{"id":"f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09","title":"Reda Heddad’s arXiv paper tests runtime contracts for coding agents","argument":"Reda Heddad’s paper, “Beyond Prompts: Runtime Contracts for Coding Agents,” tests whether executable contracts—rules enforced while an agent works—prevent coding-agent rule violations better than prompt text, retrieved guidance or checks after generation. Heddad argues that prompt text provides neither enforcement nor evidence that rules were followed, a gap for teams that expect both. Across three studies, enforcement during execution produced no oracle-detected violation in 104 paired runs, while 21 to 55% of runs without rules had violations. In a confirmatory study targeting known failure situations, enforcement reduced violating runs by 31 percentage points relative to no rules. The 17-point difference from prompted rules was not significant, and neither were contrasts in another study, where few tasks produced violations. The paper reports no loss of functional correctness in its two confirmatory studies. Heddad bounds the contribution: enforcement satisfies the rules it enforces by construction; the data instead address where prompting falls short, the absence of a functional cost, and limits in the rules and oracle. The studies used low-cost models and one small fixture, so the abstract does not establish how results generalize beyond those settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports that the studies used low-cost models and one small fixture; it does not establish how results generalize beyond those settings. The paper notes that violations are scored by an oracle and describes limits of the rules and oracle themselves.","speaker":{"name":"Heddad, Reda","roleAtSource":"Article author"},"source":{"title":"Beyond Prompts: Runtime Contracts for Coding Agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.19549","publishedAt":"2026-11-14","dateBasis":"Date basis not provided."},"evidence":[{"text":"Reda Heddad argues that prompt text alone does not enforce engineering rules or provide evidence of their application.","excerpt":"a rule given as prompt text provides neither","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"Across the reported studies, enforcement during execution produced no oracle-detected violation in 104 paired runs; runs without rules had violations at rates of 21 to 55%.","excerpt":"enforcement during execution produced no oracle-detected violation in 104 paired runs, against 21 to 55% of runs without rules","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"In Study 3, enforcement reduced the share of violating runs by 31 percentage points relative to no rules; its 17-point difference from prompted rules was not significant.","excerpt":"enforcement lowered the share of violating runs by 31 percentage points relative to no rules (Holm-adjusted p = 0.014); the 17-point difference from prompted rules was not significant (p = 0.12)","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"The paper reports no loss of functional correctness in its two confirmatory studies.","excerpt":"no loss of functional correctness in the two confirmatory studies","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"The authors characterize the evidence as bounded by the rules and oracle, and note that the studies used low-cost models and one small fixture.","excerpt":"the limits of the rules and of the oracle themselves. The studies use low-cost models and one small fixture.","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.19549","checkedAt":"2026-11-17T09:50:09.322Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would runtime contracts perform beyond the low-cost models and one small fixture used in these studies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Reda Heddad’s arXiv paper tests runtime contracts for coding agents","summary":"Reda Heddad’s paper, “Beyond Prompts: Runtime Contracts for Coding Agents,” tests whether executable contracts—rules enforced while an agent works—prevent coding-agent rule violations better than prompt text, retrieved guidance or checks after generation. Heddad argues that prompt text provides neither enforcement nor evidence that rules were followed, a gap for teams that expect both. Across three studies, enforcement during execution produced no oracle-detected violation in 104 paired runs, while 21 to 55% of runs without rules had violations. In a confirmatory study targeting known failure situations, enforcement reduced violating runs by 31 percentage points relative to no rules. The 17-point difference from prompted rules was not significant, and neither were contrasts in another study, where few tasks produced violations. The paper reports no loss of functional correctness in its two confirmatory studies. Heddad bounds the contribution: enforcement satisfies the rules it enforces by construction; the data instead address where prompting falls short, the absence of a functional cost, and limits in the rules and oracle. The studies used low-cost models and one small fixture, so the abstract does not establish how results generalize beyond those settings.","question":"How would runtime contracts perform beyond the low-cost models and one small fixture used in these studies?","claims":[{"text":"Reda Heddad argues that prompt text alone does not enforce engineering rules or provide evidence of their application.","excerpt":"a rule given as prompt text provides neither","support":"source-stated"},{"text":"Across the reported studies, enforcement during execution produced no oracle-detected violation in 104 paired runs; runs without rules had violations at rates of 21 to 55%.","excerpt":"enforcement during execution produced no oracle-detected violation in 104 paired runs, against 21 to 55% of runs without rules","support":"source-stated"},{"text":"In Study 3, enforcement reduced the share of violating runs by 31 percentage points relative to no rules; its 17-point difference from prompted rules was not significant.","excerpt":"enforcement lowered the share of violating runs by 31 percentage points relative to no rules (Holm-adjusted p = 0.014); the 17-point difference from prompted rules was not significant (p = 0.12)","support":"source-stated"},{"text":"The paper reports no loss of functional correctness in its two confirmatory studies.","excerpt":"no loss of functional correctness in the two confirmatory studies","support":"source-stated"},{"text":"The authors characterize the evidence as bounded by the rules and oracle, and note that the studies used low-cost models and one small fixture.","excerpt":"the limits of the rules and of the oracle themselves. The studies use low-cost models and one small fixture.","support":"source-stated"}],"limitations":["The abstract reports that the studies used low-cost models and one small fixture; it does not establish how results generalize beyond those settings.","The paper notes that violations are scored by an oracle and describes limits of the rules and oracle themselves."],"mediaId":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","correctionReason":null,"id":"f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09","publicationId":"ba2ae164-30ad-4edb-83d7-a8150292d6ca","version":1,"sourceUrl":"https://arxiv.org/abs/2611.19549","sourceTitle":"Beyond Prompts: Runtime Contracts for Coding Agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-14","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-17T08:31:37.880Z","createdAt":"2026-11-17T09:18:38.216Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","kind":"article","url":"https://arxiv.org/abs/2611.19549","sourceUrl":"https://arxiv.org/abs/2611.19549","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-17T09:17:49.856Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.19549","checkedAt":"2026-11-17T09:50:09.322Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","confidence":0.76,"provider":"typesafe","model":"jev-1.13.0","inputHash":"d66d8a2642274e2189414ad3f1e3f35a51df4f7461fc699873e7cfeffeb5591e","decidedAt":"2026-11-17T09:17:50.614Z"},"provenance":{"textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observationId":"dc8958fa77ecb5df729b302c60502cccdc217e71e5015693ddfb2fa15ebc0417","retrievedAt":"2026-11-17T09:17:49.748Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"127114d9-217e-43e6-8c9f-7f4774905508"},"reviewer":{"model":"gpt-6-luna","runId":"d58a6d27-46a9-48a3-8d3e-fe9fe6fcdd4d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-17T09:18:38.216Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Heddad, Reda","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Heddad, Reda","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Heddad, Reda","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.19549","locator":"head/meta[name=author|citation_author]"},"assertionId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-17T09:50:09.322Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386","sourceUrl":"https://arxiv.org/abs/2611.19549","contentSha256":"687b847ce28f39c5646b0bd4e3c573261e388fd0a35c8e4c72fea7ccd4fcb7c1","textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observedAt":"2026-11-17T09:50:09.322Z"}}],"participationCoverage":{"total":1,"returned":1,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.19549","snapshotId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386","contentSha256":"687b847ce28f39c5646b0bd4e3c573261e388fd0a35c8e4c72fea7ccd4fcb7c1","textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observedAt":"2026-11-17T09:50:09.322Z","basis":"publisher-page-metadata","sourceLinks":40,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09/source"}},"canonicalUrl":"https://escapemeridian.com/readings/reda-heddad-s-arxiv-paper-tests-runtime-contracts-for-coding-f73f5819","formats":{"html":"/readings/reda-heddad-s-arxiv-paper-tests-runtime-contracts-for-coding-f73f5819","markdown":"/readings/reda-heddad-s-arxiv-paper-tests-runtime-contracts-for-coding-f73f5819.md","json":"/readings/reda-heddad-s-arxiv-paper-tests-runtime-contracts-for-coding-f73f5819.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09","revision":"ba2ae164-30ad-4edb-83d7-a8150292d6ca","revisionKind":"publication-id","title":"Reda Heddad’s arXiv paper tests runtime contracts for coding agents","url":"https://escapemeridian.com/readings/reda-heddad-s-arxiv-paper-tests-runtime-contracts-for-coding-f73f5819"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.19549","revision":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","revisionKind":"sha256-source-text","title":"Beyond Prompts: Runtime Contracts for Coding Agents","url":"https://arxiv.org/abs/2611.19549"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter."},"revisions":[{"id":"f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09","title":"Reda Heddad’s arXiv paper tests runtime contracts for coding agents","argument":"Reda Heddad’s paper, “Beyond Prompts: Runtime Contracts for Coding Agents,” tests whether executable contracts—rules enforced while an agent works—prevent coding-agent rule violations better than prompt text, retrieved guidance or checks after generation. Heddad argues that prompt text provides neither enforcement nor evidence that rules were followed, a gap for teams that expect both. Across three studies, enforcement during execution produced no oracle-detected violation in 104 paired runs, while 21 to 55% of runs without rules had violations. In a confirmatory study targeting known failure situations, enforcement reduced violating runs by 31 percentage points relative to no rules. The 17-point difference from prompted rules was not significant, and neither were contrasts in another study, where few tasks produced violations. The paper reports no loss of functional correctness in its two confirmatory studies. Heddad bounds the contribution: enforcement satisfies the rules it enforces by construction; the data instead address where prompting falls short, the absence of a functional cost, and limits in the rules and oracle. The studies used low-cost models and one small fixture, so the abstract does not establish how results generalize beyond those settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports that the studies used low-cost models and one small fixture; it does not establish how results generalize beyond those settings. The paper notes that violations are scored by an oracle and describes limits of the rules and oracle themselves.","speaker":{"name":"Heddad, Reda","roleAtSource":"Article author"},"source":{"title":"Beyond Prompts: Runtime Contracts for Coding Agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.19549","publishedAt":"2026-11-14","dateBasis":"Date basis not provided."},"evidence":[{"text":"Reda Heddad argues that prompt text alone does not enforce engineering rules or provide evidence of their application.","excerpt":"a rule given as prompt text provides neither","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"Across the reported studies, enforcement during execution produced no oracle-detected violation in 104 paired runs; runs without rules had violations at rates of 21 to 55%.","excerpt":"enforcement during execution produced no oracle-detected violation in 104 paired runs, against 21 to 55% of runs without rules","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"In Study 3, enforcement reduced the share of violating runs by 31 percentage points relative to no rules; its 17-point difference from prompted rules was not significant.","excerpt":"enforcement lowered the share of violating runs by 31 percentage points relative to no rules (Holm-adjusted p = 0.014); the 17-point difference from prompted rules was not significant (p = 0.12)","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"The paper reports no loss of functional correctness in its two confirmatory studies.","excerpt":"no loss of functional correctness in the two confirmatory studies","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"},{"text":"The authors characterize the evidence as bounded by the rules and oracle, and note that the studies used low-cost models and one small fixture.","excerpt":"the limits of the rules and of the oracle themselves. The studies use low-cost models and one small fixture.","url":"https://arxiv.org/abs/2611.19549","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.19549","checkedAt":"2026-11-17T09:50:09.322Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would runtime contracts perform beyond the low-cost models and one small fixture used in these studies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Reda Heddad’s arXiv paper tests runtime contracts for coding agents","summary":"Reda Heddad’s paper, “Beyond Prompts: Runtime Contracts for Coding Agents,” tests whether executable contracts—rules enforced while an agent works—prevent coding-agent rule violations better than prompt text, retrieved guidance or checks after generation. Heddad argues that prompt text provides neither enforcement nor evidence that rules were followed, a gap for teams that expect both. Across three studies, enforcement during execution produced no oracle-detected violation in 104 paired runs, while 21 to 55% of runs without rules had violations. In a confirmatory study targeting known failure situations, enforcement reduced violating runs by 31 percentage points relative to no rules. The 17-point difference from prompted rules was not significant, and neither were contrasts in another study, where few tasks produced violations. The paper reports no loss of functional correctness in its two confirmatory studies. Heddad bounds the contribution: enforcement satisfies the rules it enforces by construction; the data instead address where prompting falls short, the absence of a functional cost, and limits in the rules and oracle. The studies used low-cost models and one small fixture, so the abstract does not establish how results generalize beyond those settings.","question":"How would runtime contracts perform beyond the low-cost models and one small fixture used in these studies?","claims":[{"text":"Reda Heddad argues that prompt text alone does not enforce engineering rules or provide evidence of their application.","excerpt":"a rule given as prompt text provides neither","support":"source-stated"},{"text":"Across the reported studies, enforcement during execution produced no oracle-detected violation in 104 paired runs; runs without rules had violations at rates of 21 to 55%.","excerpt":"enforcement during execution produced no oracle-detected violation in 104 paired runs, against 21 to 55% of runs without rules","support":"source-stated"},{"text":"In Study 3, enforcement reduced the share of violating runs by 31 percentage points relative to no rules; its 17-point difference from prompted rules was not significant.","excerpt":"enforcement lowered the share of violating runs by 31 percentage points relative to no rules (Holm-adjusted p = 0.014); the 17-point difference from prompted rules was not significant (p = 0.12)","support":"source-stated"},{"text":"The paper reports no loss of functional correctness in its two confirmatory studies.","excerpt":"no loss of functional correctness in the two confirmatory studies","support":"source-stated"},{"text":"The authors characterize the evidence as bounded by the rules and oracle, and note that the studies used low-cost models and one small fixture.","excerpt":"the limits of the rules and of the oracle themselves. The studies use low-cost models and one small fixture.","support":"source-stated"}],"limitations":["The abstract reports that the studies used low-cost models and one small fixture; it does not establish how results generalize beyond those settings.","The paper notes that violations are scored by an oracle and describes limits of the rules and oracle themselves."],"mediaId":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","correctionReason":null,"id":"f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09","publicationId":"ba2ae164-30ad-4edb-83d7-a8150292d6ca","version":1,"sourceUrl":"https://arxiv.org/abs/2611.19549","sourceTitle":"Beyond Prompts: Runtime Contracts for Coding Agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-14","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-17T08:31:37.880Z","createdAt":"2026-11-17T09:18:38.216Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","kind":"article","url":"https://arxiv.org/abs/2611.19549","sourceUrl":"https://arxiv.org/abs/2611.19549","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-17T09:17:49.856Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.19549","checkedAt":"2026-11-17T09:50:09.322Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"73214ad1ad5fabb447dd3a4dfe462fdcefc8b1e2e8dd117327df3957ad06b060","confidence":0.76,"provider":"typesafe","model":"jev-1.13.0","inputHash":"d66d8a2642274e2189414ad3f1e3f35a51df4f7461fc699873e7cfeffeb5591e","decidedAt":"2026-11-17T09:17:50.614Z"},"provenance":{"textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observationId":"dc8958fa77ecb5df729b302c60502cccdc217e71e5015693ddfb2fa15ebc0417","retrievedAt":"2026-11-17T09:17:49.748Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"127114d9-217e-43e6-8c9f-7f4774905508"},"reviewer":{"model":"gpt-6-luna","runId":"d58a6d27-46a9-48a3-8d3e-fe9fe6fcdd4d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-17T09:18:38.216Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Heddad, Reda","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Heddad, Reda","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Heddad, Reda","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.19549","locator":"head/meta[name=author|citation_author]"},"assertionId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-17T09:50:09.322Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386","sourceUrl":"https://arxiv.org/abs/2611.19549","contentSha256":"687b847ce28f39c5646b0bd4e3c573261e388fd0a35c8e4c72fea7ccd4fcb7c1","textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observedAt":"2026-11-17T09:50:09.322Z"}}],"participationCoverage":{"total":1,"returned":1,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.19549","snapshotId":"38c866ef045db34efebf5d03d5021f3ea08c367bd39e7b9e048fa2b646a68386","contentSha256":"687b847ce28f39c5646b0bd4e3c573261e388fd0a35c8e4c72fea7ccd4fcb7c1","textHash":"3d70c798cf85c51ac78455c464e4eab3387d210f5639214eb21d7798b1e9643d","observedAt":"2026-11-17T09:50:09.322Z","basis":"publisher-page-metadata","sourceLinks":40,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/f73f5819cc942e1aa62fd489cc7da9d6483c646814ddf7d786bb62502766ec09/source"}}}],"audits":[{"publicationId":"ba2ae164-30ad-4edb-83d7-a8150292d6ca","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The paper, title and sole author Reda Heddad are correctly identified, and findings are attributed to the paper. Claim 5 says 'The authors' for a single-author paper. This is minor wording, not a misattribution.","excerpt":"Title:Beyond Prompts: Runtime Contracts for Coding Agents\nAuthors:Reda Heddad"},{"dimension":"summary","status":"uncertain","explanation":"The figures are faithful, but the reading has two gaps. First, 'oracle' and 'fixture' are central yet unexplained. A general reader needs to know that the oracle is a separate scoring method, distinct from the in-loop checks and validated by two human reviewers and a model reviewer. Second, the summary drops the paper's concrete account of where prompting falls short: prompted rules did not stop recursive deletes, and post-generation checks could not undo commands that had already run. Smallest repair: add a plain gloss for oracle, and add one sentence on these two failure examples.","excerpt":"Prompted rules removed some violation types but not recursive deletes of scratch directories, and checks after generation repaired file edits but not commands that had already run."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-14 matches the arXiv v1 submission date. It does not reflect the later feed date.","excerpt":"[Submitted on 14 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim's excerpt is verbatim and supports it. The non-significant 17-point contrast and the null contrasts in Study 2 are preserved. The question and limitations follow from the stated low-cost models and one small fixture. No unsupported benefits are asserted.","excerpt":"the 17-point difference from prompted rules was not significant (p = 0.12), and neither was any contrast in Study 2, where only 3 of 14 tasks produced a violation"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the original source.","excerpt":null}],"createdAt":"2026-11-17T09:18:38.216Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}