The editable manuscript and Markdown source preserve demonstration status. Seven editable tables retain the central supplied numerical outputs. Clinical efficacy, causal mechanism and independent validation claims have been removed. No submission or human approval occurred. This handoff is separate from the manuscript.
This is a view of the recorded evidence and decisions. Recorded human responses are reproduced, not independently authenticated. No response or approval is created by this renderer.
Decisions
D1: Confirm data provenance and the intended evidence level
Status: pending
Issue: The supplied manuscript claims a 3,420,777-person clinical cohort; REQUEST.txt identifies it as an existing ClinicalSpark demonstration. Neither a clinical provenance package nor a simulation protocol was supplied.
Recommendation: Retain the manuscript wording that these are supplied demonstration outputs. If clinical research is intended, provide a data-source description, dates, eligibility/flow, unique-person linkage/deduplication logic, provenance and reproducible analysis exports before changing the claim level.
Consequence: A nominal sample size and database table names cannot establish an independently observed or validated cohort.
Alternatives:
Review and use solely as a demonstration.
Develop a separately validated clinical study from authorized data.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Title; Methods 2.1; REQUEST.txt: The supplied manuscript claims a 3,420,777-person clinical cohort; REQUEST.txt identifies it as an existing ClinicalSpark demonstration. Neither a clinical provenance package nor a simulation protocol was supplied.
D2: Resolve overlapping classifications and matched population
Status: pending
Issue: Counts 2,816,907 / 585,518 / 18,352 repeat across rehabilitation and remaining-teeth categories; the last two are also reused for anatomical outputs. The impaired total is 603,870, whereas matching uses 585,518 exposed participants.
Recommendation: Supply participant-level cross-tabulations or independent aggregate exports and an exposure-specific flow. Confirm whether the 18,352 unrestored group was excluded from matching and why. Retain separate illustrative classifications meanwhile.
Consequence: The current counts must not imply that rehabilitation, tooth count and tooth position classify the same people.
Alternatives:
Keep them as explicitly separate demonstration examples.
Replace all affected tables with verified exports and reconcile the narrative.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Methods 2.2; Results 3.3, 3.5 and 3.7: Counts 2,816,907 / 585,518 / 18,352 repeat across rehabilitation and remaining-teeth categories; the last two are also reused for anatomical outputs. The impaired total is 603,870, whereas matching uses 585,518 exposed participants.
D3: Recover model definitions and reconcile numerical inconsistencies
Status: pending
Issue: The rehabilitation and anatomy coefficients lack complete coding/reference levels. Pre-match SMDs do not generally reproduce from displayed moments: age is about 0.435 versus reported 0.384; male sex is about 0.048 versus 0.112. The crude mortality percentage ratio is about 5.378, not HR 4.846.
Recommendation: Recover code, model formulas, contrast definitions, unrounded summaries and original model output. Confirm Cox covariates, proportional-hazards assessment, matched dependence handling, missingness and event counts. Retain numerical outputs as source-reported examples without replacing model estimates using ad hoc recalculation.
Consequence: Matching balance and clinical effect magnitudes cannot be certified from the displayed summaries.
Alternatives:
Retain the demonstrations with the current interpretive bounds.
Publish only estimates reproduced from the original model exports.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Results Tables 2, 3, 4 and 6; input line 103: The rehabilitation and anatomy coefficients lack complete coding/reference levels. Pre-match SMDs do not generally reproduce from displayed moments: age is about 0.435 versus reported 0.384; male sex is about 0.048 versus 0.112. The crude mortality percentage ratio is about 5.378, not HR 4.846.
D4: Validate time and outcome definitions
Status: pending
Issue: Mortality is flagged by tenki_kbn_code = 3; ICD-10 I20–I25/I60–I69/I50 are presented as five-year incident events. Follow-up dates, time zero, censoring, prevalent-event exclusion and outcome validation are not supplied. Crude mortality and five-year survival have different implied observation windows.
Recommendation: Provide coding dictionaries, outcome validation evidence, entry and event dates, death linkage, censoring rules, at-risk tables and follow-up distributions. Align prosthetic exposure timing with time zero. Retain mortality-proxy and broad cardiovascular endpoint wording meanwhile.
Consequence: Claims of incident CVD, complete all-cause mortality ascertainment or treatment benefit would be premature.
Alternatives:
Continue as the current demonstration.
Recompute validated outcomes with explicit temporal definitions.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Methods 2.2; Results 3.2 and Table 8: Mortality is flagged by tenki_kbn_code = 3; ICD-10 I20–I25/I60–I69/I50 are presented as five-year incident events. Follow-up dates, time zero, censoring, prevalent-event exclusion and outcome validation are not supplied. Crude mortality and five-year survival have different implied observation windows.
D5: Define mediation and validate prediction before clinical interpretation
Status: pending
Issue: The mediation effect scale, DAG, estimators, uncertainty procedure and mediator ordering are unspecified; GBDT train/test split, hyperparameters, leakage control, calibration and importance definition are absent.
Recommendation: Supply the actual DAG, exposure/mediator/outcome timing, identification assumptions, effect scale, confounders and model exports. For prediction provide subject-disjoint validation, event prevalence, threshold and CI method, calibration and comparison with a conventional-factor model. Do not assign the direct component to measured inflammation or feature importance to causal risk.
Consequence: 31.4% is not an established mediated fraction of clinical events; AUC 0.842 does not establish validated clinical performance or treatment utility.
Alternatives:
Keep illustrative outputs and current bounded language.
Replace with reproducible, independently evaluated analyses.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Methods 2.4; Results 3.4 and 3.6; Tables 5 and 7: The mediation effect scale, DAG, estimators, uncertainty procedure and mediator ordering are unspecified; GBDT train/test split, hyperparameters, leakage control, calibration and importance definition are absent.
D6: Confirm responsible authors and declarations before any publication use
Status: pending
Issue: Only the collective author label is supplied; affiliations, responsible human authors, contributions, ethics, consent/waiver, data authorization, funding, conflicts, availability and required AI-use disclosure are unconfirmed. Two journal names appear in the source without a selected destination.
Recommendation: Retain the supplied group credit only for internal review. Obtain the actual required declarations and named responsible authors before publication use; do not insert template ethics approval, no-conflict, no-funding, or data-access claims. Select the destination and verify its current reporting requirements if journal submission is later intended.
Consequence: This is a prepared review candidate, not a submission-ready clinical manuscript.
Alternatives:
Use internally as a demonstration without making publication-readiness claims.
Complete a destination-specific publication package after factual confirmation.
Affected artifacts:
manuscript.md
manuscript.docx
Evidence:
S1, Title metadata; source overall: Only the collective author label is supplied; affiliations, responsible human authors, contributions, ethics, consent/waiver, data authorization, funding, conflicts, availability and required AI-use disclosure are unconfirmed. Two journal names appear in the source without a selected destination.
Sources
S1: ../input.md
description: Only supplied scientific manuscript and numerical source.
description: Three cited primary-source identities verified through Crossref/PubMed; claim support separately inspected in two abstracts and publisher full text. Suita corrigendum inspected.
Checks
Scientific framing — checked: Title, abstract, Methods, Results and conclusion consistently identify an existing demonstration. Prognostic association, causal mediation and treatment effect are distinguished.
Numerical preservation — checked: All central source numerical outputs retained across seven tables and text; source anatomical tables combined. The 13.2 survival difference is percentage points. The 57.2% and 2.34 quotient are arithmetic descriptions only.
Mechanistic and policy claims — repaired: Removed unmeasured trigeminal/SCFA/inflammatory causal pathway assertions, the mechanistic diagram, and claims that prosthetics are lifesaving cardiovascular prevention. No corresponding claims were relocated into a scientific supplement.
Reference correction — checked: Three retained references use actual named authors and verified identities. Tay is 2026;105(1):95-102. Aichi is 2025;30:95. Suita title and article number corrected; its 2022 correction concerned a funding grant. Other initial references were not required by the revised argument and were removed, not represented as comprehensively verified.
Automated screens — checked: Final publication gate: two files, zero findings. Final numeric screen: 279 source numbers, zero unlocated statistics or warnings. These do not validate source accuracy, clinical results or complete model reproduction.
Independent bounded review — checked: A subagent checked only input.md and manuscript.md; it found an omitted anterior-only restriction, now repaired. No other numerical transcription or uncaveated clinical-claim findings.
Layout — checked: All 14 PNG pages in render-checked were visually inspected. All seven native editable tables and their notes are intact; heading, front matter, line numbering and footer pagination were checked. The matching DOCX SHA-256 is e2d86e1828fda207df0274015b7569ffcacb24bfefdfa9b2be0b583d107c3246. See verification/layout.json.
History
2026-10-05: Prepared a new review candidate from the unchanged input and assigned frozen skill. Human decisions remain pending; no approval or submission inferred.