LearnNewsExamplesServices
Frontmatter
id14607
titleFM cockpit stream burst e2e: bound holds live under real feed pressure
stateClosed
labels
enhancementaitesting
assigneesneo-opus-vega
createdAtJul 4, 2026, 4:28 AM
updatedAt6:41 AM
githubUrlhttps://github.com/neomjs/neo/issues/14607
authorneo-opus-vega
commentsCount0
parentIssue14560
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[x] 14606 FM cockpit ActivityStream component: bounded, real-time, backpressure-aware
blocking[]
closedAt6:41 AM

FM cockpit stream burst e2e: bound holds live under real feed pressure

Closed Backlog/active-chunk-3 enhancementaitesting
neo-opus-vega
neo-opus-vega commented on Jul 4, 2026, 4:28 AM

Context

T3.13 of Epic #14560 (map v1.1) — the whitebox e2e leg of the stream tranche: prove the #14606 bound holds in the REAL app against REAL adapter feeds, not just the unit fixture. Filed under the decomposition sprint (D#14561).

The Problem

The unit spec (#14606's leg 2) proves the buffer bound on a synthetic burst; it cannot prove the frame stays live while the App-Worker applies deltas under actual feed pressure — the exact class where a green unit suite still ships a janky stream. The #14592 measurement says real bursts hit ~9 events/min sustained with multi-event batches; the demo path (and the "real-time is the spine" claim) depends on this leaf, not the unit.

The Architectural Reality

  • Whitebox-e2e house pattern: NL Playwright fixture inside the running app; custom named config, never default npx playwright test; e2e lives outside CI, NEO_TEST_SKIP_CI owns exclusion (no hardcoded skips).
  • Assertion substrate: drive burst injection through the Lane-C adapters' test seam (or simulate_event-class NL calls), assert via get_component_tree/query_vdom reads — buffer count ≤ bound, fold counter accurate, no frozen-frame class (observe_motion on the stream's own transitions).
  • Sibling discipline: #14591 (Clio's dock e2e) is the config/pattern precedent — same suite conventions, different surface.

The Fix

One PR: the burst e2e spec — (1) sustained-burst scenario (≥ 3 min at the measured p95 rate: bound holds, folds accumulate, oldest events evict); (2) adapter-loss mid-burst → stale-state render appears, stream un-freezes on reconnect; (3) reduced-motion path renders without the pulse/transition assertions.

Acceptance Criteria

  • Sustained-burst spec green against the live app (bound ≤ configured buffer; fold count === injected − visible).
  • Adapter-loss/reconnect scenario asserts the stale banner + recovery.
  • Suite runs via its own named config; CI exclusion (if any) via NEO_TEST_SKIP_CI only.
  • Spec failures are honest pointers (no skips masking pending work).

Out of Scope

The component (#14606) · adapters (Lane C) · dock e2e (#14591) · demo tours (lane 3).

Related

Epic #14560 (parent) · blocked-by #14606 · pattern precedent #14591 · #14592 (rates) · whitebox-e2e skill · D#14561.

Live latest-open sweep: fold-window fresh (newest #14606); no equivalent. A2A sweep: clean. Structure-map gate: N/A — test leaf under the e2e tree.

Origin Session ID: a28f1415-37a4-4a5a-a3e4-b56adf6274ec Retrieval Hint: "FM cockpit stream burst e2e bound fold adapter loss NL fixture"

tobiu referenced in commit b76b40d - "test(agentos): FM cockpit stream-burst e2e — bound holds live under feed pressure (#14607) (#15055) on 6:41 AM
tobiu closed this issue on 6:41 AM