{ "version": "historical-v1", "status": "Superseded by PDF-key-v2 reanalysis; preserved as historical output, not the intended scoring key.", "artifacts": { "concussion-analysis.json": { "schema_version": 1, "source": { "file": "4-sutrix-cleaned-output.csv", "sha256": "501042c4be0c3c7f40eab1cc1907d0e23607b609212af8fbcd846209c88c7e37", "runner": "microservices/orchestrator/src/evals/runners/gtsurvey-consistency.ts", "verified_on": "2026-09-10", "scope": "Exploratory reanalysis of the saved cleaned study dataset. Not a reproduction of the original biostatistician model or a validation of all source cleaning." }, "dataset_rows": 348, "regression": { "coefficients": { "Knowledge_Score_Total": { "beta": 0.025437, "ci": [ 0.87765, 1.198873 ], "or": 1.025764, "p": 0.749192 }, "Q15": { "beta": 0.927843, "ci": [ 1.278206, 5.003956 ], "or": 2.529048, "p": 0.007699 }, "Q43": { "beta": 0.45108, "ci": [ 1.137346, 2.167259 ], "or": 1.570007, "p": 0.006099 }, "Q44": { "beta": 0.244852, "ci": [ 0.676278, 2.412961 ], "or": 1.277432, "p": 0.450514 }, "intercept": { "beta": -3.425978, "ci": [ 0.002558, 0.413315 ], "or": 0.032517, "p": 0.008264 } }, "converged": true, "excluded_rows": 72, "n": 276, "outcome": "Q21" }, "verify_warnings": [] }, "concussion-exploration.json": { "search_space": { "examined": 447, "families": [ { "name": "outcome_predictors", "size": 446 }, { "name": "instrument_pairs", "size": 1 } ] }, "correction": { "method": "fdr_bh", "alpha": 0.05 }, "fdr_surviving": 4, "screened_candidates": 4, "permutations": 1000, "bootstraps": 1000, "seed": 42, "bootstrap_retention": 0.5, "candidate": { "outcome": "Q46", "predictor": "Q48", "kind": "contingency", "test": "chi_square", "statistic": 25.775549, "effect_size": { "name": "cramers_v", "value": 0.307272, "ci": null }, "p_value": 0.00024512736791815204, "family": "outcome_predictors", "n": 273, "id": "Q46__Q48__contingency", "relationship": "Q46 \u2194 Q48", "q_value": 0.041741236247923585, "survives_fdr": true, "robustness": { "stable": false, "note": "No re-fit available to assess stability." }, "rationale": "Exploratory association (chi_square): cramers_v = 0.307272 (n=273). Hypothesis-generating; confirm before drawing conclusions.", "validation": { "effectSize": 0.307272, "permutationP": 0.001998001998001998, "bootstrapStability": 0.999, "n": 273, "surfaced": true, "reason": "Same-data resampling screen: permutation p=0.0020; 999/1000 bootstrap samples retained at least half the observed association magnitude. Not independent replication." } }, "scope": "Exploratory / hypothesis-generating. These are associations, not causal claims. Multiple-comparison correction (FDR) controls data-driven false positives but does not address confounding, data quality, or causation. Confirm in an independent sample or pre-registered follow-up before drawing conclusions." }, "concussion-methods.json": { "schema_version": 1, "title": "Concussion survey: saved-data reanalysis record", "verified_on": "2026-09-10", "input": { "sha256": "501042c4be0c3c7f40eab1cc1907d0e23607b609212af8fbcd846209c88c7e37", "rows": 348, "availability": "Restricted cleaned input; participant records are not published. The fingerprint identifies the file used, not raw-to-clean accuracy." }, "model": { "method": "Complete-case logistic regression with intercept; odds ratios and 95% Wald confidence intervals", "outcome": "Q21: self-reported lifetime care-seeking for a suspected concussion", "predictors": [ "Knowledge_Score_Total", "Q43", "Q44", "Q15" ], "complete_cases": 276, "excluded_rows": 72, "exclusion_rule": "Missing at least one model value", "coefficient_p_values": "Not multiplicity-adjusted", "specification_status": "Exploratory reanalysis; no preregistration claim" }, "exploration": { "comparisons": 447, "correction": "Benjamini-Hochberg FDR at 0.05 within each family", "families": { "outcome_predictors": 446, "instrument_pairs": 1 }, "fdr_survivors": 4, "resampling_screen_candidates": 4, "published_candidate": "Q46 hypothetical care timing by Q48 hypothetical online information-seeking", "candidate_n": 273, "permutations": 1000, "bootstraps": 1000, "seed": 42, "retention_rule": "999 of 1000 bootstrap samples retained at least half the observed association magnitude", "refit_stability": "Not assessed: no re-fit available. The saved robustness.stable=false flag is distinct from bootstrap retention.", "published_scope": "One aggregate candidate and search counts, not a complete comparison ledger" }, "reproduction": { "command": "bun scripts/verify-landing-study.ts", "engine_revision": "8fe322611419d641c5fd8eee600c6cc71b140ce8", "engine_worktree_changes": false, "runtime": { "bun": "1.3.10", "python": "3.14.3" }, "scope": "Local rerun matched the input digest, saved regression, exploration and resampling values. The public exploration reason is an editorial clarification of the runner's rounded resampling summary, checked against the exact values. Requires the private repository and authorized input access; this is not a self-contained public reproduction package." }, "limitations": [ "Associations do not establish causation or temporal direction.", "Hypothetical care timing is not observed care delay.", "Resampling is not independent replication or a probability that a finding is true.", "The historical scoring code has been recovered and its output linked to the saved model columns. Its key disagrees with the supplied PDF on Q31 and Q37. Source-key reconciliation and reviewed reanalysis remain pending. The separate calculation uses the historical saved score, not a newly reconstructed score.", "No validation of every raw-to-clean decision or reproduction of the original biostatistician model.", "No claim of independent scientific validation, customer delivery, publication outcome or time savings." ] }, "concussion-model-check.json": { "scope": "Same-cleaned-input implementation cross-check; not independent sample replication, clinical validation, or raw-cleaning proof.", "method": "Unweighted, unpenalized Bernoulli logistic maximum likelihood; intercept; predictors treated as supplied numeric values; Newton steps with backtracking; model-based inverse-information standard errors; two-sided normal Wald tests and 95% confidence intervals.", "outcome": "Q21=1 versus Q21=0", "predictors": [ "Knowledge_Score_Total", "Q43", "Q44", "Q15" ], "csv_sha256": "501042c4be0c3c7f40eab1cc1907d0e23607b609212af8fbcd846209c88c7e37", "python": "3.14.3", "counts": { "total": 348, "complete_cases": 276, "excluded": 72, "missing_required_field": 72, "invalid_numeric_or_outcome": 0 }, "coefficients": [ { "term": "Intercept", "coefficient": -3.4259783556134376, "standard_error": 1.2971841536987292, "odds_ratio": 0.03251745179001104, "coefficient_wald95ci": [ -5.9684125781790165, -0.8835441330478586 ], "odds_ratio_wald95ci": [ 0.0025582992974878785, 0.4133154678008123 ], "wald_z": -2.6410886579555926, "p_two_sided": 0.008264009064535756 }, { "term": "Knowledge_Score_Total", "coefficient": 0.025437342335142693, "standard_error": 0.07956519835307126, "odds_ratio": 1.0257636323029415, "coefficient_wald95ci": [ -0.1305075808596626, 0.181382265529948 ], "odds_ratio_wald95ci": [ 0.8776498395832527, 1.1988733796783368 ], "wald_z": 0.31970437907116456, "p_two_sided": 0.7491924398625496 }, { "term": "Q43", "coefficient": 0.45108035555784465, "standard_error": 0.16448402129097875, "odds_ratio": 1.5700074358478846, "coefficient_wald95ci": [ 0.12869759779520684, 0.7734631133204825 ], "odds_ratio_wald95ci": [ 1.1373461361777968, 2.167258735235478 ], "wald_z": 2.7423962037009395, "p_two_sided": 0.006099272305859266 }, { "term": "Q44", "coefficient": 0.24485187447572357, "standard_error": 0.32449706378653753, "odds_ratio": 1.277432078893603, "coefficient_wald95ci": [ -0.3911506836348866, 0.8808544325863337 ], "odds_ratio_wald95ci": [ 0.6762782442972207, 2.412960537984201 ], "wald_z": 0.7545580586109506, "p_two_sided": 0.4505141970617434 }, { "term": "Q15", "coefficient": 0.9278429897706079, "standard_error": 0.3481624416877388, "odds_ratio": 2.529048107304899, "coefficient_wald95ci": [ 0.24545714329311308, 1.6102288362481025 ], "odds_ratio_wald95ci": [ 1.2782055027383465, 5.0039561833836 ], "wald_z": 2.6649715152295923, "p_two_sided": 0.007699484417322271 } ], "diagnostics": { "converged": true, "iterations": 6, "log_likelihood": -145.95479683384025, "gradient_max_abs": 1.2587153541687712e-14, "history": [ { "iteration": 1, "log_likelihood": -147.68803761202096, "step_max_abs": 2.460845326006448, "step_scale": 1 }, { "iteration": 2, "log_likelihood": -145.9880191840642, "step_max_abs": 0.8249331787100699, "step_scale": 1 }, { "iteration": 3, "log_likelihood": -145.95481691765576, "step_max_abs": 0.1368939001564218, "step_scale": 1 }, { "iteration": 4, "log_likelihood": -145.95479683384852, "step_max_abs": 0.003303961731458614, "step_scale": 1 }, { "iteration": 5, "log_likelihood": -145.95479683384025, "step_max_abs": 1.9890082355522776e-06, "step_scale": 1 }, { "iteration": 6, "log_likelihood": -145.95479683384025, "step_max_abs": 8.033657762025613e-13, "step_scale": 1 } ], "alternate_start_max_abs_coefficient_difference": 1.687538997430238e-14, "finite_difference_gradient_max_abs_error": 1.074340610029978e-06, "finite_difference_information_max_abs_error": 8.040698958211578e-06 }, "limitations": [ "Uses the existing cleaned input; does not verify raw recoding, consent, sampling, construct validity, or cleaning provenance.", "No survey weighting, clustering, robust variance, interactions, nonlinearity assessment, out-of-sample evaluation, or causal interpretation.", "Agreement would establish this model calculation only; it would not validate the full engine or every reported analysis." ], "verification": { "performed_on": "2026-09-10", "authorship": "Separate AI-agent implementation, computed before opening the public model artifact; no Sutrix model-engine code read or reused.", "comparison": "All 25 published beta, odds ratio, confidence-limit and p-value entries agree at six-decimal rounding.", "maximum_absolute_difference": 4.972616534093532e-07, "synthetic_check": "Two-group binomial coefficients and covariance agree with closed-form values within 1e-10.", "reproduce": "python3 concussion-model-check.py --csv /path/to/authorized-cleaned-input.csv --output result.json", "scope": "Same-cleaned-input calculation check, not independent scientific review or independent sample replication." } }, "gt-study-definitions.json": { "title": "College concussion survey: selected variable definitions", "scope": "Questionnaire and preparation excerpt. No participant records or statistical results are included.", "source": "TBI Data Dictionary (filled).xlsx, Sheet1", "source_sha256": "967fc14e2697f369062bd94ff2743ec737932069ac35d8c225d77362839acee7", "variables": [ { "id": "Q15", "source_row": 37, "question": "Did you play competitive sports in high school?", "responses": [ "Yes", "No" ], "recoded_variable": "Q15_R", "recoded_source_row": 189, "source_recoding": { "Yes": 1, "No": 0 } }, { "id": "Q44", "source_row": 78, "question": "Have you ever received any concussion education or training sessions prior to entering college?", "responses": [ "Yes", "No" ], "recoded_variable": "Q44_R", "recoded_source_row": 249, "source_recoding": { "Yes": 1, "No": 0 } }, { "id": "Q48", "source_row": 83, "question": "Imagine you slip and hit your head and afterwards don't feel right and suspect you may have a concussion.\n\nIf you suspected you had a concussion, would you search online for information before seeking professional help?", "responses": [ "Yes", "No" ], "recoded_variable": "Q48_R", "recoded_source_row": 254, "source_recoding": { "Yes": 1, "No": 0 } }, { "id": "Q21", "source_row": 52, "question": "In your lifetime, have you ever sought medical attention for a suspected concussion?", "measures": "Self-reported lifetime care-seeking", "responses": [ "Yes", "No" ] }, { "id": "Q43", "source_row": 77, "question": "How confident are you in your answers about your concussion knowledge? (where 1 is not confident at all and 5 is very confident)", "measures": "Confidence in answers, not an objective knowledge score", "responses": [ 1, 2, 3, 4, 5 ] }, { "id": "Q46", "source_row": 80, "question": "Imagine you slip and hit your head and afterwards don't feel right and suspect you may have a concussion. How soon after hitting your head would you seek healthcare?", "measures": "Intended timing in a hypothetical scenario", "responses": [ "Immediately", "Within a few hours", "The same day", "The next day", "Within the week", "I would not seek healthcare", "Only if problems persisted after a week" ] } ], "scoring_provenance": { "saved_predictor": "Knowledge_Score_Total is described in the saved Sutrix analysis as a 0\u201319 score over Q23\u2013Q41. This is distinct from the original manual Knowledge_Score_Total_R.", "item_ids": [ "Q23", "Q24", "Q25", "Q26", "Q27", "Q28", "Q29", "Q30", "Q31", "Q32", "Q33", "Q34", "Q35", "Q36", "Q37", "Q38", "Q39", "Q40", "Q41" ], "dictionary_rows": "57\u201375", "historical_code_sha256": "b6d4387cec595a19a750a540cb89aaf1c451cafbe8343d2f84730d33451c4cc4", "historical_code_key": { "Q23": "True", "Q24": "False", "Q25": "False", "Q26": "False", "Q27": "True", "Q28": "False", "Q29": "True", "Q30": "False", "Q31": "False", "Q32": "False", "Q33": "True", "Q34": "True", "Q35": "True", "Q36": "True", "Q37": "True", "Q38": "False", "Q39": "True", "Q40": "False", "Q41": "True" }, "aggregation": "Historical operations 189\u2013207 score recognized keyed answers 1 and opposites 0; operation 208 casts integers; operation 209 sums Q23\u2013Q41. Any blank/unparsable item makes the total blank; no prorating. Expected range 0\u201319 assumes valid binary items and was not enforced by the historical code.", "linkage": "The recovered production cleaned artifact and saved repository input match row-for-row on all model/exploration columns and all 19 knowledge items across 348 rows. This does not validate every raw-to-clean operation.", "unresolved": "The recovered historical code differs from the supplied answer-key PDF on Q31 and Q37. The PDF keys both oppositely. The historical key is documented here, not asserted to be the authoritative instrument key. Reconciliation and a reviewed reanalysis remain pending.", "scope": "Source definitions and saved-predictor metadata do not establish raw-to-clean fidelity." }, "coding_example": { "id": "Q46_R", "source_row": 251, "recorded_mapping": { "Immediately": 1, "Within a few hours": 2, "The same day": 3, "The next day": 4, "Within the week": 5, "Only if problems persisted after a week": 6, "I would not seek healthcare": 0 }, "review_note": "These are category codes, not elapsed time. Never seeking care must not be interpreted as earlier than immediate care. Any grouping or model treatment needs an explicit analysis decision." } } } }