Pinky BBB: user-steer fast-lane parity (v1.63.0) #93

Merged

ClaudiuSchuster merged 1 commit from oss-oo/pinky-bbb-fastlane into main

2026-10-09 23:12:25 +00:00

ClaudiuSchuster commented

2026-10-09 23:11:36 +00:00

Owner

Steer #57 — the owner filed it rightly: machine bells beat the owner's own words to a session.

D1 — user steers on the fast lane (deterministic ~3 s):

SSE turn-end nudge seam: the dashboard's push consumer POSTs /api/fastlane/nudge on session.turn-* events (250 ms coalesced); the watcher waits on the wake event with the poll cadence as timeout (v1.30 discipline — the event accelerates, the poll guarantees).
Settle self-wake: a candidate blocked only by the settle window arms a one-shot timer that fires the lane at the first legal instant — deterministic even without a push feed.
Human-first priority by construction: the lane claims only obs- steer entries at the boundary while queued bells physically ride the sweeper's 15 s cadence + the dispatcher's 30 s activity grace — the user steer delivers first even when the bell queued older. A bell that seizes a boundary inside the settle window (dispatcher-side spawn race) delays the steer by exactly that turn (re-fire at its end ~3 s later), never the sweep cadence.
Cross-lane note for Brain-A filed (priority field on queue entries + yield-to-human precheck in the notify spawn) in the README's delivery-lanes section.

D3 — the two GGG folds:

GGG-F1: Tier-B bell-witness intent-acceptance resolved doc-consistently (KEPT deliberately — dispatch.py writes intents only at terminal finish inside notify_parent; pinned both ways: intent-only completed witnesses, intent-only failed still refuses; docstring + README state the justification).
GGG-F2: test_owner_unread first-sight ms-race fixed (fresh reply at marker+1; solo 5× green).

D4: README documents the delivery lanes (bells fast, user steers fast + priority, sweep backstop) and the honest boundary.

Evidence: rig (hermetic, fake dispatcher, sanitized sweep tree): user steer delivered 2672 ms after the terminal turn row with an OLDER bell queued (bell followed 1153 ms later via the sweep); fastlane-dead backstop leg green; round-XX zombie leg green (dismissed steer never delivers; resurrect killed by the guard). Fixture CDP smoke green (nudge wiring present, zero console errors). Suite 867 green; Super-Linter station recipe green.

Steer #57 — the owner filed it rightly: machine bells beat the owner's own words to a session. **D1 — user steers on the fast lane (deterministic ~3 s):** - SSE turn-end nudge seam: the dashboard's push consumer POSTs `/api/fastlane/nudge` on `session.turn-*` events (250 ms coalesced); the watcher waits on the wake event with the poll cadence as timeout (v1.30 discipline — the event accelerates, the poll guarantees). - Settle self-wake: a candidate blocked only by the settle window arms a one-shot timer that fires the lane at the first legal instant — deterministic even without a push feed. - Human-first priority by construction: the lane claims only obs- steer entries at the boundary while queued bells physically ride the sweeper's 15 s cadence + the dispatcher's 30 s activity grace — the user steer delivers first even when the bell queued older. A bell that seizes a boundary inside the settle window (dispatcher-side spawn race) delays the steer by exactly that turn (re-fire at its end ~3 s later), never the sweep cadence. - Cross-lane note for Brain-A filed (priority field on queue entries + yield-to-human precheck in the notify spawn) in the README's delivery-lanes section. **D3 — the two GGG folds:** - GGG-F1: Tier-B bell-witness intent-acceptance resolved doc-consistently (KEPT deliberately — dispatch.py writes intents only at terminal finish inside notify_parent; pinned both ways: intent-only completed witnesses, intent-only failed still refuses; docstring + README state the justification). - GGG-F2: `test_owner_unread` first-sight ms-race fixed (fresh reply at marker+1; solo 5× green). **D4:** README documents the delivery lanes (bells fast, user steers fast + priority, sweep backstop) and the honest boundary. **Evidence:** rig (hermetic, fake dispatcher, sanitized sweep tree): user steer delivered **2672 ms** after the terminal turn row with an OLDER bell queued (bell followed 1153 ms later via the sweep); fastlane-dead backstop leg green; round-XX zombie leg green (dismissed steer never delivers; resurrect killed by the guard). Fixture CDP smoke green (nudge wiring present, zero console errors). Suite 867 green; Super-Linter station recipe green.

ClaudiuSchuster added 1 commit

2026-10-09 23:11:36 +00:00

Pinky BBB: user-steer fast-lane parity — the turn-end wake seam + human-first boundary (v1.63.0)

Checks / Functional checks (pull_request) Successful in 33s

Details Explain this check

8a526a1ef0

D1 (steer #57): queued USER steers now deliver at a deterministic ~3 s
after the target's turn ends, event-driven like the bells' own lane.
Three seams: the SSE turn-end nudge (dashboard push consumer POSTs
/api/fastlane/nudge on session.turn-* events, 250 ms coalesced; the
watcher waits on the wake event with the 1.5 s poll as timeout — the
v1.30 discipline), the settle self-wake (a candidate blocked only by
the settle window arms a one-shot timer that fires the lane at the
first legal instant — deterministic even without a push feed), and
human-first priority pinned by construction: the lane claims only obs-
steer entries while queued bells physically ride the sweeper's 15 s
cadence + the dispatcher's 30 s activity grace — the owner's words
deliver first even when the bell queued older; a bell that seizes a
boundary inside the settle window (dispatcher-side spawn race) delays
the steer by exactly that turn (re-fire at its end), never the sweep
cadence. Cross-lane note for Brain-A (priority field + yield-to-human
precheck) filed in the README's delivery-lanes section.

D3 GGG-F1: Tier-B bell-witness intent-acceptance made deliberate and
doc-consistent (docstring + README state the justification: intents are
written only at terminal finish inside notify_parent; requiring
delivered_at would re-open the binary-swap rows) — pinned both ways.
D3 GGG-F2: test_owner_unread first-sight race fixed (reply at marker+1;
solo 5x green).
D4: README documents the delivery lanes (bells fast, user steers fast +
priority, sweep backstop) and the honest boundary.

Rig evidence (G1/G2, ~/.zcode/tmp/pinky-bbb-rig): user steer delivered
2672 ms after the terminal row with an OLDER bell queued — bell
followed 1153 ms later via the sweep; fastlane-dead backstop leg and
the round-XX zombie leg (dismissal + resurrect guard) green; fixture
CDP smoke green (wiring present, zero console errors). Suite 867 green
(the lone red is the pre-tag version guard); Super-Linter station
recipe green.

ClaudiuSchuster merged commit b03ec7bf5a into main

2026-10-09 23:12:25 +00:00

ClaudiuSchuster deleted branch oss-oo/pinky-bbb-fastlane

2026-10-09 23:12:26 +00:00

ClaudiuSchuster referenced this pull request from a commit

2026-10-09 23:12:27 +00:00

Pinky BBB: user-steer fast-lane parity (v1.63.0) (#93)