When a machine initiates a consequence, natural language collapses several different questions into one word: responsible. MICRM is a formal reference model that refuses the collapse — it keeps material structure, authority, accountability, and the evidence for any of it as separate objects, so a conclusion is only ever as strong as the weakest step that was actually established.
This is the formal companion to the argument in the essays The Wall Was Never There and The Assumed Control: those name the failure in prose; this is an attempt to give the analysis a semantics precise enough to challenge.
The distinction it enforces
MICRM separates two things that everyday reasoning merges, and evaluates them independently:
- The structural relation — that one traversal is materially ancestral to a consequence, that authority and accountability terminate at the same principal, that a contribution class attaches at a particular point. This is a claim about what the system did.
- The evidentiary warrant — the support for asserting that relation. The same record can corroborate one finding and merely assert another. Support is evaluated per basis and relative to the principal the proposition would advance — because a party often controls the very record that favors it, and a search can look complete only because the party who benefits defined the universe being searched.
Keeping these apart is the whole point. An asserted relation does not silently become an established finding, and an upstream structure that looks plausible does not promote a downstream conclusion on its own.
What it outputs — and refuses to
MICRM produces structural findings and candidacy states, not liability. It deliberately does not convert unresolved evidence into an exculpatory finding: a high rate of structural non-exclusion is not a high rate of accountability — it can simply be the model refusing to treat missing or interested evidence as a favorable exit. Where a finding rests on absence (a negative inference), MICRM requires completeness over the named population rather than leaving that as an unstated premise, and the trust root where the support recursion terminates declares its assumptions instead of concealing them.
Legal fault, moral blame, and apportionment are explicitly outside the formal semantics. They belong to whatever overlay consumes MICRM's structural output — the model's job is to make the structure and the warrant inspectable, so disagreement can be aimed at a specific step rather than the whole conclusion.
How it was tested
The validation is the part I'd point a skeptic to first, because it was built to fail.
Fact patterns were authored outside the lane that wrote the semantics, sealed and published by hash before the deriving side read a single byte. Every derivation ran against frozen rule bytes it cites; no fact pattern was edited after a result was seen; where a rule was missing, the branch stopped rather than inventing one mid-derivation. Eleven sealed variants across five independently authored patterns exercised selected mechanisms of the model.
The independently authored challenge found twelve rule defects against frozen bytes — none of which the project's own self-check had caught.
That result is reported as evidence consistent with the model's own claim that self-checking is not a valid verification method — not as proof the model is sound. The process also caught subtler traps: a variant that matched its pre-sealed prediction for the wrong reason (the predicted mechanism never ran), which is why every later derivation checks the route, not just the result.
What it does not establish
The paper is deliberate about its ceiling, and so is this page:
- It does not establish that MICRM's semantics are correct — only a reproducible record of what was tested, against which rule bytes, with which outcomes.
- Substantial surface remains untested, named individually: the graph model under load, the class layer end to end, multi-representation candidacy, attachment and propagation, burdens, and the overlay interface.
- It is designed for cross-substrate use but does not establish substrate neutrality — that stays a substrate-specific validation question.
- The evidentiary architecture is itself a surveillance surface; privacy and disclosure gaps are identified but not normatively resolved here.
- No third party has reproduced any finding. The archive is an internal product of the project it documents.
The two artifacts
The paper is expository; the formal semantics (definitions D1–D30) are normative and carried by a separate specification, cited by version and content hash. Where the two disagree, the specification governs and the paper is defective — no normative content exists only in the prose. That separation is what lets the paper explain the model without quietly becoming a second source of truth for it.
The full paper carries the complete related-work differentiation, the validation record, the privacy and adversarial analysis, the checked figures, and the provenance appendix. The paper and the specification blocks it cites are bound to a public repository release, with the presented commit and per-artifact SHA-256 hashes recorded, so the reproduction procedure is executable from the published release.