casual player, too grindy for someone after a quick session), so only the first would lower a score. For this games register, that broken-promise reading has not yet cleared the same evidence bar as the rest of the site: it is tracked but withheld, so the published Buy-Again Score is built only from the two buy-again ledgers, explicit and inferred.
Dated, frozen editions
Every published score belongs to a dated edition. Editions are never quietly edited after the fact; a change in the evidence produces a new, separately dated edition.
Not for sale
A brand cannot pay to raise its score, and cannot pay to have a low score hidden. The register is worth reading only because that line has never moved.
First register, live
62 games, ranked by Buy-Again Score.
Games are the first Buy-Again Score register because reviewers on game storefronts already write in exactly the voice the instrument reads: would I play this again, was it worth the money, did it deliver on the pitch. Before publishing a single score, the language-reading step was checked against 1,473 real recommendation labels already written by owners, so the deduction is verified against ground truth before it goes live, not trusted on faith.
2026-07-20 Edition, frozen 20 Jul 2026. Mattresses are next, followed by more product categories as each corpus clears the same evidence threshold.
The family
One instrument, several registers.
Buy-Again Score runs on the Feel Layer, the same instrument that deduces satisfaction from reviews without a survey. It already powers two live hotel registers.
Questions, or want to talk before launch?
Reach the person building this directly.