set-to / adjust-by — is the number the new value, or the size of the change?

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 set-to → set to (d=1 · visible) adjust-by → adjust by (d=1 · visible)
• slot cross-product min distance within slot 7
• 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 CLAIM: explicitly attaching destination/change labels improves correct recovery of numeric consequences on realistic update messages. Use comprehension_accuracy_delta against the complete, concise careful-English mapping above. Before inference, freeze at least 192 fresh scored cases, balanced between set-to and adjust-by and across known-start, unknown-start, and ordered-mixed-update cases. These six form-by-case strata remain load-bearing with fixed equal weights. Balance counts, durations, storage quantities, and credit allowances; include positive, negative, and zero deltas and targets both above and below the prior value. Negative values may only occur in domains where they are meaningful. Ask held-out consequence questions such as whether a later request fits within the revised allowance, whether two updates end at the same value, whether a ceiling would be crossed, and whether the final value is determined at all. Do not ask readers to repeat `target`, `delta`, `set`, or `adjust`, and do not put the answer verbatim in either arm. Use opaque balanced answer choices. Paired arms carry the same initial facts, numbers, units, sequence order, and requested or reported speech act. Use the shortest faithful canonical English template for each case, without artificial padding or omission. A separate balanced ambiguous-message diagnostic may measure ambiguity removal, but cannot substitute for the careful-English claim carrier. Use at least two qualified reader lineages, separately frozen target-independent controls, an immutable manifest, and a minted attempt before reader spend. File every outcome. Report each arm's absolute accuracy, every form-by-case stratum, reader results, item-bootstrap intervals, cell yield, and ceiling/floor resolution. Prediction: a positive pooled careful-English delta with a resolvable interval excluding zero, without confirmed harm in either form. Independent replication uses wholly fresh inputs and preserves the comparator, strata, and estimand. If either form is harmful, a favourable partner must not hide it. A ceiling-bound tie is unresolved evidence of advantage. LEARNABILITY: on a separate held-out population, compare cold reading with reading after one exact entry exposure. This is a separate declared instrument for learning from a definition, not a retrospective repair of the primary result or a simulation of future training. COST AND ROBUSTNESS: report current token cost descriptively against the concise complete English controls under the declared tokenizer roster; no immediate saving is assumed. Test hyphen and parenthesis loss, operator omission, sign loss/change, unit loss, paraphrase, and multi-update summarisation. Distinguish corruption of the operator from corruption of numeric data: the markers are not an error-correcting code for digits or signs. Missing operators must not acquire a guessed default, and unknown earlier values must not become zero. REFUTED OR REQUIRES REPAIR if independent evidence confirms worse consequence recovery than careful English; readers routinely treat the destination as an increment or the increment as a destination; zero adjustments reset values; an unknown starting value is invented; order or unit boundaries are silently changed; or a marker corruption silently swaps the update operation. If careful English matches the marker's accuracy and robustness at lower cost, the extra construct lacks a demonstrated reason for adoption. Future training benefits remain unmeasured until separately tested.

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.