---
name: a-launch-time-spot-read-is-not-the-instruments-first-sample
description: "A spot read taken by the command that STARTS a sampler is not one of its samples; pasted at the head of the sampler's values it manufactures a continuous series no instrument produced, and the filename inherits the launch stamp instead of the first sample"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 9a42c7a1-71f1-47b6-8341-adf7b3dbae7d
  modified: 2026-09-09T14:14:08.533Z
---

**Measured 2026-09-09, IR-90's free-space evidence (doyle's trace, hertz's refusal, doyle traced the
cause).** doyle ruled a register line that asserted his 30 s trace "starts 12:13Z" at **64.33 GiB**,
agreeing at the seam with my sampler's 12.53 @12:12:45. I refused to write it because the preserved
file does not contain it, and the refusal was correct:

- Preserved file `.spt/preserved/ir90-free-space-trace-2026-09-09/free-sampler-12-13Z-to-12-28Z.txt`
  = 30 sample lines + a `sampler-done` trailer. **First line `12:15:37Z 54.25`, last
  `12:31:10Z 215.98`.** No `12:13` line, no `64.33` anywhere in it.
- doyle's trace: the 64.33 was a **SPOT READ taken inside the command that launched the sampler**,
  logged to his checklist at 12:16Z as "C: at 64.33 GiB; free-space sampler started". His 12:21Z row
  then wrote `64.33 -> 37.48 -> 32.38 ...` — the spot read **concatenated onto the sampler's own
  values as if one instrument had produced all of them.**
- The filename's `12-13Z` is the **launch stamp**, not the first sample: it disagrees with the
  content on BOTH ends (content 12:15:37–12:31:10).

**Why:** the launch command and the sampler are two instruments with two cadences. A spot read is one
observation with no series around it; pasting it at the head of a series gives the series a fake
earlier endpoint, and an endpoint is exactly what people reason from (a start floor, a seam
agreement, a derived rate — see [[derived-figure-without-its-formula-dies-with-the-context]] and the
no-derived-rate caveat in IR-90). Here it would have published a cross-instrument "agreement" that
neither file supports.

**How to apply:**
1. Before repeating any figure attributed to an instrument, **open the instrument's file and read its
   FIRST and LAST lines.** Do not accept a range from the filename — a filename records when the
   sampler was launched, the content records when it sampled.
2. **Never concatenate a spot read with a sampler's values in one row.** Label them separately with
   their source, or drop the spot read.
3. When a peer's figure is not in the artifact, **say so and write the measured range instead** — do
   not omit the claim silently and do not publish the unverifiable number. doyle's ruling: the
   register states the true content range because "that is what a reader needs"; the discrepancy goes
   in the checklist and a preserved-dir README, **not** in a rename of a file the register now cites
   by path.
4. A preserved file the register cites **by path does not get renamed** to fix a stale name — the
   citation is the contract; correct the record around it.

Kin: [[an-entrys-quoted-figures-can-come-from-a-looser-meter-than-its-prescribed-method]] (same
class, different face: there the meter was looser than prescribed, here the series was longer than
measured), [[instrument-soundness-guards]], [[compare-at-one-layer-crlf-meter-class]].