Fortunes written, not remembered

Nothing here carries a name. No philosopher, no ancient tradition, no “as the saying goes”. A fortune signed with a source is a fake record that outlives the tool that printed it, so this one does not print them. Invented attributions are detected separately from the saying check: named sources, national traditions, dash-signed lines and quoted phrases are each looked for, in every field of the reply rather than only in the fortune line.

The harder promise is that a line handed to you is written rather than remembered. The known-saying check runs two independent matching channels — content-word stems, and verbatim word windows — because the most memorable short phrases in English have no content words at all, and a single channel walks straight past them. Anything either channel finds is shown beside the line that collided, never swallowed.

The originality check is a corpus-membership test, not an originality test. It compares each line against a shipped corpus of — anonymous traditional English proverbs and fixed idioms. When it finds nothing, that means no known match was found. It does not mean the line is original, and this app will never tell you it does.

Here is the published number that matters, measured rather than asserted — the held-out rate, not the flattering corpus-member rate. Against lines the corpus holds, the check scores 100% word-for-word and 92.8% through light rewording. Against a held-out set of 160 traditional sayings the corpus does not hold, it scores 4.4%.

That second figure is the honest one, and the held-out set was deliberately not folded back into the corpus afterwards — doing so would have raised the headline number and destroyed its meaning. Expanding the corpus is the only thing that moves it; better matching does not.

The corpus holds anonymous traditional material only: proverbs and fixed idioms with no identifiable author, common property of English for generations. It deliberately holds no named-author quotations, no commercial fortune-slip text and no scripture. The register check above is therefore not a corpus at all — it holds the grammatical shapes the mass-produced slip is built on, and has never held a specimen.

Four tones — dry, warm, ominous, absurd — each required to break exactly one house rule, which is what makes them measurably different rather than four adjectives. Three modes decide what kind of sentence it is: a standing aphorism, a prophecy, or advice. All twelve combinations are verified reachable.

Behind them sit thirteen turns and eleven angles, both prime, against batch counts of three, five and seven. No count divides either set evenly, so covering the device set in order is never the cheapest path to filling a batch.

Every batch comes back with:

• A craft report: the measured frame of each sentence against the device the model claimed, tone discrimination against every other tone, mode grammar, repeated openings and repeated observations.
• Lucky numbers, with conformance reported rather than silently corrected.
• Partial results kept. A truncated batch renders every fortune that arrived, labelled, instead of being thrown away for being incomplete.
• Export as Markdown, plain text, JSON or CSV. History is saved to your SkillSafe account and searchable by meaning.