Bottleneck map loop

Use this loop to answer one operating question before dispatch or recovery: what is limiting us right now?

The loop has two inputs. The issue side is read-only on stdout:

python tools/issue_triage.py --json

The fleet side has one sharp edge: tools/fleet_bottleneck.py report|json refreshes tools/_registry/BOTTLENECKS.txt, bottlenecks.json, and fleet_bottleneck.prom. That is fine when the tools lane or control-pane loop owns the refresh. If tools is held by another worker, collect the same evidence without writing artifacts:

PowerShell form:

@'
import importlib.util, json
from pathlib import Path
p = Path("tools/fleet_bottleneck.py")
spec = importlib.util.spec_from_file_location("fleet_bottleneck", p)
m = importlib.util.module_from_spec(spec)
spec.loader.exec_module(m)
snap = m.collect(audit=False)
ranked = m.rank_bottlenecks(snap)
r = snap.get("registry") or {}
print(json.dumps({
    "generated_utc": snap.get("generated_utc"),
    "health": ranked.get("health"),
    "headline": ranked.get("headline"),
    "counts": r.get("counts"),
    "top_bottlenecks": ranked.get("bottlenecks", [])[:8],
}, indent=2, default=str))
'@ | python -

Do not mutate GitHub from this loop. Label, close, assign, and comment actions belong to /issue-triage after operator approval.

Cadence

Run the loop:

• at the start of every dispatch window;
• after a fleet stall, rate-limit event, or auth failure;
• after a large issue-filing burst;
• before declaring the operating process healthy.

Decision Order

Classify before choosing an action:

Field	Values	Meaning
Evidence scope	public, operator-private	Public evidence can be tracked or filed publicly: GitHub issue numbers/titles/labels, public docs, commit refs. Operator-private evidence includes account names/dirs, auth state, session IDs, transcript details, reset times, private host paths, and token/cost details; public notes should aggregate or redact it.
Horizon	transient, semi-durable, structural	Rate limits, auth failures, temporary API errors, and at-cap pressure are transient unless they repeat. Watchdog absence, auto-resume backlog, stale telemetry, and surfacing backlog are semi-durable. Missing labels, orphan P0/P1, broken routing, and missing ownership are structural.
Weight	dispatch-gating, process-debt, strategic	Transient CRITICAL/HIGH rows gate dispatch now but should be rechecked before becoming strategy. Semi-durable rows become process debt if they persist. Structural rows carry strategic weight because time will not clear them.

Keep two decisions separate:

• Can dispatch safely continue now? Transient account/auth/rate-limit pressure can answer no.
• What durable loop should improve next? Structural ownership/taxonomy debt can remain the top durable problem even while dispatch is temporarily gated.

Then use the first true condition:

Evidence	Bottleneck	Next process
Fleet health is CRITICAL or HIGH from transient account pressure	Capacity gate	Pause/cap dispatch; recheck after reset/relogin before treating it as strategic debt.
Fleet health is CRITICAL or HIGH from recovery/telemetry rows	Semi-durable recovery debt	Fix watchdog, auto-resume, and surfacing backlog before broad dispatch.
Issue actions include mechanical stale/dormant work	Backlog gardening	Ask the operator to approve the /issue-triage action manifest.
Orphan P0/P1 count is nonzero	Ownership	Claim or explicitly defer top P0/P1 rows before launching more workers.
needs_priority or needs_area dominates	Taxonomy debt	Run a focused /issue-triage pass for priority or area labels.
Fleet is below HIGH and top issues are owned	Execution	Dispatch one issue per worker with issue-bound commits and witness gates.
The same blocker repeats across passes	Process debt	File or update one small process issue naming the missing guard.

Do not recommend "more workers" while the fleet is CRITICAL/HIGH from account, auth, watchdog, auto-resume, or surfacing failures. More workers are only useful after the constrained operating loop is healthy enough to absorb them. Also do not let a transitory rate-limit reset hide durable backlog debt: record both the current gate and the structural problem that will remain when the reset clears.

Evidence Note

When the operator asks for a map, write a tracked dated note:

docs/notes/BOTTLENECK-MAP-YYYY-MM-DD.md

Record:

• exact commands and UTC timestamps;
• the top system bottlenecks and fixes from fleet_bottleneck.py;
• issue counts, action-manifest shape, and top P0/P1 rows from issue_triage.py;
• a scope/horizon/weight row that says which evidence is public vs operator-private and which bottlenecks are transient vs structural;
• the next durable process to run;
• concrete next-pass done conditions.

Standing Hook

The permanent control-pane hook belongs in tools/control_pane.loops.json and must wait for the tools lane to be safe:

python tools/fleet_control_pane.py loop-add bottleneck-map --scope repo --status-cmd "<read-only status command>" --apply

The hook should report ACTION when fleet health is CRITICAL/HIGH, orphan P0/P1 is nonzero, or label-debt counts regress above the last baseline. It should report PASS only when counts are flat or improved and fleet health is below HIGH.