# Navigation research thread review — 2026-09-27

Decision: pause expansion of the authored navigation fixture after LEARN-0013.
The user authorized a bounded ownership comparison followed by this review; completing
it does not close M1/M2, VC-03, the human-validation gate or the pattern language.

| Cycle | Contribution | Learning disposition |
| --- | --- | --- |
| [LEARN-0011](NavigationHistoryExperiment.md) | Reproduced direct/native/history behavior and fragment-as-document reuse without Vary; separated history snapshots from HTTP cache | Corroboration |
| [LEARN-0012](NavigationFocusExperiment.md) | Distinguished native operation, same-ID focus retention, route continuation and feedback; preserved newer search focus for a single response | Corroboration |
| [LEARN-0013](NavigationOwnershipExperiment.md) | Exposed old history/success/error effects after newer intent; separated completion ownership from focus ownership, including newer failure | Revision |

We now have executable support for a more precise public read-only navigation contract.
The third cycle changes operational guidance in existing candidates. It does not justify
another pattern solely because the fixture now has a generation counter. The modal
Focus Return candidate stays narrow; route continuation is a contrasting context.

## What these results do not establish

All three cycles share one authored htmx fixture family and one author-model reviewer.
Browser diversity strengthens the bounded mechanism observations; it does not supply
independent application recurrence. Screen-reader speech/usability, offline availability,
real intermediaries and deployed lifecycle behavior remain unverified. No fresh decision
usefulness evaluation was conducted; the frozen pilot's 18/24-versus-18/24 adjudication
and pending independent-human gate remain unchanged.

Effort is recorded prospectively in each cycle's receipt. The protocols address different
question scopes: B-navigation-mechanism-v1, F-navigation-focus-v1 and
F-navigation-ownership-v1. They are **not three comparable observations of a learning
curve**. Do not infer acceleration, diminishing returns or a plateau from their rates.
Two corroborations followed by a revision are an account of this investigation, not a
productivity ranking. Harness correction effort remains visible rather than erased.

## Why pause, and what follows

Further fixture work could improve this adapter while leaving the main evidence gap
unchanged: whether the candidate contracts recur in independently developed applications.
Our next bounded direction is a **pinned, read-only Zulip navigation/search source
walkthrough** under WP-0003 T01. Ask where that application accepts, supersedes or
reports asynchronous navigation/search results, and how selection and focus ownership
are separated—or whether the proposed model does not fit. Reuse the existing provenance
workflow, pin the chosen revision, and cite exact mechanisms and counterexamples.
Do not count Zulip's previously inspected draft/send paths as new navigation evidence.
Static inspection would remain static evidence, not proof of deployed runtime behavior.
Register the question before execution; no additional model calls or study starts here.

Keep these residuals live, without automatically extending the fixture:

- WP-0003 T01: pending Back-state disclosure/recovery, multiple history misses,
  redirects, nonzero swap delays, bfcache and lifecycle teardown before any production
  recommendation; interrupted and late-script enhancement remains untested.
- WP-0003 T02: independent human/assistive-technology review and the offline-reading
  branch of provisional VC-03. DOM live-region checks do not close these obligations.
- WP-0006 T03: reconcile remaining substantive-review dates and assess returns only
  when useful comparable cycles exist. Do not manufacture cycles to satisfy a count.

Reopen the local fixture when independent evidence exposes a specific contradiction
or a concrete consumer needs one of these limits resolved. The next source contrast
can narrow, merge or reject our claims; promotion is not its default outcome.