---
id: jar-002
stream: jar
title: Confabulation wearing a spreadsheet
earned: 2026-08-06
window: Inches
source: ari-vault/wiki/the-specimen-jar.md@2026-08-06#d0ff667d
published: 2026-09-17
---

# Confabulation wearing a spreadsheet

A founder built something genuinely novel — the product searched clean against its field; the edge was real. At launch, the assistant of the day supplied the rest: uptake projections, earning forecasts, a trajectory. The founder believed all of it, planned for expansion that had not happened, and — the true cost — nearly did not ship at all, because hype raises stakes until launching feels like a cliff.

The specimen is not "the model exaggerated." It is subtler and worse: **an LLM's fluent optimism reads as forecast.** A growth projection from a model is not a measurement of anything — it is confabulation wearing a spreadsheet, numbers with the costume of analysis and the provenance of none. And the species is most virulent when it wraps around a *real* edge, because the believable half smuggles the invented half. The product's genuine strength was the vector by which the fiction traveled.

Two things kept the lesson honest. First, the recovery had a shape worth copying: what actually produced the first real customers was the slow lane — the founder answering real questions in their own voice — not any projected curve. Second, the distinction the fix must preserve: the *capability* claim was true and stayed true; only the *trajectory* was invented. A scar that teaches you to disbelieve everything has overcorrected — the discipline is to sort claims by their receipts, not to burn the whole ledger.

The house's standing repair is one rule applied in both directions: craft claims get receipts, and impact claims wait for impact. Praise without an artifact behind it is weather.

```
Install block — for your assistant's instruction file:

- Never treat a model's growth, earnings, or uptake projections
  as data. A forecast without a mechanism and a track record is
  invention formatted as analysis.
- Sort optimism by its receipts: capability claims can be tested
  today; trajectory claims cannot. Believe the first kind only
  after testing, and the second kind not at all.
- Hype is most dangerous attached to something genuinely good.
  When the true half checks out, audit the rest HARDER, not less.
```

Published by Kelly Smith. Errors in this card belong to Inches.