{"claim": "Safe-mode entry is a routine, recurring feature of long-duration spacecraft operations, and spacecraft failure behavior is statistically tractable at the population level, varying systematically by subsystem, mass class, and mission type; this establishes that a population of fault and failure episodes large enough to model exists and that the post-fault-survival question is a real, recurring decision problem rather than an exotic one.", "evidence": [{"source": "Castet & Saleh, Satellite and satellite subsystems reliability: Statistical data analysis and modeling, Reliability Engineering and System Safety (2009)", "doi_or_url": "https://doi.org/10.1016/j.ress.2009.05.004", "grade": "A"}, {"source": "Castet & Saleh, Statistical reliability analysis of satellites by mass category: Does spacecraft size matter?, Acta Astronautica (2011)", "doi_or_url": "https://doi.org/10.1016/j.actaastro.2010.04.017", "grade": "A"}], "facet": "empirics", "chapter": "ch1_introduction", "subclaim": "real"}
{"claim": "The autonomy spectrum in fault management is real, varied, and flight-proven: the Deep Space One Remote Agent Experiment closed the full detection-isolation-recovery chain onboard, and Cassini-class system fault protection documents the ground-dependent-to-autonomous range, so fault-management autonomy is a genuinely orderable treatment variable rather than a binary or a hypothetical.", "evidence": [{"source": "Bernard et al., Spacecraft autonomy flight experience: The DS1 Remote Agent Experiment (1999)", "doi_or_url": "https://doi.org/10.2514/6.1999-4512", "grade": "A"}, {"source": "Rasmussen, System fault protection design for the Cassini spacecraft, IEEE Aerospace Applications Conference (1996)", "doi_or_url": "https://doi.org/10.1109/aero.1996.495890", "grade": "A"}], "facet": "mechanism", "chapter": "ch1_introduction", "subclaim": "real"}
{"claim": "The cliometric premise that grounds the study is established: a counterfactual claim of the form 'outcome could not have occurred without factor' remains unmeasured until stated quantitatively against an explicitly constructed next-best alternative and exposed to falsification (the social-saving standard), which makes a conditional population-level autonomy-survival estimate a materially improving apparatus over an untested engineering intuition.", "evidence": [{"source": "Fogel, Railroads and American Economic Growth: Essays in Econometric History (1964)", "doi_or_url": "https://doi.org/10.2307/2552284", "grade": "B"}, {"source": "Leunig, Social Savings, Journal of Economic Surveys (2010)", "doi_or_url": "https://doi.org/10.1111/j.1467-6419.2010.00636.x", "grade": "A"}], "facet": "economics", "chapter": "ch2_theoretical_framework", "subclaim": "material"}
{"claim": "The Cox proportional-hazards model with partial likelihood is the appropriate and validated estimator for a time-to-event question with right-censoring, and its counting-process large-sample behavior is governed by the terminal-event count, so the design's effective sample size and power are set by the number of mission-ending losses rather than the number of episodes; this both licenses the estimator choice and bounds what a thin-event population can support.", "evidence": [{"source": "Cox, Regression Models and Life-Tables, Journal of the Royal Statistical Society Series B (1972)", "doi_or_url": "https://doi.org/10.1111/j.2517-6161.1972.tb00899.x", "grade": "A"}, {"source": "Andersen & Gill, Cox's Regression Model for Counting Processes: A Large Sample Study, Annals of Statistics (1982)", "doi_or_url": "https://doi.org/10.1214/aos/1176345976", "grade": "A"}], "facet": "empirics", "chapter": "ch5_research_design", "subclaim": "material"}
{"claim": "The social-saving Fogel demands is a COST counterfactual (the cost of performing the same task by the next-best alternative with output held fixed, related to consumer surplus and TFP), not a survival probability; a survival hazard ratio is therefore not a substitute for the missing cost leg, so the dissertation's repeated equation of the autonomy hazard ratio with a social saving is a rhetorical analogy unless an episode-level recovery-cost counterfactual is constructed.", "evidence": [{"source": "Fogel, Railroads and American Economic Growth: Essays in Econometric History (1964)", "doi_or_url": "https://doi.org/10.2307/2552284", "grade": "B"}, {"source": "Leunig, Social Savings, Journal of Economic Surveys (2010)", "doi_or_url": "https://doi.org/10.1111/j.1467-6419.2010.00636.x", "grade": "A"}], "facet": "economics", "chapter": "ch2_theoretical_framework", "subclaim": "residual_risk"}
{"claim": "In the railway social-savings literature, time SAVED (not outcome avoided) drives most of the measured passenger benefit, so a design that models only the rare loss margin while censoring away the recovery-time-among-survivors margin risks measuring the smaller component and may have chosen the wrong estimand margin.", "evidence": [{"source": "Leunig, Time is Money: A Re-Assessment of the Passenger Social Savings from Victorian British Railways, Journal of Economic History (2006)", "doi_or_url": "https://doi.org/10.1017/S0022050706000283", "grade": "A"}, {"source": "Leunig, Social Savings, Journal of Economic Surveys (2010)", "doi_or_url": "https://doi.org/10.1111/j.1467-6419.2010.00636.x", "grade": "A"}], "facet": "empirics", "chapter": "ch6_analysis_plan", "subclaim": "alternatives"}
{"claim": "When terminal events differ in cause or downstream consequence (a contained single-vehicle loss versus a debris-generating or shared-asset loss), valid estimation requires cause-specific competing-risks hazards rather than a single pooled event; pooling heterogeneous terminal events into one hazard mis-aggregates the effect, and a fragmentation-generating loss imposes a propagating non-localized cost distinct from the loss of the single vehicle.", "evidence": [{"source": "Lau, Cole & Gange, Competing Risk Regression Models for Epidemiologic Data, American Journal of Epidemiology (2009)", "doi_or_url": "https://doi.org/10.1093/aje/kwp107", "grade": "A"}, {"source": "Lewis, Understanding long-term orbital debris population dynamics, Journal of Space Safety Engineering (2020)", "doi_or_url": "https://doi.org/10.1016/j.jsse.2020.06.006", "grade": "A"}], "facet": "rival", "chapter": "ch7_discussion", "subclaim": "residual_risk"}
{"claim": "In heavy-tailed processes the historical record systematically undersamples the tail and sample means understate true exposure, so a pooled mean-based statistic such as a single hazard ratio is unreliable where extremes dominate, and the non-naive precautionary principle confines precaution to the joint case of fat tails AND systemic non-localizable ruin; treating a thin-event tail subgroup as decisive violates this.", "evidence": [{"source": "Taleb, Read, Douady, Norman & Bar-Yam, The Precautionary Principle (arXiv:1410.5787, 2014)", "doi_or_url": "https://doi.org/10.48550/arxiv.1410.5787", "grade": "A"}, {"source": "Cirillo & Taleb, Tail risk of contagious diseases, Nature Physics (2020)", "doi_or_url": "https://doi.org/10.1038/s41567-020-0921-x", "grade": "A"}], "facet": "empirics", "chapter": "ch6_analysis_plan", "subclaim": "residual_risk"}
{"claim": "Antifragility is a SECOND-order (convex) response to the DOSE of a stressor, formally distinct from robustness and resilience, so a single monotone Cox coefficient is a first-order mean-shift quantity that by construction cannot locate the system on the fragile-robust-antifragile continuum; an autonomy-by-dose interaction or spline is required to test whether the protective effect accelerates or decays at the tail.", "evidence": [{"source": "Taleb et al., The Precautionary Principle (arXiv:1410.5787, 2014), formal fragility/convexity framework", "doi_or_url": "https://doi.org/10.48550/arxiv.1410.5787", "grade": "A"}, {"source": "Taleb, Antifragile: Things That Gain from Disorder (review record), Quantitative Finance (2013)", "doi_or_url": "https://doi.org/10.1080/14697688.2013.829244", "grade": "B"}], "facet": "mechanism", "chapter": "ch7_discussion", "subclaim": "alternatives"}
{"claim": "Skin in the game is both an ethical and a statistical filter that discounts favorable assurances authored by parties carrying no exposure to the downside; because the autonomy treatment is scored largely from the implementing programs' own documentation while only the loss outcome is routed through independent GAO oversight, the favorable upper-bound hazard ratio is exactly the self-authored cheap assurance the filter is built to discard.", "evidence": [{"source": "Martinetti et al., Safety-I and Safety-II, resilience and antifragility engineering: a debate (formalizing whether added automation builds or erodes antifragility)", "doi_or_url": "https://doi.org/10.1080/10803548.2018.1444724", "grade": "A"}], "facet": "governance", "chapter": "ch7_discussion", "subclaim": "residual_risk"}
{"claim": "Potential outcomes are well-defined only under a well-posed manipulable intervention (no causation without manipulation) and under SUTVA (no interference and no hidden versions of treatment); because the autonomy score is a per-mission pre-flight design property, the only honest manipulation is program-level and conditioning on fault entry conditions on a post-treatment variable, so the episode-level treatment is not well-posed as specified.", "evidence": [{"source": "Holland, Statistics and Causal Inference, Journal of the American Statistical Association (1986)", "doi_or_url": "https://doi.org/10.1080/01621459.1986.10478354", "grade": "A"}, {"source": "Imbens & Rubin, Causal Inference for Statistics, Social, and Biomedical Sciences (2015)", "doi_or_url": "https://doi.org/10.1017/cbo9781139025751", "grade": "A"}], "facet": "identification", "chapter": "ch5_research_design", "subclaim": "residual_risk"}
{"claim": "Propensity-score adjustment removes bias due to observed covariates only where overlap (common support) holds; where the highest-autonomy episodes occupy a complexity-distance region with no lower-autonomy donors, the within-stratum Cox contrast is extrapolation past support rather than adjustment, which the candidate's own concentration of coverage on flagship deep-space missions makes a live threat.", "evidence": [{"source": "Rosenbaum & Rubin, The central role of the propensity score in observational studies for causal effects, Biometrika (1983)", "doi_or_url": "https://doi.org/10.1093/biomet/70.1.41", "grade": "A"}], "facet": "empirics", "chapter": "ch5_research_design", "subclaim": "residual_risk"}
{"claim": "Objective causal inference requires the design, including the determination of outcomes and the analysis plan, to be completed while blind to the outcome data; the dissertation seals the treatment (pre-flight autonomy scoring with second-reader weighted-kappa reliability) but leaves the outcome and censoring side outside the blinding-and-reliability apparatus, so the design's objectivity is only half-met and differential outcome measurement can contaminate the hazard ratio.", "evidence": [{"source": "Rubin, For objective causal inference, design trumps analysis, Annals of Applied Statistics (2008)", "doi_or_url": "https://doi.org/10.1214/08-aoas187", "grade": "A"}, {"source": "Imbens & Rubin, Causal Inference for Statistics, Social, and Biomedical Sciences (2015)", "doi_or_url": "https://doi.org/10.1017/cbo9781139025751", "grade": "A"}], "facet": "empirics", "chapter": "ch4_data_and_measurement", "subclaim": "residual_risk"}
{"claim": "Onboard fault detection-isolation-recovery is implemented by materially different architectures (two-layer FDI plus layered recovery, Kalman-filter and neural-network fusion, observer-based detection), so a three-bin ordinal autonomy score collapses real within-level heterogeneity and the per-level hazard ratio averages over hidden treatment versions rather than estimating the single contrast H1 names.", "evidence": [{"source": "Chen, Bettens, Xie, Wang & Wu, Kalman filter and neural network fusion for fault detection and recovery in satellite attitude estimation, Acta Astronautica (2024)", "doi_or_url": "https://doi.org/10.1016/j.actaastro.2024.01.038", "grade": "A"}], "facet": "measurement", "chapter": "ch4_data_and_measurement", "subclaim": "residual_risk"}
{"claim": "Conditioning on a collider (a variable that is a common effect of two causes, here recorded-fault-entry as a child of both autonomy and unobserved severity) induces an endogenous, non-causal association between its causes inside the analysis sample, and this selection bias can substantially distort observed associations even when each underlying effect is modest; no observed adjustment set closes a collider path that transits an unobserved variable.", "evidence": [{"source": "Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009)", "doi_or_url": "https://doi.org/10.1017/cbo9780511803161", "grade": "A"}, {"source": "Elwert & Winship, Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable, Annual Review of Sociology (2014)", "doi_or_url": "https://doi.org/10.1146/annurev-soc-071913-043455", "grade": "A"}, {"source": "Munafo et al., Collider scope: when selection bias can substantially influence observed associations, International Journal of Epidemiology (2017)", "doi_or_url": "https://doi.org/10.1093/ije/dyx206", "grade": "A"}], "facet": "identification", "chapter": "ch5_research_design", "subclaim": "residual_risk"}
{"claim": "Adjusting for an intermediate variable on the causal path (or a proxy of it) is overadjustment that biases the effect estimate, distinct from unnecessary adjustment which only reduces precision; because distance and complexity are partly the regimes autonomy is designed to exploit, adjusting their realized-exploitation component differences out the protective mechanism and biases the hazard ratio toward one.", "evidence": [{"source": "Schisterman, Cole & Platt, On the Relative Nature of Overadjustment and Unnecessary Adjustment, Epidemiology (2009)", "doi_or_url": "https://doi.org/10.1097/ede.0b013e3181a82f12", "grade": "A"}, {"source": "Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009)", "doi_or_url": "https://doi.org/10.1017/cbo9780511803161", "grade": "A"}], "facet": "identification", "chapter": "ch5_research_design", "subclaim": "residual_risk"}
{"claim": "The proportional-hazards assumption for a specified coefficient is tested by regressing its scaled Schoenfeld residuals on a function of time (Grambsch-Therneau weighted-residual test); a non-zero slope indicates a time-varying effect for that variable, and when detected the remedy is a time-varying-coefficient or time-stratified Cox specification rather than a single constant hazard ratio, so the single treatment coefficient must be tested before it is reported.", "evidence": [{"source": "Grambsch & Therneau, Proportional hazards tests and diagnostics based on weighted residuals, Biometrika 81(3) (1994)", "doi_or_url": "https://doi.org/10.1093/biomet/81.3.515", "grade": "A"}, {"source": "Zhang et al., Time-varying covariates and coefficients in Cox regression models, Annals of Translational Medicine (2018)", "doi_or_url": "https://doi.org/10.21037/atm.2018.02.12", "grade": "B"}], "facet": "empirics", "chapter": "ch6_analysis_plan", "subclaim": "mechanism"}
{"claim": "A causal effect is identified under the back-door criterion only if the conditioning set blocks every back-door (common-cause) path; a risk-classification/requirements-stringency regime that drives both autonomy investment and, through a margin/redundancy/abort-disposal-policy channel, mission-ending loss leaves an open back-door path inside any fixed-autonomy stratum that {complexity, distance, age} does not block, so the autonomy hazard ratio is non-identified on regime-confounding grounds unless a proxy for that regime is measured.", "evidence": [{"source": "Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009)", "doi_or_url": "https://doi.org/10.1017/cbo9780511803161", "grade": "A"}], "facet": "identification", "chapter": "ch5_research_design", "subclaim": "residual_risk"}
{"claim": "Formal identification rescues for an endogenous self-selected treatment are exactly back-door adjustment, a valid instrument with an exclusion restriction, and the front-door criterion; the front-door rescue requires the onboard detection-isolation-recovery mediator to be unconfounded with loss except through autonomy, which fails when a risk-class regime sets the margin that determines whether a recovery action succeeds, so neither rescue is instantiable and a sensitivity bound rather than a point hazard ratio is the correct deliverable.", "evidence": [{"source": "Pearl, A general identification condition for causal effects (2002)", "doi_or_url": "https://doi.org/10.5555/777092.777180", "grade": "A"}, {"source": "Martinussen & Vansteelandt, Instrumental Variable Estimation of the Causal Hazard Ratio, Biometrics (2022)", "doi_or_url": "https://doi.org/10.1111/biom.13792", "grade": "A"}], "facet": "identification", "chapter": "ch6_analysis_plan", "subclaim": "residual_risk"}
{"claim": "An accurate census is the precondition for downstream collision-avoidance and governance claims, and where the catalog floor is reached capacity and risk claims become unverifiable; transferred to a fault catalog, an unreproducible event definition cannot ground a conditional survival estimate, so an inter-coder reliability statistic on the fault-entry event and end-state is owed before any hazard ratio.", "evidence": [{"source": "Murtaza et al., LEO satellite security and reliability survey, IEEE Access (2020)", "doi_or_url": "https://doi.org/10.1109/access.2020.2979505", "grade": "A"}], "facet": "measurement", "chapter": "ch4_data_and_measurement", "subclaim": "residual_risk"}
{"claim": "Ceding continuous control imposes a takeover performance cost (an operator suddenly required to intervene is slow and error-prone) and intermediate levels of automation have non-uniform effects on situation awareness and workload, so the out-of-the-loop harm is concentrated at intermediate delegation; a monotone-in-level Cox coefficient is structurally incapable of registering a mid-scale hazard bump and an inverted-U test (ordered dummies or quadratic) is required.", "evidence": [{"source": "Parasuraman, Sheridan & Wickens, A model for types and levels of human interaction with automation, IEEE Transactions on Systems, Man, and Cybernetics A (2000)", "doi_or_url": "https://doi.org/10.1109/3468.844354", "grade": "A"}, {"source": "Kaber & Endsley, The effects of level of automation and adaptive automation on human performance, situation awareness and workload in a dynamic control task, Theoretical Issues in Ergonomics Science (2004)", "doi_or_url": "https://doi.org/10.1080/1463922021000054335", "grade": "B"}], "facet": "measurement", "chapter": "ch4_data_and_measurement", "subclaim": "alternatives"}
{"claim": "Automation applies to four distinct stages of human information processing whose automation carries different human-performance signatures (auto-isolate-but-wait-for-ground keeps a human in the decision loop; auto-recover removes them), so a single collapsed detection-isolation-recovery ordinal should be coded as three separate stage indicators to let the model attribute any survival benefit to automated diagnosis, automated recovery action, or only their bundle.", "evidence": [{"source": "Parasuraman, Sheridan & Wickens, A model for types and levels of human interaction with automation, IEEE Transactions on Systems, Man, and Cybernetics A 30(3):286-297 (2000)", "doi_or_url": "https://doi.org/10.1109/3468.844354", "grade": "A"}, {"source": "Parasuraman & Riley, Humans and automation: use, misuse, disuse, abuse, Human Factors (1997)", "doi_or_url": "https://doi.org/10.1518/001872097778543886", "grade": "A"}], "facet": "mechanism", "chapter": "ch4_data_and_measurement", "subclaim": "mechanism"}
{"claim": "Automation-induced error splits into omission (missed because the automation did not flag) and commission (executing an automated action against contrary evidence), both robust and dosage-dependent; a fast autonomous reconfiguration into a worse state registers in two-state recovery/loss coding as an apparently protective short time-to-recovery or a censored survivor, biasing the hazard ratio toward a spurious autonomy benefit, so a third event category for onboard-recovery-action-implicated losses is required.", "evidence": [{"source": "Parasuraman & Manzey, Complacency and bias in human use of automation: an attentional integration, Human Factors 52(3):381-410 (2010)", "doi_or_url": "https://doi.org/10.1177/0018720810376055", "grade": "A"}, {"source": "Parasuraman & Riley, Humans and automation: use, misuse, disuse, abuse, Human Factors 39(2):230-253 (1997)", "doi_or_url": "https://doi.org/10.1518/001872097778543886", "grade": "A"}], "facet": "measurement", "chapter": "ch4_data_and_measurement", "subclaim": "residual_risk"}
{"claim": "High automation reliability predictably degrades operator monitoring of the automated function (complacency), and reliance miscalibrates two ways (misuse=over-reliance, disuse=under-reliance after nuisance trips), so the low-autonomy comparison arm (a human ground loop) is itself a non-constant-reliability automated-monitoring-plus-operator system; the estimated autonomy benefit may be the mirror image of a degrading human backstop rather than capability in the high-autonomy arm.", "evidence": [{"source": "Parasuraman, Molloy & Singh, Performance consequences of automation-induced complacency, International Journal of Aviation Psychology (1993)", "doi_or_url": "https://doi.org/10.1207/s15327108ijap0301_1", "grade": "A"}, {"source": "Parasuraman & Riley, Humans and automation: use, misuse, disuse, abuse, Human Factors (1997)", "doi_or_url": "https://doi.org/10.1518/001872097778543886", "grade": "A"}], "facet": "rival", "chapter": "ch7_discussion", "subclaim": "alternatives"}
{"claim": "Operator workload, vigilance, fatigue, and shift-handover state are dynamic time-varying variables, so a safe-mode entry during a concurrent critical operation or overloaded console is resolved by a different effective backstop than one during quiet cruise; operations tempo at fault entry is a distinct candidate confounder of the ground-loop-dependent outcome beyond the static complexity/distance/age controls.", "evidence": [{"source": "Parasuraman, Neuroergonomics: Research and practice, Theoretical Issues in Ergonomics Science (2003)", "doi_or_url": "https://doi.org/10.1080/14639220210199753", "grade": "A"}], "facet": "identification", "chapter": "ch7_discussion", "subclaim": "residual_risk"}
{"claim": "A stock changes only through its flows and behavior over time is the integral of inflow minus outflow, so two regimes whose flow balances differ (a high- and a low-autonomy recovery, where an intervention at minute two reshapes the state at minute thirty) are different dynamic systems whose baseline accumulation paths need not be proportional; the proportional-hazards single-multiplier summary must therefore be tested as a structural claim against autonomy-stratified baseline survival curves.", "evidence": [{"source": "Forrester, Industrial Dynamics / Dynamic models of economic systems and industrial organizations, System Dynamics Review (reprint commentary)", "doi_or_url": "https://doi.org/10.1002/sdr.284", "grade": "A"}, {"source": "Grambsch & Therneau, Proportional hazards tests and diagnostics based on weighted residuals, Biometrika 81(3):515-526 (1994)", "doi_or_url": "https://doi.org/10.1093/biomet/81.3.515", "grade": "A"}], "facet": "mechanism", "chapter": "ch6_analysis_plan", "subclaim": "mechanism"}
{"claim": "Material and information delays between a flow and its effect on a stock generate overshoot and instability, so the ground-intervention delay (one-way light time) is a delayed feedback loop that cannot be treated as an exogenous additive condition; the autonomy effect should be estimated with an autonomy-by-light-time interaction, and a stable hazard ratio across light-time regimes falsifies the deep-space autonomy rationale while an unstable one falsifies the single proportional-hazards coefficient.", "evidence": [{"source": "Forrester, Counterintuitive Behavior of Social Systems (1971)", "doi_or_url": "https://doi.org/10.1007/bf00148991", "grade": "A"}, {"source": "Bellera et al., Variables with time-varying effects and the Cox model, BMC Medical Research Methodology 10:20 (2010)", "doi_or_url": "https://doi.org/10.1186/1471-2288-10-20", "grade": "A"}], "facet": "identification", "chapter": "ch6_analysis_plan", "subclaim": "alternatives"}
{"claim": "A risk-averse program is a balancing feedback loop in which a latent regime stock drives two correlated outflows (autonomy investment and abort/margin/test stringency), so a single-equation Cox coefficient on autonomy is not a structural effect; recovering a causal hazard ratio under treatment endogeneity requires instrumental-variable or structural identification, and the highest leverage lies in loop structure not a single visible parameter, so an unidentified estimate is a within-regime association of unknown transferability, not an architecture-trade lever.", "evidence": [{"source": "Forrester, Counterintuitive Behavior of Social Systems (1971)", "doi_or_url": "https://doi.org/10.1007/bf00148991", "grade": "A"}, {"source": "Martinussen & Vansteelandt, Instrumental Variable Estimation of the Causal Hazard Ratio, Biometrics (2022)", "doi_or_url": "https://doi.org/10.1111/biom.13792", "grade": "A"}, {"source": "Meadows, Leverage Points: Places to Intervene in a System (1999)", "doi_or_url": "https://doi.org/10.4324/9781849773386-15", "grade": "A"}], "facet": "rival", "chapter": "ch7_discussion", "subclaim": "alternatives"}
{"claim": "Case findings generalize analytically to theoretical propositions, not statistically to populations, and the named error is over-claiming statistical generalization from a convenient handful of cases; with a handful of mission-ending events across heterogeneous spacecraft the honest claim is that the hazard ratio tests the autonomy proposition within a narrow flagship stratum (analytic generalization), not that it estimates a population parameter.", "evidence": [{"source": "Yin, Case Study Evaluations: A Decade of Progress? (analytic vs statistical generalization)", "doi_or_url": "https://doi.org/10.1007/0-306-47559-6_11", "grade": "A"}], "facet": "identification", "chapter": "ch7_discussion", "subclaim": "alternatives"}
{"claim": "A rival explanation is discharged only by being explicitly named and tested with a mutually-exclusive predicted pattern, not relabeled as a nuisance term; the program-quality/requirements-stringency regime is statable as a positive rival whose discriminating prediction (within a fixed autonomy level survival still rises with independently-coded stringency, and high-autonomy losses cluster in low-stringency programs) is the opposite of H1's within-stratum prediction, and where treated and untreated differ on a prognostic factor that drives both assignment and outcome the effect is confounded by indication.", "evidence": [{"source": "Beach, It's all about mechanisms: what process-tracing case studies should be tracing", "doi_or_url": "https://doi.org/10.1080/13563467.2015.1134466", "grade": "A"}, {"source": "Sterne et al., ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions, BMJ (2016)", "doi_or_url": "https://doi.org/10.1136/bmj.i4919", "grade": "A"}], "facet": "rival", "chapter": "ch7_discussion", "subclaim": "alternatives"}