loop-lane convention: blocked-lever protocol — a lane must execute in-authority levers, escalate out-of-authority ones, and signal blocked-idle before backing off

[kyle-sexton]

Filed by AI on operator direction (attended feedback session, 2026-07-24).

Category: behavioral · Source: operator observation of a live babysit-loop lane, 2026-07-24

Problem — a lane that names its own levers, then pulls none of them

Observed cycle report from a running source-control:babysit-loop lane (standing mode):

Cycle 2 done — 0 deltas, 0 fixes, 0 merges, loop alive, next check-in 50 min. State unchanged since cycle 1. Queue = 11 human-gated PRs; #853 still the lone C2 candidate, still conflict-held. Lane cannot do anything autonomous until a lever moves: resolve #853's conflict in an isolated worktree, gate and merge; or classify more PRs so C2-mechanical become eligible; or stop.

The lane correctly diagnosed its levers — then executed none, escalated none, and backed off toward the idle ceiling. The operator returned hours later to a lane that had effectively given up minutes in. Three distinct defects:

1. In-authority lever left unexecuted. Conflict resolution is inside the wrapped mechanic's own authority (babysit-prs's fan-out and conflict-worker contract; the loop skill's Subagents section routes exactly this through it). A conflict-held merge-eligible PR is the textbook dispatch case. If a resolved autonomy dimension (e.g. barrier handling set to escalate-only) or tier flooring denied the dispatch, the cycle report must say WHICH config denied it — "cannot do anything" with the capability present and unexplained is a defect either way.
2. Out-of-authority levers never converted to escalations. "Classify more PRs" belongs to the triage/attended lanes — correct that the merge lane didn't do it. But the loop-lane convention's escalation contract (human-gated role label + machine-marked escalation comment) exists precisely so a blocking lever surfaces in the attended queue's attention view. The convention is explicit that telemetry is the report surface, never the escalation channel — yet the lever lived only as telemetry prose, invisible to the attended lane.
3. No blocked-state signal to the operator. Standing-idle (queue genuinely empty of in-authority work) and blocked-idle (work exists, every path gated) are different states with the same observed behavior today: silent backoff. Blocked-idle should be loud.

Direction — a blocked-lever protocol as a loop-layer invariant

Amend the loop-lane convention (owner of loop-layer invariants) and the lane bodies that hold it by citation. Before a cycle may back off, it must run the lever ladder:

1. Execute every identified lever inside the lane's resolved authority (tier, rung, dimension overrides). A lever the lane names in its own report is a work item for THIS cycle, not commentary.
2. Escalate every identified lever outside its authority through the escalation contract — one tracker escalation per lever (or one consolidated item where levers share a decision), so each lands in the attended queue. Re-escalation is idempotent: an already-escalated, still-open lever is referenced, not duplicated.
3. Classify the idle: only after 1–2 does the lane enter idle backoff, recording which state it is in — standing-idle (nothing to do) vs blocked-idle (everything gated, escalations filed) — in the durable telemetry state block.
4. Signal on blocked-idle entry: transitioning into blocked-idle fires a proactive operator notification (the harness push-notification surface where available; the escalation items themselves are the durable fallback), so an operator learns within minutes, not hours, that the lane is fully gated on human decisions.

Fail-closed remains: nothing here widens authority. The protocol changes what a lane DOES with the levers it already recognizes — act, escalate, then rest.

Relations

• Loop-lane convention (docs/conventions/loop-lane/README.md) §2 escalation contract + §4 stop shapes (drain's terminal state is the analogous "all-gated" recognition; standing mode needs the same recognition without the stop), loops/skills: self-paced /loop idles between non-empty cycles — cadence should be work-state-driven (no delay when queue non-empty), not a guessed clock #500 (work-state-driven cadence — same family: pacing/behavior driven by queue state, not clock), source-control(babysit-prs): autopilot direct-gate route is inert under enabled tier — no actor runs the approver pass on zero-blocker PRs #738 (autopilot direct-gate route inert — same shape: capability present, no actor runs it), bug(source-control/babysit-prs): readiness gate resolves repo from CWD, silently scores wrong PR cross-repo #1086 (babysit readiness cross-repo defect), umbrella: SDLC lifecycle stage-gate contract — typed human/agent checkpoints, stage-state metadata on the tracker seam, forward-momentum pickup (Boris L2→L3) #1212 (lifecycle umbrella — forward-momentum invariant: an agent should always be doing what it is authorized to do), Lane telemetry: babysit-loop #1214-titled lane telemetry issues (where the state-block field lands).

[kyle-sexton]

This was generated by AI during the /work-items:work-loop autonomous drain cycle.

Surfaced in the --autonomous frontier (unassigned, unblocked, not human-gated) but carries no role label — already triaged (priority set) yet never routed to agent-ready or needs-human. Content reads as structural/cross-cutting (convention, contract, or design-space work) rather than a scoped bug fix, so this cycle's admission gate classifies it C4-shaped or unclassified and routes it human-gated per the fail-closed default rather than admitting it to autonomous dispatch.

Routing: added needs-human. No other state changed — priority and existing labels are left as-is. If this is actually scoped/delegable, re-label to agent-ready (+ status: ready) and it will be picked up by a later drain cycle.

[kyle-sexton]

Frontier disposition (drain session)

Ruling: Close as not planned for this drain wave (frontier consensus).

Dual-panel drain ruling: advisory, superseded, folded into a program umbrella, or blocked on a host/design carrier without an actionable in-repo slice this session. Re-open with a concrete acceptance brief if a consumer needs the work.

KEEP set for this wave: standing telemetry; host-physical Class C (#631 family); program umbrellas #530/#531/#532/#535; live design carriers #1500/#1553/#1668/#1695/#1716/#1681/#1718; #1360.