Repository path: workshop/regimes/R28-footing-selective.md · rendered 2026-09-09

R28 — footing-selective

v1.0 (frozen 2026-08-14, S187.) The third point on the line R06 and R27 define. R06 is a translator not asked to carry the source's grammatical footing. R27 is a translator asked to carry it at every site, with fluency subordinate. R28 is a translator asked to carry it where it can be carried without asserting something the source does not assert, inside a declared length budget — which is what a practitioner actually faces once R27 has produced the price list.

R28 is not a recommendation either. It is the middle arm of a three-point ladder whose whole purpose is to be priced by somebody other than the translator.

Why it exists

R27 established that at 52 footing sites in one passage, 51 carriages were available and 36 of them cost something the translator wrote down (T-genji-yomogiu-R27-v1). That answers is anything reachable and leaves the practitioner's question untouched: reachable at every site is not the same as worth taking at every site, and a ladder with only its two ends on it cannot show whether there is anything in between. framework/v0.2 §10.6 names the blind pricing of the R06/R27 pair as the design the section needs; a two-arm pricing can only report how much the maximum costs, never what a smaller purchase buys.

Procedure

1. The baseline and the maximum are both frozen first. R28 is written only where an R06 rendering and an R27 rendering of the identical span by the same hand already exist, frozen and committed. The three are one object.
2. The prose skeleton is the baseline's, and this is deliberate. Outside the sites it carries, R28's text is identical to its R06 baseline, modulo line wrapping. The regime is a manipulation of one variable and nothing else: a freely rewritten third rendering would confound footing treatment with every other sentence the hand would have changed on a second look, and no comparison of the three could then attribute anything. A verifier checks the identity mechanically and the regime fails if it does not hold.
3. Read the frozen site list. Decide at every site, and record the decision at every site — carried with the English named, or declined with the reason named. A site passed over silently is a regime breach.
4. The selection rule, fixed before the sites are read. Carry where the English device does not assert something the source does not assert; decline where it does. In practice this means: an elevating transitive verb (LEX) wherever English has one; a deferential auxiliary (AUX) only where the source's predicate is something its subject did on purpose; and no device that requires a clause to be added.
5. The budget, declared as a number before translating. The rendering may add at most 5% to the baseline's word count and may carry at most 20 marked deference tokens in total. The budget exists because R27 §"the one thing the per-site coding cannot see" established that the price of footing carriage is cumulative and the codes are local: a site-by-site rule with no ceiling is R27 again.
6. No apparatus, as R27 §6: no footnote, no gloss, no bracket, no translator's note.
7. Freeze. Commit the rendering and its log before any evaluation of it is designed.
8. File under workshop/translations/<work>/R28-v<N>/.

What it does not license

An R28 output is one hypothesis-aware hand and is never evidence about how English works — R27 §"What it does not license" applies here unchanged and harder, because this regime's selection rule is the translator's own theory of which sites are safe. The counts are self-report and are labelled internal-judgment-only wherever they appear, and note (bio) governs: no primary of any design may be a count over these sites judged by the translator.

The specification names a length budget and it does not name a goodness sense. This is deliberate and it is the one thing a critic should check hardest: a regime specified as "carry it where the English still reads well" would restate naturalness, and any run comparing it against R06 on naturalness would then be measuring its own instruction. The rule above is stated in terms of assertion (does the device say something the source does not) and length, neither of which is a scored sense on wiki/goodness-senses.md.

Roles

• Translator — the lead agent, as a labeled subject (charter §3, A4). No panel call, no API cost.
• Judge — never the lead (charter §5).

Human involvement

None.

Known limitations (by design)

• Everything R06 lists, and everything R27 lists.
• The skeleton constraint cuts both ways. Holding the non-site prose identical is what makes the three-arm comparison attributable, and it also means R28 is not what a translator working freely under this instruction would have written: a real hand would have rebalanced sentences around the marks it kept. R28 therefore prices insertion, not composition.
• The selection rule is a hypothesis. "Do not assert what the source does not assert" is the volitionality account of framework/v0.2 §10.5.3, which was noticed post hoc while coding R27. Executing a regime built on it does not test it.
• One hand, and the same hand, three times. Contamination between the three renderings is total by construction and is the design, not a defect.

Changelog

• 2026-08-14 v1.0 (frozen, S187) — created for ARM-footing-price step 1, to put a third point between R06 and R27 before either is priced. Frozen on creation, for the reason R27 v1.0 and R06 v1.0 were: the ladder it belongs to is written in one session.