Rendered from docs/decisions/0052-a-document-is-proposed-at-the-state-it-will-hold-and-the-merge-activates-it.md in the Headwater corpus. Every document on this half of the site is typed by the taxonomy the descriptor names: corpus.json.

A document is proposed at the state it will hold, and the merge activates it

Context

HW-DR-0034 rules that acceptance is the merge onto main. A provenance block on a branch states a proposal, and the merge decides whether the stamp was true. That ruling reaches the provenance block and no other part of a document.

The state facet had no such ruling, and the two facets moved apart. headwater new writes the opening state of the regime the kind binds, which is draft. No document, no skill, no hook and no rule then says what moves a document off that state, or when. So one event carries a declared meaning for warrant and no meaning at all for status.

The corpus measured the gap on 2026-09-06. Of 277 typed documents, 66 stand at draft. HW-DR-0047 and HW-DR-0048 are merged, deployed and serving requests. Both stand at draft, whose declared guidance reads "the document is being written or argued over, and nothing may rely on it". The obligation shelf splits at HW-OBL-0126: records 0001 to 0125 stand at current or discharged, and records 0126 to 0161 stand at draft bar two. One kind, one shelf, one meaning, and no record of a change in practice.

Four whole shelves stand at draft entire. They are the nineteen interface contracts, the four probes, the one tutorial whose fixture suite blocks CI, and the requirement beside its acceptance criterion. A second session met the same absence while writing two new records. It looked for what draft means on the decision shelf, and found nothing to copy but the deviation.

Decision

An author writes the state the document will hold once the branch lands. For a document proposed as finished, that state is current. The merge activates it, the way the merge activates the acceptance stamp beside it.

draft on a branch states that the document is not proposed as finished. It is the honest value for work a reader is asked to comment on rather than rely on. It is not a placeholder that an author leaves because the scaffolder wrote it.

The ruling is symmetric with HW-DR-0034 and needs no actor that promotes a state at merge time. Nothing moves the value between the branch and main. The state a proposal carries is the state the merged document carries, and the merge decides whether that value was true.

A document already on main at the wrong state is corrected rather than left. The correction moves status and stamps status_since with the date of the correction, because that is the date the document entered the state.

Consequences

What a writer meets. headwater new writes draft, because a regime opens at its initial state and a scaffolder reports the value it derived. The author moves the value before proposing the document. The authoring skill states this beside the paragraph that states the acceptance stamp, so a writer meets both readings of the merge in one place.

What this corpus owes. The 66 documents standing at draft are merged, and this ruling says what they should read. The change that carries this record corrects them and stamps each one with the date of the correction. The dwell each document spent at draft stays in the history rather than in the facet, which is the ordinary cost of a correction.

What no rule reads. HW-OBL-0030 records that no taxonomy declares the provenance block, so nothing reads the acceptance stamp. The state facet is declared, and lifecycle.state.not_admitted and lifecycle.transition.not_permitted both read it. Neither asks whether a merged document is finished, because neither has a merge to read. So this ruling rests on the process this repository runs rather than on a control the engine holds.

What this makes reportable. #569 holds a rule that would report a live document resting on one at a state whose role is initial. That rule is worth having once a draft document is rare, and it is noise while two thirds of a shelf stands there. This ruling is what makes the population small enough for the rule to say something.

What reopens this. A kind whose documents are argued over on main for weeks would need draft to mean what it says on a merged document. No kind of this corpus works that way today.