Case packet
Neutral case identifier: case-10d4bb9800d9e7
The possible directed relationship from Artifact A to Artifact B is under review.
Artifact A
A proximate target is checked for achievement, not for warrant
An improvement pathway rarely steers on the thing it is for. Between what ultimately counts as better and the machinery that fires, four things get called the objective, and only two of them are ever the same:
- The improvement objective — what counts as better, [declared as a parameter of the analysis].
- The proximate target — a property pursued because it is held to serve that objective. Retain commitments addressably. Keep coupling low. Widen reflective coverage.
- The operational criterion — what the pathway can actually score. Notes carry rationale, scope, and review evidence.
- The oracle — the mechanism and evidence that applies the criterion.
Two of these levels are proxies, and the theory covers only one of them. The criterion-to-objective relation is the well-worked case: an oracle is [already a stand-in graded by how reliably it discriminates], and the consequences of getting it wrong are understood down to the selection pressure it exerts. The target-to-objective relation carries a claim of the same shape — this property produces that outcome — and nothing in the ordinary machinery is arranged to evaluate it.
The daylight between a target and an oracle
Both stand in for something they are not, which is why the two levels collapse so easily. What separates them is position.
An oracle is a proxy in the verdict position. Candidates arrive; it emits accept or reject. Being a bad proxy therefore means mis-ranking, and mis-ranking leaves a trace: accepted candidates that later disappoint. That trace is what makes the check improvable — a composite oracle [is calibrated by asking whether the code became easier to change or the note supported a better decision], which is only possible because there is a verdict stream to compare against outcomes.
A proximate target is a proxy in the goal position. Nothing is submitted to it. It acts earlier, by shaping which candidates get generated and which designs get built at all, and it is wrong in a different way: not by ranking two things incorrectly, but by the linking claim being false — the property is achieved and the objective does not move.
This is not the bottom of the oracle spectrum. A no-oracle quality is one where verdicts are wanted and unavailable; a proximate target is not in the verdict position to begin with.
Achievement checks make targets look verified
The trap is that a target usually does admit a check — of the wrong proposition. Whether artifacts carry rationale is mechanically decidable. Whether carrying rationale makes later reasoning better is the claim that licensed the target, and it is not what the validator ran.
So a strengthened achievement check delivers exactly the signature of a verified property: a green gate, a rising number, a policy the system enforces. It supplies no evidence for the linking claim, and by supplying that signature it removes the pressure to look for any. The pathway's own selection dynamics then finish the job — a target with a strong achievement check outcompetes the delayed objective it was adopted to serve, for the same reason any strongly discriminated quality outcompetes a weakly discriminated one. The mechanism is not new; the level it operates on is. Underselection was stated over qualities of candidates, and it applies one level up, over what the loop is aiming at.
How a mistaken target is found
The tell is composition failure, transposed. For an artifact it is components that are individually sound and do not add up to the capability, [which is what marks a spec as a proxy theory rather than a definition]. For a target it is the property achieved, its check passing, and the objective sitting where it was.
Reading that signal requires two things the pathway does not supply by default.
The linking claim has to be stated, and stated faithfully. An unrecorded justification cannot be found false; the target simply persists as a settled commitment whose reason nobody holds. This is a design-rationale problem before it is an evaluation problem, and the rationale surfaces where the reason would live [do not enforce continuity from a shipped commitment back to what justified it]. Recording one is not sufficient either, [since a legible but unfaithful rationale makes the repair confidently wrong rather than merely uninformed] — a plausible reason attached to a target that is not the reason it was adopted leaves the composition-failure signal unreadable in the same way as no reason at all.
The objective needs a reading independent of the target. Where the only available measure of artifact quality is that artifacts carry rationale, the linking claim is unfalsifiable by construction — the proxy has been substituted for the thing, and the substitution is invisible because both are called the objective.
Why the levels collapse
A proximate target is a design choice justified by a claim, not [a constraint inherited from a consumer or a substrate] — but it presents like one. Structural properties are stable, statable, and cheap to check for achievement, and those are the same properties that make something look like a settled goal rather than a contestable bet.
The consequence runs back into how architectures get compared. A profile of structural properties selects no ordering by itself; that much is a claim about the descriptive space. But a declared objective does not repair this on its own either, because an objective stated over outcomes cannot rank architectures without some claim connecting structure to outcome. The linking claim is doing that work wherever architectures are being ranked, whether or not anyone has written it down.
Scope
- The claim needs the objective and the target to be different things. Where the property is what is wanted — a compliance requirement that artifacts carry provenance, full stop — it is terminal rather than proximate, and there is no linking claim available to be wrong.
- Nothing here says proximate targets are a mistake. When the objective is delayed, contested, or unmeasurable at the moment of decision, a structural property may be the only thing there is to steer on. The claim is about what the resulting green gate does and does not establish.
- The four levels are a reading, not a fixed depth. A target can stand in for another target, and the chain terminates wherever the analysis declares its objective.
Open Questions
- What makes a structural property a good proximate target before the outcome evidence arrives. Linking claims with [explanatory-reach] — ones that say why the property produces the outcome, and so predict where it would stop — are the obvious candidate. The same discriminator is already argued one level down for retained lessons, [where an abstraction asserted without its condition clause is what over-generalization means]; whether it survives the transfer from lessons to targets is the conjecture.
- Whether calibration against delayed outcomes can reach the linking claim, or only ever reaches the criterion. Asking whether notes carrying rationale were the ones that supported better decisions is an ordinary empirical question; whether any pathway can afford to run it is a different one.
- Whether a reflective pathway can hold its own linking claims addressably, which would make them revisable like any other retained theory rather than settled by the absence of a check.
Relevant Notes:
Artifact B
Self-improvement is relative to a declared objective
A [self-improving system] is defined by change that is causally responsive to [evidence bearing on an improvement objective]. The objective is therefore already load-bearing in the membership test. What has not been stated is its standing in the analysis: it is a declared parameter, on the same footing as the declared boundary and the declared assessment horizon. An attribution of self-improvement is elliptical until all three are named, and comparative attributions are elliptical in a way that no additional precision about the mechanism repairs.
The pattern is established. The same substrate is being improved under one boundary and self-improving under another, [which the boundary cases work through case by case]; the same pathway is currently self-improving over one horizon and only dispositionally self-improving over another. The objective is the third parameter of that kind. Its distinctive feature is that it carries two conditions rather than one, and they are separable.
Indexed and antecedent
Indexed is a condition on the attribution. A claim about self-improvement names the objective it is relative to. Failing it makes the claim incomplete, not wrong — the same failure as reporting an autonomy grade without declaring the boundary it was assessed against.
Antecedent is a condition on the pathway. The objective must be identifiable independently of the particular change it is invoked to license, and evidence bearing on it must have causally shaped that change. Failing it makes the claim false, or empty.
Keeping them apart matters because they are violated by different mistakes and repaired by different means.
Undeclared but antecedent. A parametric learner descends a loss the analysis never names. The pathway is genuinely improvement-directed; only the report is deficient. Naming the objective fixes it.
Declared but not antecedent. An analyst names an objective the pathway is not responsive to — attributing "improvement toward maintainability" to a system whose only operative evidence is latency. Clarification cannot fix this; the attribution is simply false. Under the definition's exclusion for evidence of the wrong thing, the pathway is directed at whatever its evidence is actually diagnostic of.
Neither. The objective is fitted to the change after the fact. This is the failure that makes the category vacuous, and it needs its own treatment.
Antecedent under more than one objective. A single pathway can be responsive to evidence bearing on several objectives at once — a gate that checks structural validity and prose quality supplies both. Indexing is then not bookkeeping: different declared objectives yield different, simultaneously true attributions about the same substrate, exactly as different declared boundaries do. This is why indexing cannot be discharged by simply reading the objective off the pathway.
Neither condition requires the objective to be represented inside the system. Ashby's Homeostat has essential-variable bounds built into its wiring, [as the ultrastability account describes]; nothing in it declares or stores them, and they are still specifiable without reference to any particular reorganization. Antecedence is a condition on identifiability, not on retention.
The causal clause does not close the post-hoc gap
Requiring that evidence causally shape the change already blocks the crude version of a fitted objective: a criterion invented after the change cannot have shaped it, because the causal history is fixed.
What it does not block is redescription. Take gradient steps on a loss and relabel the target as "whatever these weights now do better." The causal story is untouched — the same evidence shaped the same change — and the relabelled criterion is trivially satisfied by the change that occurred. Any operative self-change admits such a relabelling, so if this counted, evidence-responsiveness would separate nothing.
The blocker is antecedence read as independent specifiability: an objective whose only available specification refers to the change it licenses is not antecedent, however intact the causal path. This is the condition the causal clause was tacitly relying on, and it needs stating separately because the two come apart in exactly this case.
What follows
Comparison requires a shared index, and sometimes cannot be had even with one. "This system is more self-improving than that one" is unanswerable until both readings are indexed to the same objective. Indexing is necessary but not sufficient: comparison needs an ordering over states, and objectives differ in how much ordering they induce. A loss or an expected-running-time objective induces a rich one. A viability bound induces only acceptable versus unacceptable, leaving two acceptable configurations incomparable — enough evaluative direction to make change improvement-directed, not enough to rank outcomes. So a fully indexed comparison can still have no answer, for reasons internal to the declared objective. This obstacle is independent of the commensurability obstacle: [comparing per-function autonomy profiles fails for want of a shared decomposition] even where the objective is fixed and richly ordered. Both must be cleared.
The analysis is generic over objectives, not independent of them. The questions the analysis asks — what the update architecture is, what reflection covers, whether the pathway is cumulative, what the methodology settles, who performs which function, what warrant exists — keep their form as the objective varies. Several of their answers do not. Warrant is already stated to be objective-, risk-, and threshold-relative, [since it is bounded by what an oracle can establish]; cumulativity is dependence across episodes in improvement-relevant information, which is objective-relative by construction. The cluster has been doing this consistently; declaring the objective a parameter names the invariant rather than changing the practice.
The profile does not select an order. That coverage, dynamics, governance, and allocation do not collapse into one number is a claim about the descriptive space, and [the placements bear it out]. It is not the claim that no ordering exists, and it entails something sharper about orderings — because the dimensions move independently, and can move in opposite directions under one engineering change, the profile cannot determine how to trade them off. A declared objective is what can, and the ontology should not. Two qualifications keep that from promising more than it delivers. An objective inducing only a partial order settles no trade-off it does not reach, as above. And an objective stated over outcomes does not rank architectures at all without a claim connecting structure to outcome, [since a property pursued as a goal is checked for achievement rather than for warrant]. This is also how the dimensions become goals without becoming grades: greater computational autonomy, wider reflective coverage, and stronger warrant are available as objectives, singly or in combination — as proximate ones, each carrying a linking claim — and moving a gate from a person to a model advances the first while leaving the third where it was.
Open Questions
- Checking antecedence. Independent specifiability is not mechanically testable. Whether it can be operationalized beyond a case-by-case argument — some test on the specification's dependence on the change — is open.
Relevant Notes:
Under-review context phrase
the declared objective the target level stands in for, and the indexing this note adds a level below