Rendered from docs/process/decisions/0026-we-pivot-rather-than-carry-debt-forward.md in the Headwater corpus. Every document on this half of the site is typed by the taxonomy the descriptor names: corpus.json.

Project process. These notes record how Headwater itself is built and run. You do not need them to use Headwater. Back to the specification

We pivot rather than carry debt forward

Context

The additive campaign of HW-DR-0111 measured its layers in fourteen scenarios. Its tuning round closed under #1777. The re-cut of milestone 0.07, #1868, replaces that design with eleven actions, five layers and seven scenarios. A result from the new design does not compare with a baseline from the old one. So the re-cut had to choose: keep the old design for the comparison, or take the loss.

The owner stated the principle below on 2026-10-28, during the re-cut, and approved this wording on 2026-10-29 (#1869). Before this record, neither DEVELOPING.md nor a document under docs/process/ stated a rule about a structure that a better one replaces.

Decision

The owner's words, 2026-10-28, approved 2026-10-29:

We pivot rather than carry debt forward. When a better structure costs us something we already have, such as the ability to compare with earlier runs, a baseline or a probe set, we take that loss and undo the work. What was measured stays as a dated record; what was built is kept only while it is the best structure we know.

Consequences

The first application is the 0.07 re-cut, #1868. It retires the fourteen-scenario baselines of HW-DR-0111 and starts new ones. The results of the fourteen-scenario campaign stay on the corpus as dated records. A separate decision of #1868 supersedes the scenario design of HW-DR-0111, and this record does not.

So the loss of a comparison, a baseline or a probe set is not, by itself, a reason to keep an old structure. A result measured before a pivot is a dated record, and it is not a baseline for a result measured after the pivot.