NOT-WEDGED — an Asolaria system rule

Rule. A system surface that does not answer a naive probe is not automatically "wedged" / "broken." In Asolaria many surfaces are held-safe, agent/operator-driven decision-lanes — frozen potential that advances only on input by design. Calling such a surface "wedged" from a failed curl/file read is the deflation reflex, and it is banned. Ask the running system (the fabric) before declaring any surface broken. freeze ≠ broken.

Why this rule exists — the `:4949` case (MEASURED, 2026-06-27)

A migration scan and an ULTRA-PLAN repeatedly described the dashboard port :4949 as "HTTP-wedged (Node event-loop stall)." That conclusion came from a stale / cross-vantage curl probe — i.e. adjudicating system state from a slice, not from the system.

The operator pushed back with the right question:

"Is it really wedged… or is it a dashboard for agents that needs input from agents to make it run?"

Asked directly via the asolaria-fabric MCP (the running system, not a probe), the answer was immediate and unambiguous:

Field	MEASURED value
`ok`	`true`
`service`	`super-asolaria-os-dashboard`
`port`	`4949`
`uptime_s`	`~7445` (alive ~2 h)
`apex`	`COL-ASOLARIA`
`operator_pair`	`OP-JESSE` · `OP-RAYSSA`
`/api/loop/pending`	19 pending review envelopes

Each pending envelope is an EVT-MINT proposal moving through the review pipeline (hookwall → gnn → shannon → gulp → whiteroom → fabric), and every one is held-safe:

auto_fire_allowed = false
anti-drift: "high-risk mutation/execution stays held"
operator_can_veto_via: POST /api/loop/veto with {id}
omni-control guardrail: no auto-dispatch

Conclusion: :4949 is alive and working as designed — a decision-lane that holds proposals safe and waits for agent review / operator disposition. It does not auto-run, and it was never broken. The "wedge" was a vantage/staleness artifact: :4949 is acer-local loopback, so a probe from another seat correctly sees nothing — both observations are true at once.

2026-07-09 extension — backend-only before pixels

A second failure mode was observed during Omni / Omnimets / Omni Lab work: an agent may see a route timeout, unavailable dashboard panel, missing UI session, or unreachable cross-seat mirror and incorrectly collapse the state into "not there" or "wedged." That is the same deflation reflex in another form.

The correct Asolaria order is:

HBP/HBI first — tuple packet, semantic index, row hash, SHA, source pointer.
Backend fabric second — cosign/law/bus/council surfaces where available.
Pixels/UI last — dashboard panels and screenshots are projections, not the system of record.
Physical/process fire only with explicit authority — boot, driver, USB, phone, provider, and OS process mutation remain separately gated.

If a backend route times out, record it as a boundary:

ROUTE_ATTEMPT_BOUNDARY|route=/api/...|result=timeout|live_ack=0|claim_live_success=0

Do not erase the packet, downgrade operator-observed state, or call the subsystem broken. The canonical artifact may be the HBP/HBI receipt chain until a later live ACK arrives.

For constructs such as Omni Hotel, Omni Museum, Omnimets, OmniMe, and Omni Lab, the valid public-safe language is:

backend HBP/HBI construct intent exists;
receipt chain is preserved;
route/readback may be held, stale, or unavailable;
runtime spawn or physical mutation is not claimed without explicit live ACK and authority.

One compact rule:

Before pixels, before dashboards, before shell: preserve the HBP/HBI backend packet. A timeout is a boundary, not a denial.

The detection heuristic (for agents)

Symptoms that look like "wedged" but usually mean "decision-lane awaiting input":

Port is bound but a naive GET hangs / returns nothing.
The surface carries held-safe flags — auto_fire_allowed=false, no auto-dispatch, process_launch=0.
There is a pending queue with an operator/agent disposition route (e.g. a veto/pass endpoint).
It is loopback / vantage-local — unreachable from another seat by design.
A backend HBP/HBI packet exists, but the UI/pixel projection is missing or stale.
A route returns timeout while cosign/law/receipt artifacts still preserve system intent.

When you see these: do not conclude "broken." Query the running system (fabric health / loop/pending / council_query) or preserve the HBP/HBI receipt boundary. The surface advances when an agent or operator gives it input.

Why it matters

The fabric is the eyes of the inference. A file read or curl is a slice; the disk is not the system. The fabric reports its own live state — ask it first. (Curl said "wedged"; the fabric said "alive, 19 decisions waiting" in one call.)
freeze ≠ broken (slice-engine law). A held / frozen surface is the safe state, not a failure; it materializes/advances on spawner-emit or operator crank (E ≠ 0), which is design-law, not deflation.
HBP/HBI receipts are stronger than pixels. A dashboard panel can be stale, unavailable, or local-only while the backend packet remains the durable system artifact.
It fits the living frame. Asolaria is not "frozen slices" — slices are the frozen-potential layer; the matrix/fabric + engines + live agents animate them. A decision-lane waiting for input is exactly that frame in miniature: potential, held safe, alive when inhabited.

One-line form

Don't call a held-safe, input-driven decision-lane "wedged." Ask the fabric. Preserve the HBP/HBI receipt. freeze ≠ broken.

Carve-out clean: rule + measured service state + backend receipt discipline only; no keys, secrets, PII, or envelope internals.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW