frontend-patterns/workplans/FEP-WP-0006-language-evolution.md

---
id: FEP-WP-0006
type: workplan
title: "Make ongoing language learning observable"
domain: infotech
repo: frontend-patterns
status: blocked
owner: codex
topic_slug: canon
created: "2026-09-26"
updated: "2026-10-04"
state_hub_workstream_id: "528d24c1-c2c4-5914-a065-992b0dd877c8"
---

# Make ongoing language learning observable

User-approved baseline, learning ledger and generated evolution report. Finite
investigations support an ongoing language; no composite maturity score or automatic
promotion. M1/M2 and independent human validation remain open in WP-0009/WP-0003.

## Establish the versioned evidence baseline

```task
id: FEP-WP-0006-T01
status: done
priority: high
state_hub_task_id: "a23a0de7-9f53-5135-b412-3e556725ad9b"
```

Seventeen research identities, current profiles and immutable baseline pin candidate
content and distinguish documentary/static/runtime evidence, implementation families,
review provenance and limits. Full-contract independence and historical effort remain
unknown. See research/evolution/README.md and snapshots/baseline-2026-09-26.json.

## Record learning and generate transparent reporting

```task
id: FEP-WP-0006-T02
status: done
priority: high
state_hub_task_id: "5085abed-813d-585a-ba3f-b67727e93c4d"
```

Selected retrospective ledger, prospective cycle template, generated evidence/case/
usefulness report, validation and tests. Preserve the frozen pilot and all historical
experiment receipts. Intent/protocol/program now describe continuous learning.

## Measure the first comparable prospective cycles

```task
id: FEP-WP-0006-T03
status: wait
blocking_reason: "Needs useful comparable prospective evidence and an explicit protocol/returns review from WP-0003 or consumer-led WP-0008; largest declared group has two cycles, one with unknown effort."
priority: medium
state_hub_task_id: "2c5c4bae-a0a9-54bf-ac86-49b2803fdd8c"
```

Before WP-0003 T01's next navigation/history/cache investigation, register one bounded
cycle using templates/LearningCycle.md. Record effort, outcomes and counterevidence;
update profiles/report and snapshot at the review boundary. Continue with comparable
questions only where they are useful, not to meet a count target. After at least three
comparable cycles review returns within that stream/protocol; no trend claim before
then and no automatic plateau declaration. Preserve independent confirmation value.

Reconcile substantive-review dates from original artifacts; baseline assessment dates
must not fabricate freshness. Future usefulness work retains stable development anchors
and preregisters fresh tasks under WP-0004; it does not alter the consumed pilot.

## First prospective cycle completed — 2026-09-27

LEARN-0011 preregistered at f3697c1; htmx 2.0.11 navigation/history/cache experiment
meets 20/20 registered checks (18 browser rows and repeated explicit HTTP checks).
Both browsers reproduce missing-Vary fragment-as-document failure. Outcome is
corroboration, not a new discovery. Active/execution effort recorded in
research/evolution/cycles/2026-09-27-navigation-effort.json; one cycle cannot show a trend.
Candidate profiles/report updated; hypothesis status and provisional VC-03 retained.

WP-0003 T01 retains keyboard/focus/error-feedback and delayed/interrupted enhancement
questions; T02 retains independent walkthrough review and offline-reading branch.
WP-0006 T03 remains progress for comparable-window review and remaining substantive
review-date reconciliation. Choose a new comparison group for materially different
questions. M1/M2 and independent-human validation remain open. See
research/reviews/NavigationHistoryExperiment.md.

## Measured keyboard/focus cycle — 2026-09-27

LEARN-0012 registered at 2b1c9a4 in new comparison group F-navigation-focus-v1.
22/22 native/raw/managed rows complete; raw rows are observations, not fulfillment
of managed obligations. Same-ID focus retained by raw htmx, while authored policy
adds heading/title/status continuation and preserves newer search focus. Outcome
corroboration, hypotheses retained; no AT or human validation, no trend across the
two different prospective groups. See research/reviews/NavigationFocusExperiment.md.

WP-0003 T01 next: delayed history restoration and competing-navigation ownership;
interrupted/late enhancement remains open. T02 retains AT/independent review and
offline branch. WP-0006 T03 remains progress for comparable windows and remaining
review-date reconciliation. No M1/M2 completion.

## Completion ownership compared; navigation fixture paused — 2026-09-27

LEARN-0013 preregistered at e484cae; 24 completed rows after a retained keyboard-helper
correction. Prior policy meets full obligations in 0/12, authored guard 12/12. Record
one revision: check navigation completion ownership separately from focus permission,
including history, errors and newer-request failure. Same authored family; hypotheses
and provisional VC-03 retained. See research/reviews/NavigationOwnershipExperiment.md
and NavigationThreadReview.md. Three distinct cycle groups do not establish a trend.

Agreed pause: do not automatically extend the navigation fixture. Next bounded
WP-0003 T01 question is a preregistered, pinned read-only Zulip navigation/search
source contrast: locate selection/result/focus ownership or counterexamples outside
the authored fixture; prior draft/send evidence does not count as navigation support.
T01 retains pending history disclosure, multiple misses, redirects, delayed swaps,
bfcache/lifecycle and interrupted/late enhancement before any production claim.
T02 retains independent human/AT review and offline VC-03 branch. WP-0006 T03 retains
review-date reconciliation and useful comparable-cycle windows; no M1/M2 closure.

## Independent Zulip navigation source study completed — 2026-09-27

LEARN-0014 preregistered at f3ac569; twelve files at existing Zulip revision
5b06fa3d34f28fa01f459fbf3ba053e3c0858e9c, static ranges/hash manifest and ZN01–ZN08
review retained. Main list-identity guards coexist with off-screen cache updates;
shared feedback, auxiliary callbacks and deferred focus limit generalization. One
revision: effect-specific applicability and explicit client-fragment addressability.
No Zulip app/tests executed, no no-script/full-contract/bug claim. Same Zulip family,
new navigation paths; all candidates remain hypotheses. Local fixture stays paused.

WP-0003 T01 follow-up: preregister a bounded caller/transport/test study of auxiliary
target lookup and anchor continuations, including deferred focus/search wiring only
if needed to establish applicability. Record unknowns; reproduce before claiming a
defect. T02 retains independent human/AT and offline-case obligations. WP-0006 T03
retains remaining review dates and comparable-window work; this static protocol is
not pooled with runtime cycles. No M1/M2 closure. See
research/reviews/ZulipNavigationSourceStudy.md and ZulipNavigationSources.json.

## Scope–intent assessment — 2026-09-27

Updated SCOPE.md to distinguish current research capabilities, intended reference
capabilities and exclusions. The timestamped assessment in
[history/](../history/2026-09-27T010237+0200-scope-vs-intent.md) prioritizes usefulness,
independent recurrence, contrasting coverage/coherence, reference access and
comparable learning evidence. Recommendations map to existing WP-0002/0003/0004/0006
tasks; statuses and M1/M2 gates are unchanged. This documentation assessment is not
a measured empirical cycle. T03 retains review-date and comparable-window work.

## Authority/projection cycle completed — 2026-09-27

LEARN-0015 preregistered at 2df1d94; six primary documents across four families,
six seed dispositions and two synthetic VC-01/02 decision traces. One classification
revision: separate scoped acceptance, projection placement and delivery schedule.
Zero new drafts; all seventeen hypotheses and candidate profiles unchanged. No
application execution, full-contract recurrence, human validation or M1/M2 closure.
See [review](../research/reviews/AuthorityProjectionComparison.md). Fifth distinct
prospective group; no comparable-window trend.

WP-0003 T01 next: preregister one bounded pinned-application study of shared-state
acceptance versus local editing and a projection transition, using the review's six
questions. Select the application by the needed evidence; retain contrary/unknown
paths. Existing Zulip auxiliary continuation residual stays open; local navigation
fixture remains paused. WP-0002 T03 retains independent usefulness review; WP-0003
T02/T03 gates and WP-0006 T03 review dates/comparable measurement remain open.

## Pinned acceptance/projection study — 2026-09-27

LEARN-0016 preregistered at ff0c1d4; ten pinned Zulip source/test files inspected.
[Review](../research/reviews/ZulipAcceptanceProjection.md) corroborates the six-question
aid: tentative local rendering, locked server validation/rendering and event-driven
replacement have different meanings. Response/event ordering, lost-response recovery
and event ownership of a newly opened edit remain unknown; no app/upstream tests run.
Source manifest retains hashes, ranges and the failed web-tool retrieval. No candidate,
profile, 0.1 reference or maturity change; no new independent implementation family.

WP-0003 T01 next: preregister a bounded response/event ownership reproduction around
an edit reopened after HTTP success, an event before response, response lost after
acceptance and navigation away before failure. Include echo-pending eligibility guards
in the setup; do not infer a defect from callback inspection. Reuse pinned application
infrastructure if feasible within a new budget. Existing auxiliary-navigation residual
stays open, local navigation fixture stays paused, and WP-0003 T04 retains storage/
receiver validation. WP-0006 T03 retains review dates/comparable-window work; this sixth
prospective group supplies no learning trend. Human validation and M1/M2/M3 stay open.

## Closeout audit — 2026-09-27

T03 review-date reconciliation is complete: [dated provenance report](../research/reviews/ReviewDateReconciliation.md)
resolves fourteen unknown dates and corroborates the three known dates from original
artifacts. Current-profile validation requires review provenance/scope and prevents
a date later than its assessment. Historical baseline and 0.1 release are unchanged.

T03 remains progress, workplan active: LEARN-0011–0016 belong to six different
comparison groups, none with three comparable cycles. The returns review needs useful
further prospective research under the existing task; do not manufacture repetitions
or pool incompatible protocols to close it. There is no external blocker on that
research and no claim that a metadata reconciliation is a new learning cycle.

## Second B-independent-navigation-source-v1 cycle — 2026-09-27

LEARN-0017 preregistered at `884803c`, a naturally-motivated follow-up to LEARN-0014
(not a manufactured repetition): it answers ZN07's open question about the target-
lookup and anchor-navigation continuations' missing guard. Comparison group
B-independent-navigation-source-v1 now has two cycles, still short of the three
needed for a comparable-window returns review. See
[research/reviews/ZulipTargetLookupCallPath.md](../research/reviews/ZulipTargetLookupCallPath.md).

T03 remains progress: reopening the paused local navigation fixture to manufacture
a third cycle there was considered and explicitly declined, per the existing agreed
pause and this task's own rule against forcing a count. A third comparable cycle in
this or another group should only happen when a genuinely useful next question
arises. No trend claim, no M1/M2/M3 closure.

## Loose-end review — 2026-09-28

T03 waits; workplan blocked on the comparable-evidence prerequisite. The largest
prospective group still has only LEARN-0014/0017 (two cycles; 0017 effort unknown).
LEARN-0018 completes WP-0003 T04's concrete storage/receiver experiment in a new
runtime group; it cannot be pooled with those static-source inspections. Review-date
reconciliation remains complete. No qualified three-cycle returns window exists.

This is an evidence dependency, not a credential or implementation blocker. Resume
when useful research under existing WP-0003 T01 or consumer-led WP-0008 supplies a
comparable window with honest effort accounting. Do not restart the paused navigation
fixture or manufacture repeated studies merely to close T03. No new task/workplan;
no diminishing-returns, independent-human or maturity claim.

## Resumed measurement audit — 2026-10-04

The user explicitly resumed this workplan from backlog. Reviewed the current ledger,
registered cycle protocols and reporting implementation. T01/T02 and substantive
review-date reconciliation remain complete. Eight prospective cycles occupy seven
declared groups; B-independent-navigation-source-v1 has LEARN-0014/0017, with
LEARN-0017's effort honestly unknown. No complete three-cycle window exists.

Implemented a deterministic [window audit](../research/evolution/ComparisonWindows.json)
alongside the evolution report: ordered cycle IDs, protocol references, outcome
counts, positive/unknown/zero effort, qualitative review triggers and complete-window
effort readiness. A matching group label no longer appears to establish comparability.
Three declared entries trigger protocol review; missing effort still permits scoped
qualitative review but cannot silently produce an effort comparison. Regression cases
cover unknown/zero time, nonconsequential outcomes, mixed streams, retrospective
exclusion, ordering and stale generated audit rejection. The cycle template now asks
why a follow-up is useful and requires accounting gaps to stay visible.

This is measurement infrastructure, not a new empirical learning cycle; no ledger
outcome, candidate freshness, maturity or missing historical time is invented. Updated
the report's open-work references to the live M1 owner WP-0009 and consumer-led WP-0008.
Removed the stale exclusive dependency on parked WP-0003 T01: either owner can supply
useful comparable evidence under its existing scope. No downstream research gate is
waived. T03 remains wait and the resumed workplan is blocked on that evidence; the
paused navigation fixture stays paused. Existing work records retain the next useful
research and returns review; no replacement backlog is created.