with-action / with-entity — did ‘I saw the agent with the telescope’ name the seeing tool, or describe the agent?

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 with-action → with action (d=1 · visible) with-entity → with entity (d=1 · visible) with-action(X) → with-action() (d=1 · visible) with-entity(E, X) → with-entity(X) (d=3 · visible) with-entity(E, X) → with-entity(X, E) (d=2 · visible)
• slot cross-product min distance within slot 4
• 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

PRIMARY: preregister at least 192 fresh matched vignettes across ordinary observation, logistics, robotics, maintenance, healthcare, security, user interfaces, and data work. Balance worlds where the named thing is an instrument used by the grammatical subject and worlds where it is physically associated with the subject, direct object, or another named participant. Include clauses with two plausible entities, unfamiliar but resolvable identifiers, tempting world-knowledge defaults, and matched reversals. Compare each registered form against its complete careful-English mapping. Add a separate balanced bare-‘with’ ambiguity arm whose identical wording supports each attachment equally; do not pool that under-specified arm into the careful-English non-inferiority scalar. Ask held-out consequence questions without the words ‘action’, ‘entity’, ‘attachment’, ‘instrument’, or the marker names: Who had the telescope? What equipment did the observer use? Which item must be fetched before the task? What can be removed without changing the method? Report both forms separately and by domain and entity position. Prediction: each form is non-inferior to complete careful English within 5 percentage points, improves exact attachment recovery over balanced bare ‘with’ by at least 25 points, and keeps the two critical cross-readings—entity association inferred from `with-action`, and instrument use inferred from `with-entity`—at or below 5%. Hard negatives include an observer carrying but not using binoculars, a target wearing a camera, a robot moving a crate that has a hook attached, a clinician examining a patient who holds a scanner, tools owned by one actor but used by another, inanimate grammatical subjects, passives with omitted agents, and several entities that could satisfy E. Refuted or narrowed if readers attach either marker to the wrong participant or event, infer the forbidden cross-reading above 5%, ignore explicit E, or if either marker trails complete careful English by more than 5 points. Ceiling-bound comparisons are unresolved, not supportive. PREREQUISITE: on the same frozen semantic cells and current cl100k_base, o200k_base, and p50k_base tokenizers, compare each complete marked clause with the shortest adequate careful-English sentence that states the same event, actor, instrument or entity attachment, and resolved identifiers. The least-favourable tokenizer mean may be positive but must be at most +4 tokens. Cost versus bare ‘with’ is diagnostic only because bare wording omits the attachment decision. ROBUSTNESS: test hyphen-to-space conversion, case folding, punctuation and parenthesis loss, deletion or corruption of X, deletion of E, swapped E/X arguments, a reference to a non-participant, and nearby registered forms returned by live preflight. Damage must become invalid, ordinary explicit prose, or trigger clarification; it must not silently reverse which participant had the thing or whether it was used. Adoption is independent evidence: zero non-author use in a current post-ratification window counts against flagship status.

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.