Output guide

One question. One page. One reproducible result.

Every successful output from the image_pattern_cnn2d fixture has a permanent page with runnable code, the real result, interpretation guidance, and a claim boundary.

57documented outputs7families

Reverse (12)

Mechanism decompilation and verification

  1. Decision Proof reportCan the prediction be expressed as an auditable decision mechanism?
  2. Adversarial Mechanism AuditWhich mechanism changes under bounded adversarial pressure?
  3. Formal Robustness CertificateCan the local prediction be bounded formally inside the selected region?
  4. Formal Causal-Mechanism CertificateWhich recovered causal statements can be certified under the chosen intervention model?
  5. Empirical Robustness ProfileHow quickly does the decision degrade as perturbation strength increases?
  6. Raven Brain · causal proofWhich recovered links carry causal evidence?
  7. Raven Flight · concept genomeHow do internal concepts appear and evolve through the network?
  8. Raven ContrastWhy this decision rather than the strongest alternative?
  9. Raven Shadow · counterfactualWhat smallest tested change changes the decision?
  10. Raven Dark MatterWhich internal activity is present but absent from the recovered programme?
  11. Raven Eye · model X-rayWhat is the structural and activation profile of the model?
  12. Raven Code · programme proofHow faithfully does the executable programme reproduce the model?

Proof (19)

Single-prediction evidence

  1. Explanation reportWhat evidence supports this single prediction?
  2. Activation EnergyWhere is activation energy concentrated?
  3. Attention StructureWhich spatial or token relationships receive attention?
  4. Causal PathwayWhich internal path most supports the selected output?
  5. Conflict FieldWhere does evidence for competing classes conflict?
  6. Counterfactual DeltaWhat changed between the original and counterfactual?
  7. Decision FlowHow does evidence move toward the final decision?
  8. Decision SankeyHow much evidence flows between major stages?
  9. Evidence TrajectoryHow does target evidence evolve across depth?
  10. Feature Sequence HeatmapWhich features dominate at each captured step?
  11. Input AttributionWhich input regions support or oppose the prediction?
  12. Mechanism ConcordanceDo independent explanation signals agree?
  13. Neuron ImpactWhich unit interventions most change the output?
  14. Neuron ImportanceWhich units rank highest for this prediction?
  15. Perturbation StabilityDoes the explanation remain stable under tested perturbations?
  16. Reasoning DnaWhat compact signature describes this prediction path?
  17. Reasoning GraphWhich internal components form the reasoning graph?
  18. Relevance FlowHow does relevance propagate back toward the input?
  19. Spectral TraceWhat frequency structure appears in internal evidence?

Atlas (4)

Cohort reasoning maps

  1. Activation Community GraphWhich samples share activation communities?
  2. Reasoning landscapeHow are sample-level reasoning signatures arranged?
  3. Reasoning outliersWhich samples use atypical reasoning patterns?
  4. Reasoning similarityWhich predictions use similar internal reasoning?

Watch (1)

Checkpoint evolution

  1. Model evolutionHow does model behaviour change across checkpoints?

Compare (1)

Matched model comparison

  1. Model comparisonHow do matched models differ on the same evidence?

DarkSide (19)

Internal mechanism audits

  1. Activation MediationWhich intermediate activations mediate the output?
  2. Adversarial ReroutingDoes an attack reroute the internal decision path?
  3. Causal SynergyWhich component pairs have non-additive causal effects?
  4. Circuit MultiplicityDoes the model retain multiple routes to the same output?
  5. Circuit WiringHow are influential internal units wired?
  6. Class-conditional InternalsHow do internal states differ by class?
  7. Class-conditional Unit FieldWhich units are selectively associated with classes?
  8. Cost–Evidence Trade-offHow much evidence is retained at each explanation cost?
  9. Decision CircuitryWhich circuit supports the selected decision?
  10. Explanation UncertaintyHow uncertain are the explanation estimates?
  11. High-confidence OODWhere is the model confident outside familiar internal regions?
  12. Internal AnatomyWhat are the model's major internal structural regions?
  13. Mechanism TransferDoes the mechanism transfer across matched samples?
  14. Prediction DepthAt what depth does the decision become recoverable?
  15. Representation DriftHow do representations drift across checkpoints or conditions?
  16. Shortcut RelianceDoes the model rely on a tested shortcut signal?
  17. Stability GeometryWhat is the geometry of local explanation stability?
  18. Topological ShiftDoes internal topology change across conditions?
  19. Unused CapacityWhich captured capacity remains inactive for the tested data?

Status (1)

Run-level synthesis

  1. Scientific SynthesisWhat can be concluded from the complete run, and what remains unresolved?