docs/constraint-floor.md
Attacking Observe and H(O)
⚠️ Extended by sufficiency-test.md. Distinction fails a constructive sufficiency test: it is symmetric, and equality, ordering, time, causation and identity all require asymmetry. It is half the floor, not the floor.
Outcome: Observe is not primitive — I named the surviving verb after one of its
own parameter values. H(O) reduces to a constraint system and dissolves an open
item it had carried twice. Constraint survives as semantic, not ontological, and
the ontological floor is distinction, which is machinery already derived.
1. Reducing Observe
The proposed decomposition fails — instructively
Observe = establishing correspondence + asserting a constraint. Both halves are existing machinery, so the decomposition is legitimate to attempt.
It fails. Establishing correspondence produces a co-reference claim, and a co-reference claim is itself a constraint — it restricts which assignments are possible. So correspondence is not a second operation beside assertion; it is assertion about identity rather than about a value.
The decomposition yields one operation over two subject-matters, not two operations.
The reduction that succeeds
inference-operations.md concluded: "One verb: assert a
constraint on H, parameterized by target and warrant." Then it named the verb
Observe.
That was wrong, and wrong in a way this investigation has already caught three
times. Observation is one value of the warrant parameter — the evidential
one. Naming the verb after it is the same error as naming a fact family measure
after its observation method, and as putting kind inside a URN: a value
promoted into the thing it is a value of.
The verb is assert a constraint. Observation is that verb with an evidential warrant.
Fourth instance of one error, and the first time it appeared in my own vocabulary rather than in the code.
Confirmed by domains with no observer
| Domain | Constraint? | Observation? |
|---|---|---|
| pure mathematics | yes — axioms constrain models | no |
| cellular automata | yes — the rule constrains successors | no |
| physical law | yes — F = ma constrains trajectories |
no |
| organizations | yes | yes |
Three domains have constraints and no observers at all. Constraint is strictly more general; observation is a special case. Had I promoted Observe, these three would have falsified it immediately.
2. H(O) reduces
No implementation enumerates hypotheses — SAT solvers, CSP engines, Bayesian
inference, theorem provers and physics all operate over constraints. So does H
carry weight, or is it notation?
Test: express its four states constraint-side.
H state |
Constraint-side |
|---|---|
|H| = 0 |
the constraint set is unsatisfiable |
|H| = 1 |
every variable of interest is entailed |
|H| > 1 |
some variable is free |
| undefined | the variable is not well-formed |
All four are properties of the constraint system. Nothing requires the set to be built.
H(O)is explanatory notation for a constraint system. It reduces.
It dissolves an open item
Two documents left the same admission: "|H| is defined and uncounted; nothing
says how to count it for a real record."
It was never countable, because counting was the wrong question. The real question is "is this variable determined?" — entailment, which is decidable in many systems and is what every listed technology actually computes. An open item carried twice was an artifact of the notation, not a gap in the theory.
The architecture simplifies accordingly:
constraints (each with a warrant) → entailed truths → projection
and projection is itself constraint-assertion over a model, so the last arrow is not a different kind of step.
3. Layers — with an honest audit
The demand is that a primitive survive only if it cannot move to another layer. Applying it required auditing past rounds, and the audit is not flattering.
| Concept | Layer | Note |
|---|---|---|
| distinction | ontological | see §4 |
| constraint | semantic | a relation between a proposition and possibilities |
| entailment / determination | semantic | |
| value space | semantic | a set with an equality |
| persistence extent | ontological | how truths exist in time |
| dimension | mixed | value space (semantic) + persistence (ontological) — not a clean primitive |
| separation, identity | semantic | derived from persistence |
| warrant | epistemic | |
| evidence, provenance | epistemic | first- and second-order |
| lineage | ontological | |
| the selection rule | conventional | a sixth layer the original five omit |
| co-reference | spans three | lineage (ont.), data association (epist.), selection (conv.) |
H(O) |
computational | eliminated |
collapse() |
computational | wrongly treated as semantic — that is the bug |
| the four fact families | representational | eliminated |
What the audit shows
Several earlier "reductions" eliminated implementation strategies, not ontological primitives, and I did not always label which. Specifically:
- eliminating
Eliminateremoved a computational step; - eliminating the four fact families removed a representational enumeration;
- reducing interpolability to absent-evidence semantics was an epistemic relocation, not an elimination.
Only the reductions of constraint-as-truth-kind, relation, and capability were ontological. The reduction count across this investigation is therefore smaller than it has appeared, and the layer labels are the correction.
Two things also fall out that were not asked for. dimension is a mixed
construct spanning two layers, so dimension-algebra.md's
"two primitives" is one semantic and one ontological — not a clean pair.
And convention is a sixth layer: the post-fission selection rule is not
ontological, semantic, computational, representational or epistemic, and the
five-layer scheme has nowhere to put it.
4. Constraint or distinction?
Constraint sits in the semantic layer, so by the rule it is not the ontological primitive. What is?
§4's candidate was state distinction. It fails: mathematical objects have no states, and mathematics was one of the domains that carried constraints. So state-distinction is too narrow.
Drop "state":
Distinguishability. A value space is a set with an equality. Everything above operates over what that equality separates.
This is not a new concept — it is the equality already absorbed into value spaces in dimension-algebra.md, now recognized as the floor rather than as a component.
And it explains the investigation's most persistent anomaly. The separation principle has been reached five times by independent routes:
- missing-versus-zero (provenance)
- absent-evidence semantics (truth kinds)
- idempotent occurrences (the occurrence ledger)
- co-reference (subject identity)
- information-as-elimination (the hypothesis space)
Five arrivals at one principle was recorded each time as a curiosity. It is not. Separation is distinction, applied at five different layers. A principle that keeps reappearing across layers is a candidate for the floor, and that is the strongest evidence available here.
It also explains the chemistry result exactly: indistinguishable particles have no identity because there is no distinction to build one from. Not a limitation — the floor is absent, so nothing above it can stand.
Can distinction reduce?
Attempt: distinction requires two things and a relation, so it presupposes plurality — is plurality more primitive? No: two things are two because they are distinguishable. Plurality and distinction are the same claim.
No further reduction found. One round. By this investigation's own standard that is not promotion.
5. Standing
| Concept | Verdict |
|---|---|
| Observe | not primitive — a warrant value mistaken for the verb |
H(O) |
eliminated — notation for a constraint system |
| constraint | survives, semantic |
| distinction | survives, ontological — the current floor |
| warrant | survives, epistemic |
| convention | a sixth layer, previously unnamed |
The architecture, restated:
distinction (ontological)
→ constraint (semantic)
→ warrant (epistemic)
→ representation → projection
What is still not established
- Distinction has survived one round. The burden is to eliminate it, not to defend it, and nothing here does that.
- Whether constraint is genuinely semantic rather than ontological depends on a contested question — whether physical laws govern or describe — which I have not settled and cannot settle from here. The classification is asserted with its dependency stated, not derived.
- The layer scheme itself is now six, having grown by one under pressure. A scheme that grows when tested is not yet stable.
capability— ten rounds untouched.
The observation that this has stopped being an ontology of organizations is correct, and the raised burden is accepted: nothing above has been demonstrated outside this repository's own domain, and three of the four domains in §1 were reasoned about, not measured.
Nothing implemented.