Intuition Machine
Nothing breaks. Something is quietly rewritten.
Every pathology in the theory is a small syntactic edit to a healthy word. The rewritten version still type-checks, still runs, and still reports on schedule.
Pathologies as word edits
EDIT TYPE PATHOLOGY
──────────────────────────────────────────────────────────────────────
Letter deletion degraded audit — Δ(ν,ν), the fold removed
⦿G bypass — the V-word flattened, never reaches ω
Provenance substitution stale input dressed as current; state ossification;
context starvation; a preference presented as a wall
Operator substitution false dual; phantom pool
Index freezing one attentional column unavailable — its words
untypeable, its cells unperformed verbs
Word truncation the loop fails to close
Type coercion model output stored as fact without the audit gate
Route substitution misroute — content matching loop X delivered to Y
Route deletion route starvation — a healthy loop never engaged
Binding decay orphaned convention — grounded, but no living editor
Edit at the spine IMPOSSIBLE — quality cannot be rewritten,
only refused
The last row carries the hopeful theorem. Because spine edits are impossible and the reflexive loop's critical move is just a double descent, every quality-bypass is a routing pathology around an intact capacity. The pathologies that look most entrenched are structurally the cheapest to reverse.
A rulebook can disable a capacity without touching it
Theorem T9 protects the quality-sensing capacity: nothing the diagnostics recommend can move the standard the diagnostics measure against. But a routing rule set that simply never sends anything to the reflexive loop disables the protected capacity without editing it — the way a hospital could neutralise its untouchable trauma team by never paging them.
So no fully compiled rule set may be the sole gate on engaging quality-sensing. The invocation must retain either a live-judgment path or an alerting bypass — Beer's algedonic channels, which the theory already carried, turn out to be the existing recognition that routine routing must be bypassable.
Seven layers, in order
Each layer is a typing question about the system's words, and each presupposes the previous layer's judgment. There is no point asking whether a loop's controlled variable is aligned if the loop does not close.
1 Structural does the word exist and close?
does a selector exist; is every arrival routable?
2 Attentional is every letter index-typed?
3 Content-phase do positions match levels; provenance tags healthy?
is routing type-correct for the content's level?
4 Evolutionary are interface levels matched to stage?
5 Harness are both phase conversions and the audit gate intact?
6 CV is the system's controlled variable aligned with the
collective's? including the selector's own
7 Transparency do all convention-level traces have cross-boundary
witnesses? does every standing rule have a living editor?
Who holds the fold-witnesses
The capture indicators of the diagnostic practice — context-authorship concentration, decision-beneficiary correlation, reviewer diversity, taxonomy concentration, runbook visibility — are all measurements of one underlying thing: who holds the fold-witnesses, and how far claims sit from their folds.
To those, version 5.3 adds one forward-looking measurement of the same thing: who holds the edit rights over each standing convention, and whether those editors are currently functional.
Sixteen viability conditions state what must be true of a system's words, interfaces and perceptions for it to remain viable. They are also the source of the walls that carve any resource tradeoff's feasible region — the theory reinterprets the conditions it already had rather than adding new ones for the purpose.
Start with one workflow.
We map your constraints, build a governed AI workflow alongside your existing operations, and prove the results before autonomy expands.