Superseded choices need a historical witness; refuted beliefs lose subject-matter standing

Type: kb/types/note.md · Tags: kb-maintenance, artifact-analysis, foundations

When a system refutes a belief about some subject matter, the displaced assertion loses standing. It was wrong, and the fact that the system once held it makes nothing about the subject true. When a system supersedes a residual choice, the earlier prescription stops governing, but the choice event remains true: the system committed to X at t. At least one reliable witness to that event must survive when no other retained artifact can reconstruct it.

The asymmetry comes from the role that remains after displacement. A belief answers to evidence independent of what the system once thought. A choice event answers to the act of choosing. Supersession does not keep the old prescription current, but it cannot undo that act. Destroying every reliable witness does not make the past event false; it only makes the event unavailable to a later bounded reader.

What "no standing" rules out

"The system held B at time t" is itself truth-apt and can be worth retaining. That is a historical observation about the system, not the refuted assertion about B. It earns retention on evidence about the past, like any other belief. A time-indexed subject-matter claim that remains true also falls outside the refuted case. Replacing current guidance does not falsify a claim about an earlier time.

The choice side also contains two roles. "The system committed to X at t" remains the same true event-fact after supersession. "Do X now" may be the old choice's expired prescription and has no current authority. The obligation is to keep the event checkable, not to keep treating the displaced prescription as operative. A refuted belief must therefore re-earn retention in some other role, while a genuine choice event keeps historical standing.

Production relation selects the operation, not what must survive

Commitment, not derivation, creates new ground truth draws the production boundary at whether source material plus the consumer goal determines the retained content. A change to determining inputs calls for refresh or re-derivation. Changing an addition the inputs did not determine calls for a later commitment or supersession.

That production relation selects the maintenance operation. It does not settle what, if anything, must survive displacement. An ampliative conjecture is a committed addition because its sources do not determine it, but what it asserts remains a belief. If later evidence refutes it, the current theory should be rewritten. Preserving the old wording as history is a separate retention decision. A residual choice is also committed, but its event-fact remains true after supersession and therefore needs a reliable witness.

Timing makes that witness load-bearing. Since history has one chance to become checkable, an unrecorded choice is often not reconstructable from the artifacts it shaped. Those artifacts may show the winning option without showing that a decision was taken, which alternatives remained live, or which forces were weighed. Other contemporaneous records can satisfy the obligation when they preserve those facts. The requirement is at least one reliable witness, not every duplicate.

Two maintenance regimes follow

A record of choices grows by appending: each supersession adds a link in a chain whose earlier events stay readable and true. A body of beliefs is reworked holistically: when evidence changes the understanding, the current text states what is now believed. Prior wording remains only when a separate historical use earns its retention.

Commonplace shows that both regimes can coexist. Decision records carry lifecycle status, supersedes/superseded-by edges, and the alternatives that lost. Theoretical notes instead rewrite refuted claims, leaving version control to retain bytes without granting them standing. Its proposal archiving decision also shows that a superseded choice can leave the active frontier while its historical witness survives. This instance establishes feasibility, not unique correctness.

Scope

Where applicable requirements, constraints, and beliefs determine a selection uniquely, no residual choice was made. The apparent choice is a derivation recoverable from its determining inputs. Dropping its record costs recomputation, not the history of a free selection.

Where a belief's subject matter is the system's own commitments, the historical proposition may remain true. The same is possible for a subject-matter claim whose explicit time index remains true after current guidance changes. Neither case is a refuted belief retaining standing it lost.

Mixed artifacts require region-level treatment. An ADR can contain a choice event, an expired prescription, and truth-apt rationale. Retaining the artifact keeps a witness to the first without granting equal standing to every region. A belief that was responsibly warranted at t may likewise deserve preservation as evidence of a past knowledge state, but that historical use must earn its own place.

The obligation is against losing every reliable witness, not against deleting a duplicate or demoting history from active attention. Which maintenance operation applies remains keyed to production relation by artifact classification.

Relevant Notes: