D-0010 · When a run of words counts as found
How strong must a run of the quoted words be before it counts as found?
Rare enough, not long enough. A run counts as found when its exact words appear in only a handful of places in the whole library, and every match shows that count. How small "a handful" is will be set by measuring how often runs match by chance — not yet measured.
| Status | Ruled by the owner |
| Set | 20 Tishrei 5787 · 2026-10-01 13:27 |
| Recorded in | held 81eb092a |
| Rests on | D-0011 |
| Affects | whether a quote is found, partly found or not found, D-0012 |
| Would change if | The chance-match measurement, once it runs, sets the cut-off; a Daas Torah ruling otherwise. |
The options on the table
- Rare enough, not long enough — the count is shown; the cut-off is measured (chosen)
- Keep 5 words (the machine's earlier guess)
- Raise it to 8 words
- No cut-off — show the run's length and its count, and let the layers decide
Why
Some phrases are everywhere: "אפילו באלף לא בטל" turns up across the halachic literature, so finding it proves little about any one source. A rarer run of the same length proves a great deal. Length is not evidence; rarity is. The July 2026 checker specified exactly this measurement — how many places hold the words, and how far the best place is ahead of the next — and never computed it.
Still to measure
The cut-off: run real phrases against the whole library and against shuffled controls, and set "a handful" where chance matches become rare. Until that runs, every match shows its count and no cut-off is applied.
History
- 2026-10-01 04:45 — the machine's guess: 5 words in a row (experiments/2026-10-01-head-to-head).
- 2026-10-01 13:27 — ruled: rarity, not length; the count shown; the cut-off to be measured.
What rests on this
Generated from the register (docs/decisions/0010-min-words.md). A changed value is recorded as a new entry; the old one is never overwritten.