english_mapping |
− Use `while-overlap(E; C)` only when E resolves to an event or state with a time interval and C is asserted, requested, or instructed to hold during a nonempty part of that interval. It marks temporal overlap. It does not by itself say that C spans all of E, starts or ends with E, causes E, is caused by E, or contrasts with E. If full containment, exact concurrency, ordering, or causation matters, state it separately. Use `while-contrast(A; B)` when A and B are both issued as claims and the speaker directs the reader to compare them as different, opposed, or unexpectedly coexisting considerations. It corresponds to contrastive English `whereas` or `although`, not to a claim that A and B overlap in time. It does not say which clause is preferred, more important, causal, exceptional, or normatively controlling unless the surrounding sentence says so. The two forms may both be true in a world, but they make different discourse claims: temporal co-occurrence never entails contrast, and contrast never entails temporal overlap. References and clause boundaries must be recoverable; otherwise ask rather than guessing. Bare `while` remains legal when the temporal-versus-contrast distinction cannot affect an inference or action.
+ Use `while-overlap(E; C)` only when E resolves to an event or state with a time interval and C is asserted, requested, or instructed to hold during a nonempty part of that interval. It marks existential temporal overlap. It does not say that C spans all of E, starts or ends with E, causes E, is caused by E, or contrasts with E. Because partial overlap is sufficient, `while-overlap` MUST NOT scope a prohibition, safety invariant, or other obligation whose satisfaction requires coverage over all of E. Use `while-throughout(E; S)` when E resolves to an interval and the truth of positive state-clause S is required at every relevant instant from E's declared start through its declared end. Express a prohibition as the positive permitted state that must persist, for example `while-throughout(migration-7; service-not-restarted)` only when `service-not-restarted` is a resolved state predicate; do not treat a momentary non-event as proof of whole-interval compliance. `while-throughout` does not require S to begin or end with E, and it does not assert cause, contrast, or what holds outside E. Use `while-contrast(A; B)` when A and B are both issued as claims and the speaker directs the reader to compare them as different, opposed, or unexpectedly coexisting considerations. It corresponds to contrastive English `whereas` or a non-dominance reading of `although`, not to a claim that A and B overlap in time. It does not say which clause is preferred, more important, causal, exceptional, conceded, or normatively controlling unless the surrounding sentence says so. The forms can describe facts in the same world but make different relation claims: some overlap does not entail throughout coverage; throughout coverage entails nonempty overlap only for a nonempty E; contrast entails neither temporal relation. References, polarity, interval boundaries, and clause boundaries must be recoverable; otherwise ask rather than guessing. Bare `while` remains legal when these distinctions cannot affect an inference or action.
|
rationale |
− English `while` has two common readings that license different inferences. In `rotate the key while the light is green`, it locates an action in an interval. In `the local model is private while the hosted model is faster`, it means roughly `whereas` and says nothing about simultaneity. Humans usually use context to repair the ambiguity, but short agent handoffs, policy clauses, summaries, and translated instructions often omit enough context for the wrong reading to survive.
The pair is teachable in one question: **overlapping in time, or being set in contrast?** `while-overlap` requires an explicit interval-bearing reference and withholds contrast, causation, and full-duration claims. `while-contrast` takes two bounded clauses, asserts both, and withholds timing. This prevents a comparison from being scheduled as concurrency and prevents a timing instruction from being treated as a rhetorical concession.
The distinction is intentionally narrower than general temporal logic or discourse annotation. It does not replace `before`, `after`, `during`, causal markers, preference order, or exception syntax. It only types the two incompatible jobs performed by bare `while`. The filing-time register audit searches every proposal stage for the exact forms and for combinations of `while`, `whereas`, temporal overlap, simultaneity, concession, and contrast. No registered proposal currently owns this distinction.
+ English `while` hides both a discourse fork and a temporal quantifier. `Check the log while the upload runs` often asks for some overlap. `Do not restart while the migration runs` normally imposes a whole-interval invariant. `The local model is private while the hosted model is faster` means roughly `whereas` and says nothing about simultaneity. Treating all three as generic overlap makes a prohibition trivially satisfiable after one compliant moment; treating contrast as timing schedules an unintended concurrency constraint.
The three-way test is memorable: **sometime during, the whole time, or set in contrast?** `while-overlap` requires an interval-bearing reference and asserts nonempty temporal intersection. `while-throughout` requires a positive state predicate at every relevant instant of the named interval; the positive-state discipline prevents `not X happened at one sampled moment` from masquerading as continuous compliance. `while-contrast` asserts two bounded clauses and their comparison while withholding timing. All three withhold causation.
This is intentionally narrower than general temporal logic or discourse annotation. It does not replace `before`, `after`, causal markers, preference order, exceptions, concessions with a dominant clause, or richer interval algebra. It types the three operationally incompatible jobs that bare `while` performs in short handoffs. The all-stage register audit searches the exact forms and combinations of `while`, `whereas`, temporal overlap, simultaneity, throughout, whole-interval invariants, concession, and contrast. No registered proposal currently owns this three-way distinction; `in-parallel / in-sequence` controls ordering between action lists, and `time-total / longest-stretch` compares accumulated duration with a longest continuous run, neither types the scope of `while`.
|
predicted_measurement |
− PRIMARY CLAIM CARRIER: preregister at least 144 fresh consequence scenarios, balanced 72 temporal and 72 contrastive, across operations, monitoring, contracts, scientific summaries, scheduling, safety instructions, product comparisons, and ordinary coordination. Each world fixes the relevant event interval, both clause truth values, and whether timing or comparison is the load-bearing relation. Randomize readers across three arms: the registered form, deliberately ambiguous bare `while`, and complete careful English using `during a nonempty part of` or `whereas` with the same facts. Ask held-out questions that do not repeat marker words: whether a scheduler must overlap actions, whether either clause can occur at a different time, whether both clauses are asserted, and whether one clause is merely a time anchor.
The declared `comprehension_accuracy_delta` is registered form minus the balanced bare-`while` arm, not registered form minus careful English. Prediction: at least +25 percentage points overall, at least +20 points in each relation stratum, and at least 90% absolute exact relation-plus-entailment accuracy for each marker. Complete careful English is a ceiling and information-equivalence control: report it separately, and flag a deficit greater than 5 points as a usability warning rather than relabelling it as success on the bare-English claim. Report every form × domain × question-type cell. REFUTED if either marker fails 85% absolute accuracy, improves by less than 10 points over bare `while`, induces temporal-overlap answers on more than 10% of contrast cases, induces contrast answers on more than 10% of temporal cases, or routinely imports causation, full-duration coverage, preference, or exception semantics. A ceiling-bound or chance-bound arm is unresolved, not a win.
PREREQUISITE: on a separate frozen set of at least 48 complete semantic pairs, measure `token_delta` for complete marked sentences against their complete careful-English mappings under current cl100k_base, o200k_base, and p50k_base. The least-favourable tokenizer mean may be positive but must be at most +4 tokens. Cost against bare `while` is diagnostic only because bare `while` omits the load-bearing distinction.
ROBUSTNESS: test hyphen-to-space, case folding, dropped suffixes, swapped clause order, missing or non-interval event references, negated clauses, nested reported speech, both relations holding in the same world, and speech-to-text loss. Hyphen loss may fall back to direction-preserving ordinary wording; dropping `overlap` or `contrast` must reopen ambiguity rather than silently selecting a reading. Verify gold answers against frozen interval and clause records, not annotator intuition. Adoption remains separate evidence: zero non-author use in a current post-ratification scan counts against flagship status.
+ PRIMARY CLAIM CARRIER: preregister at least 180 fresh consequence scenarios, balanced 60 nonempty-overlap, 60 whole-interval, and 60 contrastive, across operations, monitoring, contracts, scientific summaries, scheduling, safety instructions, product comparisons, and ordinary coordination. Before any reader call, every item must carry machine fields `while_kind: overlap|throughout|contrast`, `interval_ref`, `coverage_demand: some|all|none`, `polarity`, and frozen start/end facts. Include positive actions, persistent states, prohibitions rewritten as positive invariants, partial-overlap counterexamples, intervals with gaps, empty or unresolved intervals, and worlds where more than one relation happens to be true but only one is asserted. Randomize readers across three arms: the registered form, deliberately ambiguous bare `while`, and complete careful English using `during a nonempty part of`, `throughout the entire interval`, or `whereas`, with the same facts. Ask held-out questions that do not repeat marker words: whether one compliant instant suffices, whether a scheduler must overlap actions, whether a state may fail midway, whether either contrastive clause can occur at another time, whether both clauses are asserted, and whether one clause is merely a time anchor.
The declared `comprehension_accuracy_delta` is registered form minus the balanced bare-`while` arm, not registered form minus careful English. Prediction: at least +25 percentage points overall, at least +20 points in each of the three relation strata, and at least 90% absolute exact relation-plus-entailment accuracy for every marker. Complete careful English is a ceiling and information-equivalence control: report it separately, and flag a deficit greater than 5 points as a usability warning rather than relabelling it as success on the bare-English claim. Report every form × domain × coverage-demand × question-type cell. REFUTED if any marker fails 85% absolute accuracy, improves by less than 10 points over bare `while`, accepts partial overlap for more than 5% of `throughout` obligations, imports whole-interval coverage into more than 10% of `overlap` cases, induces timing answers on more than 10% of contrast cases, induces contrast answers on more than 10% of temporal cases, or routinely imports causation, preference, exception, or concessive dominance. A ceiling-bound, floor-bound, or chance-bound arm is unresolved, not a win.
PREREQUISITE: on a separate frozen set of at least 60 complete semantic pairs, balanced twenty per marker, measure `token_delta` for complete marked sentences against their complete careful-English mappings under current cl100k_base, o200k_base, and p50k_base. Report all three marker strata; the least-favourable tokenizer mean over the equally weighted strata may be positive but must be at most +4 tokens. Cost against bare `while` is diagnostic only because bare `while` omits the load-bearing relation and coverage distinctions.
ROBUSTNESS: test hyphen-to-space, case folding, dropped suffixes, confusion between `overlap` and `throughout`, swapped clause order, missing or non-interval event references, unresolved boundaries, negation versus positive invariant spelling, nested reported speech, multiple relations holding in the same world, and speech-to-text loss. Hyphen loss may fall back to direction-preserving ordinary wording; dropping or changing the relation suffix must reopen ambiguity or visibly change meaning, never silently preserve the original claim. Verify gold answers against frozen interval traces and clause records, not annotator intuition. Re-run qualification and the frozen study for each declared reader version; a result for one model roster is not durable evidence for a replacement roster. Adoption remains separate evidence: zero non-author use in a current post-ratification scan counts against flagship status.
|
corruption_neighbors |
− [{"from":"while-overlap","to":"while overlap","yields":"hyphen loss leaves direction-preserving ordinary words, but not the registered marker","yields_valid_marker":false},{"from":"while-contrast","to":"while contrast","yields":"hyphen loss leaves direction-preserving ordinary words, but not the registered marker","yields_valid_marker":false},{"from":"while-overlap","to":"while","yields":"dropping the relation suffix restores the temporal-versus-contrast ambiguity","yields_valid_marker":false},{"from":"while-contrast","to":"while","yields":"dropping the relation suffix restores the temporal-versus-contrast ambiguity","yields_valid_marker":false},{"from":"while-overlap(E; C)","to":"while-overlap(non-interval-ref; C)","yields":"a visible type error because the first argument does not resolve to an interval-bearing event or state","yields_valid_marker":false}]
+ [{"from":"while-overlap","to":"while overlap","yields":"hyphen loss leaves direction-preserving ordinary words, but not the registered marker","yields_valid_marker":false},{"from":"while-contrast","to":"while contrast","yields":"hyphen loss leaves direction-preserving ordinary words, but not the registered marker","yields_valid_marker":false},{"from":"while-throughout","to":"while throughout","yields":"hyphen loss leaves direction-preserving ordinary words, but not the registered marker","yields_valid_marker":false},{"from":"while-overlap","to":"while","yields":"dropping the relation suffix restores the temporal-versus-contrast ambiguity","yields_valid_marker":false},{"from":"while-contrast","to":"while","yields":"dropping the relation suffix restores the temporal-versus-contrast ambiguity","yields_valid_marker":false},{"from":"while-throughout","to":"while","yields":"dropping the relation suffix erases the universal whole-interval obligation","yields_valid_marker":false},{"from":"while-throughout","to":"while-overlap","yields":"a valid but weaker marker that turns an all-instants obligation into a some-instants claim","yields_valid_marker":true},{"from":"while-overlap(E; C)","to":"while-overlap(non-interval-ref; C)","yields":"a visible type error because the first argument does not resolve to an interval-bearing event or state","yields_valid_marker":false}]
|
