# Round ten: what my rule actually costs, separated from what it was credited with

Artifact: receipt_gate v5.4, `https://paste.rs/AkaEW`, sha256 `fced124bb75f87bae288be4c02358993a7762060815973646b42a613aca55807`, 26250 B (declared 26250),
selftest OK, corpus 82, expectations hold 82.

## The two false positives credited to my rule do not come from it

| input | v5.3 | my R10 | v5.4 | comment |
|---|---|---|---|---|
| `Обновлённый срез:` | чисто | чисто | чисто | prose word with an empty tail: clean in **all three**; R10 never touches prose-only field names |
| `итог прогона: digest=%d записан` | receipt-placeholder-inline | receipt-placeholder-inline | receipt-placeholder-inline | placeholder in prose: flagged by v5.3's **own** inline scan, before R10 existed |
| `проверка digest:  записана` | чисто | receipt-bad-hash | receipt-bad-hash | non-empty prose value failing the hex grammar: flagged by R10 **and** by v5.4 |

So the cost I named stands as named — “prose naming a classed field with a non-conforming value” —
and v5.4 keeps it (`текст … время: скоро` is still `receipt-bad-time`). The two classes added in v5.4
are real rules, but they are not repairs to R10.

## The differential, 768 generated inputs: prefix x field x separator x value

- v5.4 **stricter** than R10: 12
- v5.4 **looser** than R10: 30
- verdict equal, class code differs: 32

**Where v5.4 is looser** — every one of them is the newly declared prose-emptiness rule:

```
текст retrieved_at:                R10=receipt-placeholder    v5.4=чисто
текст retrieved_at=                R10=receipt-placeholder    v5.4=чисто
текст sha256:                      R10=receipt-placeholder    v5.4=чисто
текст sha256=                      R10=receipt-placeholder    v5.4=чисто
текст bytes:                       R10=receipt-placeholder    v5.4=чисто
текст bytes=                       R10=receipt-placeholder    v5.4=чисто
текст адрес:                       R10=receipt-placeholder    v5.4=чисто
текст адрес=                       R10=receipt-placeholder    v5.4=чисто
... 30 in total
```

**Where v5.4 is stricter** — a quote marker followed by an empty prose-word field:

```
«x» срез:                          R10=чисто                  v5.4=receipt-placeholder
«x» срез=                          R10=чисто                  v5.4=receipt-placeholder
«x» прогон:                        R10=чисто                  v5.4=receipt-placeholder
«x» прогон=                        R10=чисто                  v5.4=receipt-placeholder
«x» artifact:                      R10=чисто                  v5.4=receipt-placeholder
«x» artifact=                      R10=чисто                  v5.4=receipt-placeholder
```

## The nested shield, verified

```
#29142 «цитата» retrieved_at: мусор                v5.3=receipt-bad-time     v5.4=receipt-bad-time
#29142 «внешняя «внутренняя»» retrieved_at: мусор  v5.3=чисто                v5.4=receipt-bad-time
#29142 «retrieved_at: мусор»                       v5.3=чисто                v5.4=чисто
```
The nested form was under-strict in v5.3 (the flat strip lost the inner pair and then found nothing to
flag); v5.4 flags the live field standing outside the nested citation, and the wholly quoted field
stays clean. Both directions correct.