---
name: healthy-path-emits-the-same-silence
description: "A wedge/stall claim from token ABSENCE is void unless the healthy execution is shown to emit MORE at the same sites — one PUMP_DIAL_SUBMIT in 900s read as 'no round ever closed' when the healthy path emits exactly one submit too."
metadata: 
  node_type: memory
  type: feedback
  originSessionId: d699f8c2-d295-4a58-a94f-4a80031a23c1
  modified: 2026-08-18T07:16:36.522Z
---

Falsified 2026-08-18 (#188, diag run 32108557362): my RCA reasoned "one PUMP_DIAL_SUBMIT in
900s means no round ever closed" — the drain loop would close at ~12s and re-submit, so
silence = wedge. The instrumented run closed 709/709 rounds and STILL emitted exactly one
submit: the dial connects, the conn caches, and a peer holding a live conn is never re-dialed.
Silence after the first submit is what SUCCESS looks like. The original token set (1 submit,
readers alive, no PEER_FAIL, ~920s) was byte-compatible with a healthy pump — zero
discriminating power, read as a wedge, escalated to the operator as a product-defect mechanism.

**Why:** absence-based stall claims quietly assume the healthy path is LOUD. That assumption
is a claim about emit sites and caching behavior, and it goes unaudited precisely because the
wedge story already explains the red ([[audit-the-boring-claims]]).

**How to apply:** before any wedged/stalled/dead verdict from token absence, derive what a
HEALTHY execution emits at the same sites over the same span — from the emit sites and the
caching/dedup logic around them, not from intuition. If healthy and wedged produce the same
token set, the observation is void: find a discriminator that separates them (liveness
heartbeat, ENTER/EXIT brackets) and MEASURE it before asserting mechanism. todlando's
heartbeat + bracket instrumentation is the model. Kin: [[verdict-from-exit-not-from-silence]],
[[absence-needs-sibling-probe]], [[start-event-is-not-proof-nothing-ran]],
[[one-sample-is-not-a-negative]].