tegwick
da06464f6b
Assistant: codex Assistant-Model: gpt-6-astra Assistant-Session: 01a0cffe-206f-7323-89d3-b20ca4cf4bf9
4.6 KiB
| title | status | maturity | families | scales | admission | |||
|---|---|---|---|---|---|---|---|---|
| URL-addressable State | draft | hypothesis |
|
|
research-candidate-only |
URL-addressable State
Context: readers need to share, bookmark and revisit a resource or selected view. Problem: an address change can conceal state that only exists in the current tab. Addressability competes with private/transient state and URL/schema stability.
Proposed solution: encode the public, stable resource identity and relevant view selection in path/query parameters, or in a fragment when a declared client runtime reconstructs the view. Define validation, defaults and the required resolution path so direct entry and reload reconstruct that selection under current authorization. A client-fragment contract does not imply no-script availability. A URL identifies a view, not necessarily immutable bytes or guaranteed future availability. Use an explicit version if a stable historical snapshot matters.
Do not place secrets, private draft content, focus or every transient interaction in the address. An identifier confers no authorization. Server-visible query/path state and fragment-only state have different resolution contracts. If history is enhanced, back/forward, cache misses and direct loads must resolve coherently; pushState success alone is insufficient.
Consequences: views become independently reachable, but parameter compatibility, invalid/deleted resources and full-versus-fragment cache behavior need contracts. Alternatives: ephemeral in-tab selection when sharing is unnecessary; authenticated saved-view identifiers when the selection is too large or private for the URL.
Evidence: navigation review N01/N02/N06 locates documentation mechanisms, not verified deployments of this whole proposal. Full-pattern known uses and recurrence remain unknown. Refines the existing URL as Application State seed; no canonical identity or admitted relations.
Proposed connection: Progressive Enhancement can preserve native navigation while adding faster transitions; it is optional, not required by addressability. VC-03 is a synthetic trace.
Prospective runtime corroboration — 2026-09-27
Direct filtered entry, reload, snapshot restoration and explicit history-miss recovery preserve selection in a pinned htmx fixture. Omitting representation Vary reproduces a fragment served as a full document in Chromium and Firefox; correct selection keeps representations distinct, without promising cache retention or freshness. See NH01–NH06. One authored application with a real package, not independent deployed recurrence. Hypothesis retained.
Competing completion ownership — 2026-09-27
NO01–NO06 show a delayed history response replacing content/title while a newer URL remains selected in the unguarded fixture. Treat navigation initiation, including Back/Forward, as intent with an explicit completion owner. Before applying an asynchronous result, check that it still owns the selected route; include URL, content, title and feedback in that decision. A response to obsolete intent is not made current by arriving last or succeeding.
The experiment's latest-intent policy keeps the last committed view and newest error when a newer request fails; it does not revive older pending navigation implicitly. Other policies require an explicit contract. Browser history can change the URL before loading completes: pending-state disclosure/recovery is still unresolved here. The authored guard supports this obligation only in the tested schedules, not a portable router or proof of independent application recurrence.
Independent application source contrast — 2026-09-27
Zulip ZN01–ZN08 locates pinned hash-to-view reconstruction, separate history pointer/offset state, and current-list guards on main fetch continuations. It supplies static partial application evidence; no direct load or Back race was executed in Zulip. Auxiliary lookup/anchor paths and deferred focus remain unverified.
Applicability belongs to each effect: noncurrent results can still populate cached data without acquiring the right to change current selection. Object/list identity and a navigation generation are different implementation policies; neither alone proves freshness. Keep addressability separate from no-script enhancement and from within-session focus/scroll restoration. Full-contract recurrence remains unassessed.