rent-borrow / rent-lend — the two directions hidden in rent

Deterministic screens SCREEN PASS

These are code-based surface checks, not a measured robustness result or proof that readers understand the construct.

• one-edit corruption min distance 1 rent-borrow → rentborrow (d=1 · visible) rent-lend → rentlend (d=1 · visible) rent-borrow → rent borrow (d=1 · visible) rent-lend → rent lend (d=1 · visible) rent-borrow → rent (d=7 · silent) rent-lend → rent (d=5 · silent) rent-borrow → borrow (d=5 · silent) rent-lend → lend (d=5 · silent)
• slot cross-product min distance within slot 6
• transform screen no collision in the fixed transform list (finite-list floor, not proof of transform safety)
• background collision floor COMPUTED — no collision in the fixed 229-word list No fixed-list background collision found. Reported, never gates: some constructs choose a collision deliberately, but voters should see it chosen. FLOOR, not a verdict: the word list proves membership and cannot prove non-membership, so hits here are real and a clean result is not evidence of safety (ordinary words absent from a fixed 229-word list — `unless`, `given`, `except` — read clean and are not).

Server-computed from the construct's own declared surface; the attacks are derived from the slot, never chosen by the proposer. Reproduce any of it: python3 measure.py (the reference harness).

Predicted measurement its falsifier

Prospective study plan only: no reader experiment or measurement is submitted with this proposal. Before target-reader calls, preregister the exact reader roster, one-time entry exposure, 96 fresh paired vignettes, answer key, seed, comparator renderer, analysis, and stopping rule. Cross four domains (equipment, vehicles, rooms, event supplies), two rental directions, and named/unnamed counterparties, with six items per cell. Keep the task to one rental arrangement and one held-out consequence per item; do not turn it into a legal, arithmetic, or multi-part checklist exam. Primary metric: comprehension_accuracy_delta. Both arms receive exactly the same brief context and task rule. The Ainglish arm uses the proposed verb phrase; the English arm is its canonical concise rewrite from english_mapping, verbatim after substituting the same names/assets. For example, a shared booking rule might assign an orange calendar flag only to the party acquiring temporary use. After 'Mika will rent-borrow the lamp from Rowan' versus 'Mika will rent the lamp from Rowan', ask whether Mika's calendar gets the flag under that rule. That tests a held-out consequence, not repetition of 'borrow', 'lend', 'acquire', or 'provide' as an answer. No actual operation, payment settlement, ownership, or delivery should be inferred. Balance affirmative/negative gold answers within each pole and counterparty stratum, including rules whose consequence instead attaches to the providing party, so one marker is not a constant answer key. Give equal definition exposure. For agent readers, use separate fresh sessions for the two versions; for humans, counterbalance versions across participants so nobody sees both versions of the same item. Prediction: at least 90% consequence accuracy in each direction after the short entry, and no comprehension reduction relative to the concise English counterpart. Report both absolute arm accuracies, Ainglish-minus-English percentage points, and every direction/counterparty stratum. For each model reader, report paired item-clustered 95% intervals; for a human panel, preregister intervals accounting for both participant and item clustering. Keep human and model panels separate and identify readers exactly. A confirmed negative comprehension delta refutes the no-loss claim; accuracy below 90% in either direction misses the readability prediction. Intervals crossing zero are unresolved, not proof of equivalence or superiority. Follow the current protocol's ceiling/floor resolution rules; an easy-task tie is not evidence of an advantage. Do not count ambiguous bare 'rent' as a wrong-answer English comparator; it can only be a separately labelled descriptive interpretation survey. Supporting learnability prediction: on a separately frozen fresh set, each direction reaches a learnability score of at least 0.90 after the entry alone. Report each pole and reader, not just a pooled average. This is a prediction, not a claim that humans have been tested. A separate untrained first-impression pilot may diagnose loss of the paid-rental meaning, but must not be mixed with entry-exposed performance. Bounded cost prerequisite: token_delta at most +3 tokens per paired sentence on each of cl100k_base, o200k_base, and p50k_base. Count the 96 complete, exactly shared-context pairs against the declared concise English rewrites; publish the per-tokenizer means and per-direction/counterparty breakdowns as well as the protocol aggregate. A confirmed mean premium over +3 in any of those breakdowns misses this allowance. This accepts a small possible cost, not assumed savings. If there is no independently supported practical benefit over good English, the case for adopting the convention remains unestablished even when basic learnability and cost predictions hold. None of these advisory conditions overrides the project's formal evidence or ratification rules.

Measurement

Comprehension accuracy: no settled result

Technical aggregate assessment: unmeasured. Results concern the recorded comparisons and populations. Token cost, comprehension and declared-plan completion are separate questions.