On the summary
that is not
the source.

When the matter must change, the System shall change the canonical artefact first, and only then regenerate the view that stands upon it. It shall not amend the rollup as though the rollup were the record.

A neglected Synthesis is the graver danger, for it drifts in silence while reading with full confidence; worse than a neglected Canon, which keeps gaps but keeps its facts true.

The mark shall not be carried in the head; it shall be set where the next hand reads. Where the upstream can be named, staleness is computed: the source is diffed against the view's recorded derivation, not asserted by judgment. The manual assertion is reserved for views of diffuse source that no machine can check.

A manual staleness flag is itself a Synthesis, and so earns its place only where computation cannot reach, and only where its clearing is tied to regeneration, never to a second glance that says it looks fine now. The mark shall record what is false and the source that now contradicts it, that it serve as the brief for regeneration, and not as a bare warning.

A project killed, an initiative archived, a decision superseded: each ends the truth of all that stood upon it then, and not gradually. The System shall sweep and mark those views as part of the deciding act, and shall not leave each to be caught when some later hand acts on it.

For a view bound to a project, the upstream is named even where no file exists to hash: the project's own signal of life, the archived flag of a repository, the decision record, is the proxy to diff against, and so the sweep is computed, not asserted. The failure is the silent drift above, only moved in time: a Synthesis that reads with confidence for days because the decision that killed it never reached the note. The manner of enactment is ordained apart: the change is captured at the decision and applied at the next safe write point, as the change-queue pattern provides.

Derivation cannot tell an accident from an intent: a thing misplaced for an hour yields a confident and false label, for the mechanism reads reality, and reality carries its own errors. The safeguard is therefore not the derived reading alone but the gap it opens against the authored should-be. A derived fact with no stated intent behind it is a mirror that reflects with confidence; it is not a check.

To derive is not to abolish drift but to move it: out of the datum and into the logic that computes it, and the comments that explain that logic. The trade is sound where the fact is volatile and its drift would pass in silence, for the logic then lives in one place that can be tested. It earns nothing where the fact never moves, or where no ground truth exists to project from; there the mechanism is cargo, carried past the point where it pays its weight.