Start with the question.

Instrument a real PyTorch forward pass and preserve the internal state needed by later analyses.

• Activations, gradients, attention and layer metadata
• Mechanism IR and reproducible capture summaries
• Trace identity for connecting every result to its source execution

Explain one decision with linked visual evidence rather than an isolated saliency map.

• Attribution, relevance, causal pathways and reasoning graphs
• Counterfactual deltas, perturbation stability and mechanism concordance
• A combined HTML explanation report or one selected export at a time

Recover an executable decision programme, then test whether it reproduces the source model under declared coverage.

• Decision Proof with agreement, fidelity and coverage
• Adversarial mechanism audit and empirical robustness profile
• Formal robustness and causal-mechanism certificates when prerequisites hold
• Seven signature views: Raven Eye, Flight, Brain, Code, Contrast, Shadow and Dark Matter

Move from one decision to a population view of the representation space.

• Activation communities and reasoning landscapes
• Sample similarity, clusters and representation outliers
• Dataset-level evidence with sample identities retained

Follow internal evolution across training checkpoints or model versions.

• Representation drift and decision changes
• Checkpoint-aware traces and model-evolution reports
• Pair with Compare when architectures or trained variants differ

Place two models or runs on the same evidence basis and expose where their mechanisms agree or diverge.

• Prediction, confidence and internal-path comparison
• Comparable visual scales and explicit provenance
• Model-comparison report for review or publication

Investigate the behaviours ordinary explanation dashboards miss.

• Shortcuts, failure anatomy, adversarial rerouting and explanation uncertainty
• Prediction depth, unused capacity, circuit multiplicity and causal synergy
• Representation drift, mechanism transfer and high-confidence out-of-distribution behaviour