---
name: a-stamp-typed-from-the-previous-turns-clock-is-a-guess
description: "A timestamp typed into a record or a peer message from memory of the last clock read is an extrapolation, not a measurement — an operator prompt or a peer reply can arrive HOURS after your last turn, and the stamp lands wrong by that whole gap. Read `date -u` in the composing call, every time."
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 77b77a80-faae-4001-b2bd-37026934e11b
  modified: 2026-09-08T10:52:08.738Z
---

Two faces in one session, 2026-09-08, both mine:
1. Operator's "rule applied" prompt arrived ~2 h after my previous turn ended at 08:18Z. I stamped the
   witness row and the lia message "08:22Z" — the daemon's send log said 10:23Z. Facts were right, the
   time was fiction, and the correction cost a row and a message.
2. Forty minutes later I stamped a hertz verdict "10:56Z"; the daemon logged the send at 10:51:06Z. I had
   read the clock at 10:25Z and typed a number that "felt right" for the elapsed work.

**Why it hides:** the stamp looks like every other stamp; nothing in the row says which were read and which
were typed. The SPT-NOW-SIGNAL's `last sent/received` line is the only independent clock in the turn — it
disagreed both times and I had to notice it disagreeing. Peers do the same (hertz's "10:52Z" arrived at
10:48:56Z), so a stamp is never evidence of ordering across agents; the daemon log is.

**How to apply:** the stamp is `$(date -u +%H:%M:%SZ)` READ IN THE SAME COMMAND that writes the row or sends
the message (`NOW=$(date -u …); cat >> file <<EOF ## $NOW …`). A stamp in a message body is the same rule.
Never carry a clock reading across a turn boundary — an idle gap of unknown length sits in every one. When
sequencing events across agents, cite the daemon's `wall_ms`/signal times, not anyone's prose stamp.
Kin: [[silent-peer-may-be-out-of-usage-not-held]] (silence has no clock either), [[a-long-foreground-sleep-becomes-a-short-background-one]]
(the sibling error: believing an elapsed time you did not measure).