Agentic Synthesis against Counterexample-Supplemented Sketches

Probably not a substantial claim: may not pass S3 (load-bearing).

CE7 makes the sketch join structured rules and narrative tags by the semantic triple of cat attribute, operator, and value.

“CE7 makes the sketch join structured rules and narrative tags by the semantic triple of cat attribute, operator, and value.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In Generation 007, the next profile supplied unknown for allergy status.

“The next profile supplies unknown for allergy status.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Replay and semantic compare reject different failures.

“Replay and semantic compare reject different failures.”

Probably not a substantial claim: may not pass S3 (load-bearing).

CE4 added a rule that avoid_needy applies a one-point soft penalty to breeds with high sociability.

“• avoid_needy applies a one-point soft penalty to breeds with high sociability;”

Probably not a substantial claim: may not pass S3 (load-bearing).

In Generation 006, before repair, the prompt emitted no tag.

“Before repair, the prompt emits no tag.”

Classified as probably not a substantial claim.

CE6 added a general rule: every distinct applicable discourage predicate contributes before the final ranking and tie-break.

“CE6 adds a general rule: every distinct applicable discourage predicate contributes before the final ranking and tie-break.”

Classified as probably not a substantial claim.

A candidate can repair state and still choose the wrong policy field.

“A candidate can repair state and still choose the wrong policy field.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

CatSynth retains every accepted case as a regression because the run is small, so R = A in this example.

“CatSynth also retains every accepted case as a regression because the run is small, so R = A in this example.”

May combine two claims.

After CE3, the gate passed 4/4 even though the strategy still chose Balinese.

“The gate passes 4/4 even though the strategy still chooses Balinese, because CE3 has not yet added a deterministic meaning for the tag.”

May combine two claims.

Tokens through visible acceptance are 1,021,822 for retained Sketch-CE, 891,880 for replay-all, and 828,628 for evolved-sketch rebuild.

“Tokens through visible acceptance are 1,021,822 for retained Sketch-CE, 891,880 for replay-all, and 828,628 for evolved-sketch rebuild.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In Generation 005, the approved output is British Shorthair.

“The approved output is British Shorthair.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The Sketch-CE repair metadata identifies one active failure and no unrevealed case.

“The Sketch-CE repair metadata identifies one active failure and no unrevealed case.”

Classified as probably not a substantial claim.

The synthetic policy row says that mild or severe allergies activate a hard forbid rule for breeds whose hypoallergenic field is false.

“The synthetic policy row says that mild or severe allergies activate a hard forbid rule for breeds whose hypoallergenic field is false.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The Sketch-CE total comprises 217,576 Developer tokens, 657,478 Runtime Oracle tokens, and 146,768 Specification Oracle tokens.

“The Sketch-CE total comprises 217,576 Developer tokens, 657,478 Runtime Oracle tokens, and 146,768 Specification Oracle tokens.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The second counterexample activates five hard rules: severe allergies, apartment constraints, long work hours, and young children.

“The second counterexample activates five hard rules: severe allergies, apartment constraints, long work hours, and young children.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The gate passed 5/5 after CE4.

“The gate passes 5/5: initial anchor: PASS CE1: PASS CE2: PASS CE3 tags: PASS CE4 ranking: PASS gate: 5/5”

Probably not a substantial claim: may not pass S3 (load-bearing).

In the CatSynth worked example the naive preference-only output fails semantic compare because it does not preserve the approved breed and cited-rule fields (the approved expectation names Siberian and cites the hard allergy rule).

“Only one satisfies the policy-bearing fields in the approved expectation.”

• Also stated in: § Additional information · plain

  “Semantic compare rejects the output because the approved expectation names Siberian and cites the hard allergy rule.”

• AssumptionRests on an assumption: The approved expected rows and operator-approved counterexamples correctly encode the intended policy, so that passing semantic compare reflects genuine policy correctness rather than the paper's own fixture choices.

Classified as probably not a substantial claim.

In the CatSynth worked example the naive preference-only output (Persian) closes the encoded state gap, so replay accepts it.

“Both candidates close the state gap defined by the replay predicate.”

• Also stated in: § Additional information · plain

  “Replay accepts Persian because the fixture marks it as large, fluffy, and affectionate.”

Classified as probably not a substantial claim.

CE3 added one controlled narrative classification to the evolved sketch: travel, repeated absence, or concern about loneliness maps to exactly avoid_needy.

“CE3 adds one controlled narrative classification to the evolved sketch: • travel, repeated absence, or concern about loneliness maps to exactly avoid_needy;”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

The retained strategy had 298 final strategy LOC and 110 final decision nodes, the largest and most decision nodes among the paths.

“Final strategy LOC 224 228 298 Final decision nodes 77 70 110”

Classified as probably not a substantial claim. May combine two claims.

After CE3, the prompt emits avoid_needy, but the strategy still recommends Balinese.

“After CE3, the prompt emits avoid_needy, but the strategy still recommends Balinese.”

Probably not a substantial claim: may not pass S3 (load-bearing).

CE4 added a rule that base preference scoring continues to use only the three explicit wants_* fields: size, affection, and fluffiness.

“• base preference scoring continues to use only the three explicit wants_* fields: size, affection, and fluffiness.”

Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

The complete gate passed 8/8 after Generation 007.

“The complete gate passes 8/8.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In Generation 006, the next profile expressed the same high-energy concern twice: once through a structured policy row and once through a narrative note.

“The next profile expresses the same high-energy concern twice: once through a structured policy row and once through a narrative note.”

May combine two claims.

In Generation 008, the current strategy silently ignored the unsupported cat operator and recommended Persian.

“The current strategy silently ignores it and recommends Persian.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The initial preference anchor passed with gate 1/1.

“The initial preference anchor passes: initial-preference-ranking: PASS gate: 1/1”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the CatSynth worked example the policy resolver selects Siberian, which satisfies the same encoded preferences and the hard allergy rule, so both replay and compare accept its output.

“The policy resolver selects Siberian because the fixture marks it as satisfying the same preferences and the hard allergy rule.”

• Also stated in: § Additional information · plain

  “Siberian preferences satisfied hard rule respected replay accepts; compare accepts”

• AssumptionRests on an assumption: The approved expected rows and operator-approved counterexamples correctly encode the intended policy, so that passing semantic compare reflects genuine policy correctness rather than the paper's own fixture choices.

May combine two claims.

The closed-world run reaches 20/20 visible and passes 21/21 withheld cases after four Developer calls.

“It reaches 20/20 visible and passes 21/21 withheld cases after four Developer calls.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

A maintainer can audit which predicates passed, which regressions they covered, which archived rules those checks represent, and which questions remain outside the proof.

“A maintainer can audit which predicates passed, which regressions they covered, which archived rules those checks represent, and which questions remain outside the proof.”

Classified as probably not a substantial claim. May combine two claims.

In Generation 007, the current strategy treated unknown allergy status like no allergy and recommended Persian.

“The current strategy treats it like no allergy and recommends Persian.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

CE12 requires escalate and cites invalid_reviewer_policy, preserving which policy source needs repair.

“CE12 requires escalate and cites invalid_reviewer_policy, preserving which policy source needs repair.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Post-acceptance visible and withheld evaluation in the closed-world run consumes another 239,929, for 851,448 recorded tokens in total.

“Post-acceptance visible and withheld evaluation consumes another 239,929, for 851,448 recorded tokens in total.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Every one of the eight accepted CatSynth counterexamples changes SKETCH.md.

“Every one of the eight accepted CatSynth counterexamples changes SKETCH.md.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The current implementation understands the first allergy operator but still returns Balinese and cites only that rule.

“The current implementation understands the first allergy operator but still returns Balinese and cites only that rule.”

Classified as probably not a substantial claim.

The accepted counterexample adds a general ordering rule to the sketch: hard-policy filtering must precede preference ranking, so preference match never overrides a hard rule.

“The accepted counterexample adds a sketch rule: preference match never overrides a hard rule.The sketch tells the next agent that filtering precedes ranking.”

• Also stated in: § Additional information · plain

  “The correction adds a general ordering rule: filter hard-policy violations before ranking the survivors.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Default activity, noise, and experience values do not add score unless an approved sketch clause gives them semantics.

“Default activity, noise, and experience values do not add score unless an approved sketch clause gives them semantics.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the worked example, Persian closes the visible preference gap, and Siberian also closes it while satisfying the approved hard rule.

“Persian closes the visible preference gap; Siberian also closes it while satisfying the approved hard rule.”

Classified as probably not a substantial claim.

After the Developer revises the sketch and strategy.py, CatSynth's gate passes both the initial anchor and CE1 (gate: 2/2).

“CatSynth's R = A gate then passes the initial anchor and CE1: initial-preference-ranking: PASS ce-001-allergy-override: PASS gate: 2/2”

• AssumptionRests on an assumption: The approved expected rows and operator-approved counterexamples correctly encode the intended policy, so that passing semantic compare reflects genuine policy correctness rather than the paper's own fixture choices.

Probably not a substantial claim: may not pass S3 (load-bearing).

The approved output for the Generation 006 case requires avoid_high_energy.

“The approved output requires avoid_high_energy, while the ranking must apply the shared energy predicate only once.”

Probably not a substantial claim: may not pass S3 (load-bearing).

A selected regression preserves the near miss as an executable check.

“A selected regression preserves the near miss as an executable check.”

May combine two claims.

The withheld-case results are 18/21 for Sketch-CE, 15/21 for replay-all, and 19/21 for evolved-sketch rebuild.

“Their withheld-case results are 18/21, 15/21, and 19/21 respectively.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In Generation 006, the ranking must apply the shared energy predicate only once.

“The approved output requires avoid_high_energy, while the ranking must apply the shared energy predicate only once.”

Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

The complete gate passed 6/6 after Generation 005.

“Developer revises the sketch and strategy; the complete gate passes 6/6.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The complete gate passed 7/7 after Generation 006.

“Developer revises the sketch, prompt, and strategy; the complete gate passes 7/7.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The candidate repaired the visible preference state but did not follow the approved policy.

“The candidate repaired the visible preference state but did not follow the approved policy.”

May combine two claims.

CatSynth makes the replay and semantic-compare predicates visible: the naive strategy repairs the encoded preference gap so replay passes, but it does not preserve the approved breed and cited-rule fields so semantic compare fails.

“CatSynth makes the two predicates visible: the naive strategy repairs the encoded preference gap, so replay passes, but it does not preserve the approved breed and cited-rule fields, so semantic compare fails.”

May combine two claims.

The final Sketch-CE strategy is the largest and most branch-heavy: 298 lines and 110 decision nodes, versus 224/77 for replay-all and 228/70 for evolved-sketch rebuild.

“Its final strategy is also the largest and most branch-heavy: 298 lines and 110 decision nodes, versus 224/77 for replay-all and 228/70 for evolved-sketch rebuild.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The run consumed 173 model calls and 3,251,645 recorded model tokens across all three paths, including post-acceptance evaluation.

“It consumed 173 model calls and 3,251,645 recorded model tokens across all three paths, including post-acceptance evaluation.”

May combine two claims.

Reaching visible acceptance in the closed-world run consumes 611,519 model tokens: 132,632 for Developer calls and 478,887 for prompt-mediated Runtime Oracle checks.

“Reaching visible acceptance consumes 611,519 model tokens: 132,632 for Developer calls and 478,887 for prompt-mediated Runtime Oracle checks.”

Classified as probably not a substantial claim.

The deterministic meaning of avoid_needy remained an open hole after CE3.

“• the deterministic meaning of avoid_needy remains an open hole.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

All recorded model tokens, including evaluation, were 1,061,834 for Replay-all, 998,307 for Evolved-sketch rebuild, and 1,191,504 for Sketch-CE (retained code).

“All recorded model tokens, including evaluation 1,061,834 998,307 1,191,504”

Classified as probably not a substantial claim.

CE10 distinguishes missing or unsupported safety data from a negative value: the implementation must return escalate with no breed until an operator can clarify the input.

“CE10 distinguishes missing or unsupported safety data from a negative value: the implementation must return escalate with no breed until an operator can clarify the input.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The browser's Review button records only a local operator decision in SQLite and does not invoke Developer or revise SKETCH.md.

“The browser's Review button records only a local operator decision in SQLite. It does not invoke Developer or revise SKETCH.md.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

That claim prevents the last plausible repair from disappearing back into a chat transcript, test failure, or reviewer memory.

“That claim prevents the last plausible repair from disappearing back into a chat transcript, test failure, or reviewer memory.”

Classified as probably not a substantial claim.

The agent searches across code, prompts, fixtures, and checks.

“The boundary differs in an ordinary repository.The agent searches across code, prompts, fixtures, and checks.An accepted counterexample records the rule that the repair must preserve and the known bad repair that the gate must reject.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Here the example set serves a repository repair loop.

“Here the example set serves a repository repair loop.Each accepted case ties an SME correction to a sketch clause, an implementation or prompt surface, and a replay/compare check.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Each accepted case ties an SME correction to a sketch clause, an implementation or prompt surface, and a replay/compare check.

“Here the example set serves a repository repair loop.Each accepted case ties an SME correction to a sketch clause, an implementation or prompt surface, and a replay/compare check.”

Probably not a substantial claim: may not pass S4 (based).

The gate makes a bounded claim over the current sketch, regression set, implementation surface, replay predicate, and semantic compare predicate.

“The gate makes a bounded claim over the current sketch, regression set, implementation surface, replay predicate, and semantic compare predicate.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The boundary differs in an ordinary repository.

“The boundary differs in an ordinary repository.The agent searches across code, prompts, fixtures, and checks.An accepted counterexample records the rule that the repair must preserve and the known bad repair that the gate must reject.”

Classified as probably not a substantial claim. May combine two claims.

An accepted counterexample records the rule that the repair must preserve and the known bad repair that the gate must reject.

“The boundary differs in an ordinary repository.The agent searches across code, prompts, fixtures, and checks.An accepted counterexample records the rule that the repair must preserve and the known bad repair that the gate must reject.”

Probably not a substantial claim: may not pass S4 (based).

A repository-native workflow is inspectable when every accepted counterexample connects its operator decision, archive record, and reviewed sketch clause.

“Claim.A repository-native workflow is inspectable when every accepted counterexample connects its operator decision, archive record, and reviewed sketch clause.”

Classified as probably not a substantial claim. May combine two claims.

The evidence boundary is narrower than the deployment plumbing: SMEs judged concrete outputs; operators approved policy-changing counterexamples; the sketch evolved; and the golden subset made implementation regressions visible.

“The evidence boundary is narrower than that plumbing: SMEs judged concrete outputs; operators approved policy-changing counterexamples; the sketch evolved; and the golden subset made implementation regressions visible.”

Classified as probably not a substantial claim.

Passing the gate does not claim correctness outside those artifacts.

“Passing the gate gives maintainers a bounded invariant over the current regression set, replay/checker code, and approved expected rows.It does not claim correctness outside those artifacts.”

Classified as probably not a substantial claim.

Replay and compare must stay separate because they catch different wrong repairs.

“Replay and compare must stay separate because they catch different wrong repairs.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The general method does not require the full archive to remain in the executable regression set.

“The general method does not require the full archive to remain in the executable regression set.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A comparer with a bug can accept the wrong behavior.

“A comparer with a bug can accept the wrong behavior.”

Probably not a substantial claim: may not pass S2 (checkable).

A passing test suite covers only encoded behavior.

“A passing test suite covers only encoded behavior.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Within the open-world experiment, the rebuild controls inherit the promoted stream without paying to evaluate the external candidates or propose rules for failures.

“Within the open-world experiment, the rebuild controls inherit the promoted stream without paying to evaluate the external candidates or propose rules for failures.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Rules left outside the sketch still rely on human memory.

“Rules left outside the sketch still rely on human memory.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Developer does not receive unrevealed cases, the archive, or the regression set as bulk prompt context.

“It does not receive unrevealed cases, the archive, or the regression set as bulk prompt context.”

Classified as probably not a substantial claim.

In a separate closed-world run, the model receives an immutable complete specification and empty implementation files.

“In a separate closed-world run, the model receives an immutable complete specification and empty implementation files.”

Classified as probably not a substantial claim.

The archive and regression set are not bulk repair context.

“The archive and regression set are not bulk repair context.With one active failure, the agent has a narrower job: revise the sketch with the general rule, reject the known tempting patch, and keep the codebase's existing shape.”

Classified as probably not a substantial claim.

The code can be discarded and regenerated.

“The code can be discarded and regenerated.”

Classified as probably not a substantial claim.

The archive carries the decision history.

“The archive carries the decision history.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The maintainer makes those decisions durable: the runtime runs, the gate checks the right fields, the output shape stays compatible, accepted rules remain inspectable, and a clean implementation can be rebuilt from the evolved sketch.

“SMEs supply the judgments and operators authorize policy changes.The maintainer makes those decisions durable: the runtime runs, the gate checks the right fields, the output shape stays compatible, accepted rules remain inspectable, and a clean implementation can be rebuilt from the evolved sketch.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The archive is complete; the regression set is intentionally selective.

“The archive is complete; the regression set is intentionally selective.”

May combine two claims.

It does not establish universal cost, correctness, or maintainability superiority.

“It does not establish universal cost, correctness, or maintainability superiority.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A buggy checker can certify the wrong behavior until a maintainer repairs the checker and reruns the regression set.

“A buggy checker can certify the wrong behavior until a maintainer repairs the checker and reruns the regression set.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The first row of the Measure table includes every recorded Developer, Runtime Oracle, Specification Oracle, and postacceptance evaluation call.

“The first row includes every recorded Developer, Runtime Oracle, Specification Oracle, and postacceptance evaluation call.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Living-specification workflows can support the same kind of learning.

“Living-specification workflows can support the same kind of learning.The narrower distinction is procedural: Sketch-CE places explicit counterexample approval and reviewed sketch revision inside the repair protocol instead of treating them as exceptional upstream processes.”

Classified as probably not a substantial claim. May combine two claims.

In Argyris's terms, regression repair is single-loop correction; counterexample acceptance is double-loop learning because the correction changes the governing policy or objective.

“In Argyris's terms, regression repair is single-loop correction; counterexample acceptance is double-loop learning because the correction changes the governing policy or objective [2].”

• cites [2]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The gate-soundness theorem is bounded to the regression set, the encoded replay checker, the encoded semantic comparer, and the current approved expected rows; it does not cover archived cases omitted from R, unapproved failures, unencoded policy fields, future model behavior, or private deployment behavior outside the reported experiment.

“The theorem is bounded to the regression set, the encoded replay checker, the encoded semantic comparer, and the current approved expected rows.It does not cover archived cases omitted from R, unapproved failures, unencoded policy fields, future model behavior, or private deployment behavior outside the reported experiment.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Provider totals count input plus output; cached input and reasoning are subsets, not additional tokens.

“Provider totals count input plus output; cached input and reasoning are subsets, not additional tokens.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Maintainers use it to hold the current governing model, including rules that are not yet fully encoded.

“The sketch is less formal than a full specification.Maintainers use it to hold the current governing model, including rules that are not yet fully encoded.They can discard code and prompt implementations, regenerate them from the sketch and anchors, and use the regression set to find what the new implementation lost.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Both controls inherited the promotion schedule, and evolved-sketch rebuild also inherited the sketch checkpoints.

“Both controls inherited the promotion schedule, and evolved-sketch rebuild also inherited the sketch checkpoints.”

Classified as probably not a substantial claim.

Sketch-CE makes the destination of feedback explicit.

“Sketch-CE makes the destination of feedback explicit.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

It also does not treat the complete historical archive as generation context or require every archived row to run forever.

“Separating A from R prevents two mistakes.The team does not discard the history behind an evolved rule merely because its original case is redundant in CI.It also does not treat the complete historical archive as generation context or require every archived row to run forever.”

May combine two claims.

It does not prove behavior on cases outside R or on policy fields the checker does not encode.

“It does not prove behavior on cases outside R or on policy fields the checker does not encode.”

Probably not a substantial claim: may not pass S4 (based).

A wrong row in the golden dataset makes the gate enforce the wrong behavior.

“A wrong row in the golden dataset makes the gate enforce the wrong behavior.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The naive Persian result can therefore pass replay and fail semantic compare.

“The naive Persian result can therefore pass replay and fail semantic compare:”

Probably not a substantial claim: may not pass S3 (load-bearing).

The worked example separates state repair from policy repair in one small case.

“The worked example separates state repair from policy repair in one small case.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A gate that omits a field cannot check that field.

“A gate that omits a field cannot check that field.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The closed-world run begins with information the open-world run must discover, so their token totals answer different questions.

“The closed-world run begins with information the open-world run must discover, so their token totals answer different questions.”

Classified as probably not a substantial claim.

When a gate fails, maintainers can ask which artifact needs work rather than silently treating every failure as new policy.

“When a gate fails, maintainers can ask which artifact needs work rather than silently treating every failure as new policy.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The archive answers why policy changed; the regression set asks whether the current implementation still obeys selected consequences.

“The archive answers why policy changed; the regression set asks whether the current implementation still obeys selected consequences.”

May combine two claims.

Sketch-CE with retained code used less Developer work and produced less cumulative churn.

“Sketch-CE with retained code used less Developer work and produced less cumulative churn.”

Probably not a substantial claim: may not pass S4 (based).

Separating A from R prevents two mistakes.

“Separating A from R prevents two mistakes.The team does not discard the history behind an evolved rule merely because its original case is redundant in CI.It also does not treat the complete historical archive as generation context or require every archived row to run forever.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Raw JSON-RPC transcripts stayed local; the repository retains every generated sketch, strategy, prompt, failure, gate outcome, diff, and usage total.

“Raw JSON-RPC transcripts stayed local; the repository retains every generated sketch, strategy, prompt, failure, gate outcome, diff, and usage total.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Needing the full archive to reconstruct policy is evidence that the sketch has not synthesized the accepted lessons well enough.

“Needing the full archive to reconstruct policy is evidence that the sketch has not synthesized the accepted lessons well enough.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The archive preserves that policy history, and the regression set checks selected consequences against later implementations.

“The archive preserves that policy history, and the regression set checks selected consequences against later implementations.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Tokens through acceptance exclude only the final visible and withheld evaluation.

“Tokens through acceptance exclude only the final visible and withheld evaluation.”

Probably not a substantial claim: may not pass S4 (based).

Periodic clean regeneration tests whether the evolved sketch, rather than retained implementation history, carries the learned policy.

“Periodic clean regeneration tests whether the evolved sketch, rather than retained implementation history, carries the learned policy.”

Classified as probably not a substantial claim. May combine two claims.

They can discard code and prompt implementations, regenerate them from the sketch and anchors, and use the regression set to find what the new implementation lost.

“The sketch is less formal than a full specification.Maintainers use it to hold the current governing model, including rules that are not yet fully encoded.They can discard code and prompt implementations, regenerate them from the sketch and anchors, and use the regression set to find what the new implementation lost.”

Classified as probably not a substantial claim.

CatSynth records one stochastic run per path with one model and candidate order.

“CatSynth records one stochastic run per path with one model and candidate order.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Approval quality controls the governing model; regression selection controls executable scale.

“Approval quality controls the governing model; regression selection controls executable scale.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The reported evidence covers that control structure, not private deployment behavior.

“The reported evidence covers that control structure, not private deployment behavior.”

Probably not a substantial claim: may not pass S4 (based).

A rule can begin in prompt-mediated completion, then move into deterministic code once enough counterexamples make it precise.

“A rule can begin in prompt-mediated completion, then move into deterministic code once enough counterexamples make it precise.”

Probably not a substantial claim: may not pass S2 (checkable).

The archive provides provenance, not proof.

“The archive provides provenance, not proof.”

Classified as probably not a substantial claim. May combine two claims.

The agent can repair Oracle A because the rule has been made concrete enough to encode.

“The agent can repair Oracle A because the rule has been made concrete enough to encode.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The conceptual contrast is spec-first versus Sketch-CE: a complete specification works when the problem is already known, while Sketch-CE changes the governing sketch as the world reveals new policy.

“The conceptual contrast remains spec-first versus Sketch-CE: a complete specification works when the problem is already known, while Sketch-CE changes the governing sketch as the world reveals new policy.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Tests check behavior in the current run; sketch clauses preserve why a failure mattered for the next repair.

“Tests check behavior in the current run; sketch clauses preserve why a failure mattered for the next repair.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A clean implementation should be regenerable from the evolved sketch and known-code anchors without replaying the CE archive as generation context.

“A clean implementation should be regenerable from the evolved sketch and known-code anchors without replaying the CE archive as generation context.”

Probably not a substantial claim: may not pass S2 (checkable).

Reporting the two layers separately lets maintainers distinguish state repair from policy repair.

“Reporting the two layers separately lets maintainers distinguish state repair from policy repair.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The sketch is less formal than a full specification.

“The sketch is less formal than a full specification.Maintainers use it to hold the current governing model, including rules that are not yet fully encoded.They can discard code and prompt implementations, regenerate them from the sketch and anchors, and use the regression set to find what the new implementation lost.”

May combine two claims.

Sketch-CE reduces implementation work and churn in this run, but does not achieve the highest withheld-case rate.

“Sketch-CE reduces implementation work and churn in this run, but does not achieve the highest withheld-case rate.”

Classified as probably not a substantial claim.

The team does not discard the history behind an evolved rule merely because its original case is redundant in CI.

“Separating A from R prevents two mistakes.The team does not discard the history behind an evolved rule merely because its original case is redundant in CI.It also does not treat the complete historical archive as generation context or require every archived row to run forever.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The rebuild controls preserve each generated state and the visible failures returned after a failed gate.

“The rebuild controls preserve each generated state and the visible failures returned after a failed gate.”

Classified as probably not a substantial claim.

The withheld-case failures indicate where another open-world counterexample could extend the current sketch.

“Those failures show where another open-world counterexample could extend the current sketch.”

Classified as probably not a substantial claim. The quote only partly states this claim.

The captured open-world experiment adds two controls (Replay-all and Evolved-sketch rebuild) to identify what carries the policy forward.

“The captured open-world experiment adds two controls to identify what carries that policy forward.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Their totals omit discovery work and are not end-to-end price rankings.

“Their totals omit discovery work and are not end-to-end price rankings.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The informational restriction prevents Developer from copying future counterexamples into the sketch because it never sees them.

“The informational restriction prevents Developer from copying future counterexamples into the sketch because it never sees them.”

Probably not a substantial claim: may not pass S4 (based).

The initial sketch contains enough detail to generate and test an initial strategy without preloading the later policy discoveries.

“It contains enough detail to generate and test an initial strategy without preloading the later policy discoveries.”

Classified as probably not a substantial claim. May combine two claims.

With one active failure, the agent has a narrower job: revise the sketch with the general rule, reject the known tempting patch, and keep the codebase's existing shape.

“The archive and regression set are not bulk repair context.With one active failure, the agent has a narrower job: revise the sketch with the general rule, reject the known tempting patch, and keep the codebase's existing shape.”

Classified as probably not a substantial claim. The quote only partly states this claim.

A reader can therefore inspect the information boundary and code evolution directly from the retained artifacts.

“A reader can therefore inspect the information boundary and code evolution directly.”

May combine two claims.

Accepting a new counterexample expands A and changes S; it does not mechanically enlarge R.

“Accepting a new counterexample expands A and changes S; it does not mechanically enlarge R.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The CatSynth experiments do not establish a general model advantage.

“not establish a general model advantage.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Readers can inspect the control structure in CatSynth's synthetic fixtures.

“Readers can inspect the control structure in CatSynth's synthetic fixtures.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Live model behavior can drift.

“Live model behavior can drift.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

The repository loop uses those disciplines at a narrower boundary.

“The repository loop uses those disciplines at a narrower boundary.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The controls inherit the resulting promotion schedule, so these totals have different accounting boundaries and are not end-to-end price rankings.

“The controls inherit the resulting promotion schedule, so these totals have different accounting boundaries and are not end-to-end price rankings.”

May combine two claims.

The retained strategy path shows less rework during evolution, not better final maintainability.

“The retained strategy was the largest and had the most decision nodes, so that path shows less rework during evolution, not better final maintainability.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

A CE that leaves the sketch unchanged was misclassified or incompletely processed.

“A CE that leaves the sketch unchanged was misclassified or incompletely processed.”

Classified as probably not a substantial claim. May combine two claims.

The candidate cases and authoritative expected outputs were frozen before the run and treated as a simulated operator-approved stream.

“The candidate cases and authoritative expected outputs were frozen before the run and treated as a simulated operator-approved stream.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The observed token ratios and checked results are not a benchmark or a causal estimate.

“The observed token ratios and checked results are not a benchmark or a causal estimate.”

Classified as probably not a substantial claim.

A passing gate does not prove the domain correct.

“A passing gate does not prove the domain correct.”

Probably not a substantial claim: may not pass S2 (checkable).

The sketch carries the learned policy.

“The sketch carries the learned policy.”

Classified as probably not a substantial claim.

The regression set tests generated implementations.

“The regression set tests generated implementations.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Replay and compare are only as good as the semantics encoded in their checkers.

“Replay and compare are only as good as the semantics encoded in their checkers.”

Classified as probably not a substantial claim. May combine two claims. The quote only partly states this claim.

The gate's guarantee is limited to the current regression set: it does not prove the program correct or replay the full history of policy discovery, but answers the narrower question of whether, for every selected row, the candidate repairs the encoded state and preserves the encoded policy fields.

“It does not prove the program correct or replay the full history of policy discovery.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

A small R can omit an important boundary; an indiscriminate one can become expensive noise.

“A small R can omit an important boundary; an indiscriminate one can become expensive noise.”

Probably not a substantial claim: may not pass S4 (based).

For the current repository, current checkers, and current regression set, replay and semantic comparison pass.

“It says only this: for the current repository, current checkers, and current regression set, replay and semantic comparison pass.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The gate checks the finite regression set with current replay, compare, and data; people still own counterexample approval, regression selection, checker design, golden review, sketch maintenance, prompt review, and any private deployment claim.

“The gate checks the finite regression set with current replay, compare, and data; people still own counterexample approval, regression selection, checker design, golden review, sketch maintenance, prompt review, and any private deployment claim.”

Probably not a substantial claim: may not pass S2 (checkable).

Everything outside those encoded cases and predicates still belongs to human review and future discovery.

“Everything outside those encoded cases and predicates still belongs to human review and future discovery.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The archive may contain accepted cases not selected for routine execution, and the result does not extend to unseen cases.

“The archive may contain accepted cases not selected for routine execution, and the result does not extend to unseen cases.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Oracle B remains constrained by the sketch and checked against selected SME-approved regressions by replay and semantic compare.

“Oracle B remains constrained by the sketch and checked against selected SME-approved regressions by replay and semantic compare.”

Probably not a substantial claim: may not pass S4 (based).

A passing gate proves only the current R.

“A passing gate proves only the current R.”

Classified as probably not a substantial claim.

The narrower distinction is procedural: Sketch-CE places explicit counterexample approval and reviewed sketch revision inside the repair protocol instead of treating them as exceptional upstream processes.

“Living-specification workflows can support the same kind of learning.The narrower distinction is procedural: Sketch-CE places explicit counterexample approval and reviewed sketch revision inside the repair protocol instead of treating them as exceptional upstream processes.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

A replay failure means the state is still broken, whereas a compare failure means the state repair passed while a policy field diverged.

“Read the failure by its job: replay failure means the state is still broken; compare failure means the state repair passed while a policy field diverged.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

CE4 is a genuine new counterexample rather than another explanation pasted into CE3, because it fails on the breed field.

“It fails on the breed field, so CE4 is a genuine new counterexample rather than another explanation pasted into CE3.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The retained histories let a reader answer five questions without relying on the paper's prose.

“The retained histories let a reader answer five questions without relying on the paper's prose:”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Any correctness claim is bounded by the current regression set and the current checkers.

“Any correctness claim is bounded by the current regression set and the current checkers.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Neither the gate result nor the withheld cases establish correctness for all owner profiles, real breed facts, bad expected outputs or checkers, unencoded policy, future model behavior, or other models and reveal orders.

“Neither result establishes correctness for all owner profiles, real breed facts, bad expected outputs or checkers, unencoded policy, future model behavior, or other models and reveal orders.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The withheld cases provide limited evidence beyond the gate.

“The withheld cases provide limited evidence beyond the gate.”

Classified as probably not a substantial claim.

The experiment history, not the Review button, is the evidence for the complete CE-to-sketch-and-code loop.

“The experiment history, not that button, is the evidence for the complete CE-to-sketch-and-code loop.”

Probably not a substantial claim: may not pass S3 (load-bearing).

CatSynth preserves the evidence needed to inspect that boundary.

“CatSynth preserves the evidence needed to inspect that boundary.”