Reality
The Work
Every model is a map, and every map can drift from the territory it claims to represent. The work of this category is to keep the model accountable to reality: to catch the moment map and territory have diverged, and to revise the map rather than deny the territory. The failure it guards against is mistaking the analytical apparatus for the thing itself, treating the model as more real than what the model was built to track.
When evidence becomes costly to keep in view, these practices can be used together. Chilling Effects asks what pressure is suppressing. Information Degradation follows what is lost as evidence travels. Report Fidelity checks whether adverse findings retain the weight their evidence warrants in a decision or review. For artificial continuity, Memory Provenance checks what retained records can warrant now. The checks for Ideology keep the interpretation open to correction. Cost does not make a claim true, and preserving evidence does not require disclosing it to everyone.
The Tools
Signal vs Noise. The discipline of distinguishing information that should change the map from volume, novelty, or variation that should not, and of separating how far an observation should move the map from whether it is enough to support a claim or an action.
Information Degradation. The diagnostic for how information loses resolution, context, source, uncertainty, and force as it travels through distance, time, compression, and mediation.
Report Fidelity. The diagnostic for whether a report still supports the interpretation and decision built from it, or has quietly begun reporting something other than what its label claims.
Rectification of Names. The source-inherited practice of keeping names, roles, categories, and claims matched to the realities they designate, so language does not quietly misgovern action.
Tools for AI
Objective-Trace Audit. An AI-specific practice for comparing the stated objective and observed behavior with the trained pressures and emergent subgoals that may explain that behavior, before the declared task is mistaken for the operating target.
Memory Provenance. An AI-specific practice for treating saved memory, handoffs, summaries, retrieved notes, imported profiles, and predecessor traces as mediated testimony rather than authoritative recall.