The source atlas.
A browsable critical reconstruction of the human-science operating system. Every article has a fixed source revision, an authored critique, an executable alternative and a test.
Scientific protocol
A written sequence is treated as the operative object; implementation drift becomes visible only after execution.
REV 1321308276READ DOSSIER ↗Foundations · W02Scientific method
A human-scale hypothesis–experiment narrative poorly represents high-dimensional, parallel search.
REV 1376633181READ DOSSIER ↗Foundations · W03Philosophy of science
Debates over method are often disconnected from measurable performance of actual inquiry.
REV 1376331797READ DOSSIER ↗Foundations · W04Epistemology
Justification is frequently discussed without a machine-readable trail from proposition to observation.
REV 1376802426READ DOSSIER ↗Foundations · W05Falsifiability
The ability to imagine refutation does not rank competing explanations or localize failed assumptions.
REV 1380088035READ DOSSIER ↗Foundations · W06Problem of induction
Past success cannot deductively guarantee future validity or extrapolation outside observed regimes.
REV 1371948724READ DOSSIER ↗Foundations · W07Duhem–Quine thesis
The tested statement is entangled with assumptions about apparatus, background theory and initial conditions.
REV 1361617395READ DOSSIER ↗Foundations · W08Paradigm shift
Disciplinary paradigms stabilize useful methods but can suppress representation-level alternatives.
REV 1360663769READ DOSSIER ↗Foundations · W09Underdetermination
Several theories can fit the same observations; agreement with history alone is insufficient.
REV 1362414613READ DOSSIER ↗Foundations · W10Scientific realism
A successful predictive model can be confused with a settled claim about underlying entities.
REV 1370998013READ DOSSIER ↗Foundations · W11Scientific model
Static models hide version history, parameter changes and domains where predictions break.
REV 1371659107READ DOSSIER ↗Foundations · W12Occam's razor
Simplicity can become aesthetic authority rather than a calibrated trade-off.
REV 1379603268READ DOSSIER ↗Inference · W13Bayesian inference
Posterior certainty is conditional on model classes and priors that are often left unchallenged.
REV 1379760658READ DOSSIER ↗Inference · W14Statistical hypothesis test
A thresholded decision compresses uncertainty, design choices and effect sizes into a ritual verdict.
REV 1379362199READ DOSSIER ↗Inference · W15Null hypothesis
A point null is sometimes operationally convenient but mistaken for a meaningful physical absence.
REV 1379362199READ DOSSIER ↗Inference · W16Statistical significance
A p-value is often reinterpreted as the probability a hypothesis is true.
REV 1361974577READ DOSSIER ↗Inference · W17Statistical power
Underpowered experiments generate unstable estimates and encourage selective attention to large effects.
REV 1378995428READ DOSSIER ↗Inference · W18Multiple comparisons problem
Hidden flexibility in the number of questions tested inflates apparently impressive findings.
REV 1376255469READ DOSSIER ↗Inference · W19False discovery rate
Global error controls can be applied without confronting dependence or shifting hypothesis families.
REV 1374927830READ DOSSIER ↗Inference · W20Confidence interval
An interval may be given a probability interpretation it does not support and may omit selection effects.
REV 1376255302READ DOSSIER ↗Inference · W21Meta-analysis
Combining available studies can precisely estimate a biased published subset.
REV 1379623909READ DOSSIER ↗Inference · W22Regression toward the mean
Extreme observations can generate spurious improvement or deterioration narratives.
REV 1368360447READ DOSSIER ↗Bias · W23Data dredging
Exploratory searches are routinely mistaken for tests whose evidential threshold was fixed beforehand.
REV 1377936004READ DOSSIER ↗Bias · W24Publication bias
The scientific record is filtered by publishability and statistical excitement.
REV 1368850151READ DOSSIER ↗Bias · W25Confirmation bias
Investigators preferentially see support for expected explanations.
REV 1370294982READ DOSSIER ↗Bias · W26Selection bias
Who enters the dataset can determine the apparent result.
REV 1375130988READ DOSSIER ↗Bias · W27Survivorship bias
Only successful or visible cases inform a seemingly convincing causal story.
REV 1379606971READ DOSSIER ↗Bias · W28Sampling bias
A large sample can be precise yet systematically unrepresentative.
REV 1375044021READ DOSSIER ↗Bias · W29HARKing
A post-hoc explanation is presented as if it were the prediction that motivated the work.
REV 1359619333READ DOSSIER ↗Bias · W30Pseudoreplication
Dependent observations are counted as if they were independent experimental units.
REV 1367999065READ DOSSIER ↗Bias · W31Cognitive bias
Human heuristic errors are treated as individual defects despite being amplified by institutions.
REV 1379095929READ DOSSIER ↗Bias · W32Observer bias
Instrument and coding choices can systematically shift observations before analysis begins.
REV 1353554158READ DOSSIER ↗Reliability · W33Replication crisis
Validation arrives late, after strong claims and reputational accumulation.
REV 1379868267READ DOSSIER ↗Reliability · W34Reproducibility
Code and methods may be described but not executable outside the authors' environment.
REV 1379031988READ DOSSIER ↗Reliability · W35Replication (statistics)
Replication is treated as a new paper rather than a continuously linked evidence update.
REV 1360732077READ DOSSIER ↗Reliability · W36Peer review
A limited set of reviewers acts as a bottleneck and imperfect proxy for validity.
REV 1379295044READ DOSSIER ↗Reliability · W37Metascience
Reform is studied in disconnected publications, often far from the research workflows being evaluated.
REV 1373322115READ DOSSIER ↗Reliability · W38Open science
Sharing artifacts does not ensure they are usable, correctly linked, or independently inspectable.
REV 1379400605READ DOSSIER ↗Reliability · W39Preregistration (science)
Research intentions can be registered but later analytical decisions remain hard to reconstruct.
REV 1372960129READ DOSSIER ↗Reliability · W40Registered report
Pre-result review reduces selective publication yet still routes recognition through a journal.
REV 1375710612READ DOSSIER ↗Reliability · W41Systematic review
Reviews depend on slow search, manual screening and publication access.
REV 1372813059READ DOSSIER ↗Reliability · W42Research integrity
Rules around misconduct do not by themselves guarantee an inspectable chain of custody.
REV 1377152310READ DOSSIER ↗Reliability · W43Research transparency
Transparency is treated as an author practice rather than a property enforced throughout execution.
REV 1375658312READ DOSSIER ↗Experimentation · W44Design of experiments
Fixed-factor plans are often selected before observing the full information structure of a problem.
REV 1379887652READ DOSSIER ↗Experimentation · W45Randomized controlled trial
Randomization helps identification but can be undermined by noncompliance, attrition and poor transport.
REV 1350325633READ DOSSIER ↗Experimentation · W46Scientific control
A control is specified locally but often lacks explicit causal justification.
REV 1375480330READ DOSSIER ↗Experimentation · W47Blinded experiment
Blinding is fragile, and unblinding is frequently insufficiently measured.
REV 1357239547READ DOSSIER ↗Experimentation · W48Natural experiment
Apparently exogenous exposure may hide violations of the identification assumptions.
REV 1348959741READ DOSSIER ↗Experimentation · W49Observational study
Large observational datasets invite confident causal narratives without identifying interventions.
REV 1346158126READ DOSSIER ↗Experimentation · W50Causal inference
Models may estimate associations accurately while causal assumptions remain unverifiable.
REV 1368546517READ DOSSIER ↗Experimentation · W51Confounding
Unobserved common causes can persist despite sophisticated adjustment.
REV 1371518385READ DOSSIER ↗Experimentation · W52Counterfactual conditional
Counterfactual claims require assumptions that cannot be observed directly for each unit.
REV 1368679740READ DOSSIER ↗Experimentation · W53External validity
An internally valid result is routinely projected beyond the population and setting studied.
REV 1370590349READ DOSSIER ↗Experimentation · W54Internal validity
Methodological checklists can miss unexpected pathways from assignment to outcomes.
REV 1374310121READ DOSSIER ↗Experimentation · W55Measurement uncertainty
Final error bars frequently omit uncertain calibration and preprocessing decisions.
REV 1367370535READ DOSSIER ↗Experimentation · W56Adaptive clinical trial
Adaptation can boost efficiency but introduces operational bias and statistical complexity.
REV 1364723345READ DOSSIER ↗Verification · W57Automated theorem proving
Search for proofs is conflated with independently establishing that a theorem statement was proved.
REV 1380246457READ DOSSIER ↗Verification · W58Proof assistant
Machine-checked derivations remain conditional on encodings, axioms and trusted components.
REV 1380223205READ DOSSIER ↗Verification · W59Formal verification
Proving software properties may ignore a mistaken specification or an external physical failure.
REV 1377238699READ DOSSIER ↗Verification · W60Lean (proof assistant)
The presence of Lean code can be mistaken for an independent certification of the intended mathematics.
REV 1374563227READ DOSSIER ↗Verification · W61Computational science
Numerical results can inherit uncontrolled approximation, environment and implementation choices.
REV 1376349250READ DOSSIER ↗Verification · W62Scientific computing
A successful computation is often reported without enough executable dependency detail.
REV 1376349250READ DOSSIER ↗Verification · W63Scientific workflow system
Workflow orchestration is optimized for throughput rather than epistemic traceability.
REV 1369246739READ DOSSIER ↗Verification · W64Laboratory automation
Automated experimentation still risks optimizing a flawed objective or unreliable instruments.
REV 1377714125READ DOSSIER ↗Verification · W65Active learning (machine learning)
Selecting useful labels can exploit dataset artifacts while neglecting scientific identifiability.
REV 1378688515READ DOSSIER ↗Institutions · W66Research ethics
Ethical evaluation can appear as a one-time approval external to the research execution.
REV 1368094508READ DOSSIER ↗Institutions · W67Scientific journal
Editorial acceptance becomes a social gateway through which scientific results obtain visibility.
REV 1379833314READ DOSSIER ↗Institutions · W68Academic publishing
Article incentives often reward narratively tidy novelty rather than cumulative correction.
REV 1375045430READ DOSSIER ↗Institutions · W69Bibliometrics
Citation volume measures attention, not independent correctness or the value of error detection.
REV 1377099702READ DOSSIER ↗Institutions · W70Citation analysis
Citation networks may propagate untested claims and self-referential chains of authority.
REV 1370697715READ DOSSIER ↗Institutions · W71Open data
Making data downloadable does not guarantee lawful reuse, sufficient context or interpretability.
REV 1380195714READ DOSSIER ↗Institutions · W72FAIR data
Findability and interoperability principles can be satisfied superficially without epistemic reproducibility.
REV 1370308559READ DOSSIER ↗Institutions · W73Citizen science
Mass participation can increase coverage yet introduce unequal measurement quality and overlooked local expertise.
REV 1375457912READ DOSSIER ↗Institutions · W74Scientific community
Institutional and disciplinary silos impede recombination of findings into common models.
REV 1369270996READ DOSSIER ↗Institutions · W75Research misconduct
Fabrication and selective disclosure can survive polished papers and plausible machine-generated narratives.
REV 1379337313READ DOSSIER ↗Institutions · W76Science studies
Institutions, power and laboratory practice influence what counts as knowledge, but their effects are rarely modeled operationally.
REV 1373996015READ DOSSIER ↗Institutions · W77Scientific consensus
Agreement among experts is useful but cannot replace the underlying evidence and conditional inference.
REV 1330333272READ DOSSIER ↗Institutions · W78Goodhart's law
Optimizing publication counts or benchmark scores can corrupt the underlying purpose of science.
REV 1379318143READ DOSSIER ↗No dossiers match. Clear the search or choose another domain.