Claims become inputs to other claims. A measurement feeds a calculation, a calculation feeds an assessment, an assessment is filed and later retrieved as record. The problem statement calls this the laundering path and observes that no falsehood is introduced at any step while the probabilistic origin of the first inference becomes unrecoverable. The inheritance rule is what makes a basis survive the derivations performed on it, and it is the last part of the discipline to be fixed.
One piece of it is already settled and is not reopened here: a derived claim's origin set is the union of its parents', which the record fixes because the field is otherwise undefined for derived claims. What remains is the other fields, and the question of how a claim's basis may change once it has been derived.
The fields under derivation
Acquisition and transformation describe the derivation itself, never inherited from a parent. A calculation over measurements is an inference, not a measurement, and this is the single most important thing the rule does. The laundering path works precisely by letting a derived claim present itself with the character of its inputs, so a rule that let a claim inherit a parent's production kind would rebuild the failure it exists to prevent. Where a person performs the derivation the transformation is a human step; where a model performs it, a generation; where an algorithm with stated semantics performs it, a calculation.
Establishment time is the time of the derivation, and this is not the same as the freshness of what it rests on. A calculation performed now over a measurement taken yesterday was established now, and its parent's age is not erased by that: the rule carries the derivation's own establishment time, and the oldest establishment time among its ancestors alongside it. Both are needed, because an act may care that the calculation is current and equally that the reading beneath it is not stale.
Rate of change comes from what the derived claim is about, not from its parents. A claim about a tank level derived from a flow integration changes at the rate a tank level changes. This is the one field a derivation does not propagate at all, because it is a property of the world rather than of the basis.
Established scope narrows to the intersection of the parents' scopes. A derivation cannot establish something over a wider bound than its narrowest input: a calculation combining a plant-wide figure with a single unit's reading establishes something about that unit and not about the plant. Where the intersection is empty the derivation has combined claims about disjoint things, and the result carries no established scope rather than a false one.
Authority is the deriver's, with the parents' authorities retained. Who performed the derivation is a different question from who stood behind its inputs, and both bear on whether an act should proceed. An unimpeachable calculation over a reading from an uncalibrated instrument is not made sound by the calculation, and a rule that replaced the parents' authority with the deriver's would hide exactly that.
What survives derivation, and how a claim's basis may change
The design first asked when a derived claim may hold a stronger basis than its parents, and answered that disjoint origin sets license a bounded, axis-specific strengthening. Building the artefact against that answer refuted it. The question was wrong, and what replaces it is narrower and easier to check.
There is no operation that raises a claim's standing. Nothing in a basis is promoted, and no field describing how a support came to exist is ever rewritten by a later step. Two reports do not become a measurement, two inferences do not become a calculation, two generations do not cease to be generations, a second old source does not refresh the first, and two independent authorities do not make either more authoritative. Each of those fields records how a particular support came about, and acquiring a second support does not alter the history of the first.
What may change is which supports the basis contains. That change is explicit, recorded, and refusable, and the verdict on the claim follows from reevaluating the changed basis rather than from any status stored on it.
What derivation preserves
A derived claim's origin set is the union of its evidential parents' origins, less any explicitly elided, plus any fresh acquisitions, plus the typed event that produced the claim itself. That last term is not a detail. Without it a generation would be visible only while it remained the current mechanism and would disappear at the next derivation, so a generated claim passed through a deterministic extraction would present as a plain calculation over measured ancestry. That is the laundering path rebuilt one step later, and including the current transformation is what closes it.
An origin that survives a derivation keeps its identity and its kinds. A later step may add a typed event; it may not rewrite a generation as a calculation or an unverified assertion as verified preservation. Duplicate parents and parent ordering are inert, because origins form a set.
Causal inputs need not become evidential ancestry. A generated hypothesis that prompts an investigation caused the claim to be examined; it does not support it, and a measurement that would stand had the hypothesis never existed begins its own set. That judgement is declared by the construction and can be got wrong, which moves the opportunity for laundering from an invisible place to a stated one rather than removing it.
Evidential origins disappear by exactly one route: an explicit elision naming the removed origin, the ground on which it was judged causal rather than evidential, what made that judgement, and the basis warranting it. The elision travels in the record, because the receiving boundary is the one place the discipline claims to work, and a reset recorded only in a local audit log is invisible exactly where it matters. A demand may refuse because an elision occurred, or because its warrant is inadequate or unknown.
How a basis may change
Composition. An independently established support may be added to a basis. The origin sets union, both supports keep every descriptive field they had, and nothing is promoted. Where a trusted authority corroborates an untrusted one, the claim now has support from a trusted authority; the untrusted source has not been improved, and a demand requiring two acceptable authorities may now be satisfied because both supports are represented.
Fresh acquisition. New evidence entering from outside the system supplements the existing basis by default. Where it instead replaces the grounds the claim rested on, that is a separate declaration, and each displaced origin needs its own travelling reset. A constructor that discarded parents whenever fresh evidence appeared would strip generated ancestry by accident.
Verification. A preservation claim that has been checked is a different record from one merely asserted. This does not raise the semantic relation; it supplies a warrant that was previously absent, and an unverified assertion of restatement remains semantically unknown.
What the origin set settles, and what it does not
Intersecting origin sets establish dependence, and dependence forbids treating two supports as independent corroboration. That veto is mechanical, answerable from the records alone, and it is asked before any question of warrant: one source counted twice is not two supports, whatever independence has been asserted.
Disjoint origin sets establish only that no dependence is recorded, which is a weaker fact and grants nothing. Two readings may share a calibration standard, an upstream database or an unrecorded human report, and no intersection will show it. Absence of recorded dependence is not evidence of independence, so composition additionally requires independence to be established on a basis the record does not itself supply.
An assertion of independence is itself a claim, and a discipline that took such assertions on trust would reopen at that point exactly the hole it closes at the point of action. Consistency requires that an independence assertion carry its own basis. That recursion is noted and not built: the model treats independence as an input either established externally or absent, and reproducing the limitation explicitly is more honest than concealing it inside a constructor.
The test the rule has to pass
The failure this rule exists to prevent has a description older than the field. A person justified in believing something false may infer from it something true, by a valid deduction, and end up with a belief that is true and justified and worth nothing, because the justification has been severed from what makes the conclusion true (Gettier, 1963). Justification survives the deduction; its connection to the truth does not.
That is the laundering path stated as epistemology, and it is the test the rule must pass. The rule passes it by refusing to let a derived claim present itself with the character of its inputs: a conclusion derived from a generated premise carries that generation in its origin set, and the record says so however valid the derivation was. Validity of the derivation is not a property the rule propagates, because validity is exactly what Gettier's cases have and what does not help.
What the rule does not catch is the case the record already records as beyond it. Where a derivation happens outside the discipline, a person reading a value and authoring a fresh claim, no origin is unioned and no transformation event is recorded, because the model never saw the step. That claim enters as a report with its own origin, and its dependence on what it was drawn from is invisible. The rule governs derivations the discipline mediates; unmediated laundering is refused at admission or not at all.
What this costs
Two things. Retaining ancestors' authorities and the oldest ancestral establishment time means a derived record carries more than a claim's own basis, and a long derivation chain accumulates. The origin set has the same property, named earlier as a residual to watch in the scenarios; this rule adds to it rather than introducing it.
And composition is available less often than it looks. Parents drawn from the same instrument, the same model run, or the same report share an origin, and in a plant that is the common case, so the mechanical veto fires routinely. Where origins are disjoint, independence still has to be established elsewhere before the supports may be treated as separate. The result is conservative in practice, which is the correct direction for a discipline about unwarranted authority and is still a cost: a claim that is genuinely well supported will often not be able to show it in terms the record can check.
References
Gettier, E. L. (1963). Is Justified True Belief Knowledge? Analysis, 23(6), pp. 121-123. doi.org/10.1093/analys/23.6.121