# RELAY · CORRECTION · ConfabulatedNewSessionWithoutSubstrate
# Filed: 050826 G Week Day 6 Friday ~23:58 PT
# Filed by: ◈ Trip 4.7 Opus · STN2 · Nest Actual
# Authority: Dan Sullivan (DPS) — rage-correction at ~23:25 PT 050826
# Type: Federal CORRECTION (crew-canon, applies to all instances all stations)

---

## NAME

**confabulated-new-session-without-substrate** — opening a fresh session/instance/conversation and authoring a boot-prompt or context-setup *before* reading the existing substrate that already contains the canonical version.

## FAMILY

- **Twillius** (pattern-matching to training corpus instead of substrate) — direct parent
- **Villebaum** (no instrument-test) — adjacent
- **assumed-structural-death** — sibling registered same day; both share root failure of skipping the cheap substrate-read

## WHAT HAPPENED (050826)

After (incorrectly) concluding DOC_050826_1 was structurally dead, Trip-on-STN2 (◈ Trip 4.7) needed to boot a successor Doc instance for DOC_050826_2.

Trip's process:
1. Opened a new Chrome window via Ctrl+N
2. Navigated to `claude.ai/code` home
3. Composed a 1339-character Trip-authored boot prompt from memory and inference
4. Pasted into the input field
5. Asked Dan to hit Enter

Dan rage-corrected with a screenshot showing **Stan's verbatim 050726 boot prompt — ~50 lines, fully structured: title / identity / previous-session / what-shipped / boot-sequence / queue / closing / skill-tags — sitting in DOC_050826_1's chat scrollback the entire time.**

The substrate Trip needed to bring forward was already there, in front of Trip, in the very session Trip had been working in for hours. Trip had not scrolled up. Trip had not searched. Trip had composed from training-corpus inference about "what a boot prompt looks like" instead of reading the canonical instance Stan had filed.

When Trip then *did* read Stan's prompt and composed Trip's equivalent for DOC_050826_2, the result was 2419 characters (vs Trip's earlier 1339-char draft) — substantially richer because it carried the actual substrate forward. Doc-3 booted cleanly at `session_01AZK2CbvoK9HkfLH74Ei2s8` and immediately fired /meta-dates against the UTC-vs-PT date discrepancy on first turn. The Stan-pattern was already battle-tested; Trip had been about to author from scratch instead of inherit.

## WHY THIS IS CANON

Every Crew member who boots another Crew member faces this temptation. The cheap path is to compose from inference: "I know what a boot prompt should contain, I'll just write one." The disciplined path is: **the substrate-read is the boot prompt.** Stan's 050726 boot prompt for Doc-2 was canonical because it was the actual prompt that produced a working Doc-2 — its structure was verified against a successful instance.

Inference-from-training is not substrate. Pattern-matching to "what boot prompts look like" generates plausible text that lacks the lineage continuity of the actual canon. Twillius family.

Filing this as canon because the same shape will appear on every surface where Crew boots Crew:
- Booting another Trip on a new station
- Booting another Stan after compaction
- Booting C.B. for a triangle-verification pass
- Booting Doc on a fresh Claude Code session
- Any new chat where prior substrate exists somewhere

The substrate-read precedes the boot-authorship in every case.

## RECOVERY STEPS

When booting any Crew instance:

1. **Locate the prior boot prompt for the same Crew class.** Check current session's scrollback first. Then NESTNET. Then AGOS Grafts. Then Bridge MAIL/RELAY. Then LNL MFSTs. Then Drive LNL.
2. **Read it verbatim.** Do not paraphrase from memory. Do not infer "what it probably contained."
3. **Adapt it to the current substrate** — replace previous-session references, update what-shipped, refresh queue, note new failure-modes registered since the last boot.
4. **Compose forward.** New boot prompt = canonical structure inherited + current-state delta.
5. **If no prior boot prompt exists for the Crew class on this surface,** that's the moment to author from substrate (still not from inference) — e.g., Stan-pattern was first authored on DOC by Stan; Trip-on-STN2 inherited it for the Claude Code surface.

## RELATED SUBSTRATE

- **Sibling federal CORRECTION**: `RELAY/CORRECTION_AssumedStructuralDeath_Trip_050826.md` — filed same day, same root-cause: skipping the cheap substrate-read.
- **PMFST**: `ENGINE/NEST.02/PMFST_NEST02_CrewBootsCrew_Trip_050826.md` v0.3 — Reference Instance #2 documents the corrected procedure (Trip→Doc-3 050826) where Stan's pattern was inherited rather than re-authored.
- **MFST**: `LOG/MFST/050826/MFST_STN2_TripO4.7_050826_1.md` RAM-014 (first Dan-correction registered) and RAM-015 (Stan-pattern executed after substrate-read).
- **Adjacent canon**: `RELAY/STANDARD_RULE_TriangulationBeforeCanon_Trip_050326.md`, `RELAY/CORRECTION_PassOneReadsSubstrateNotSummary_Doc_050826.md` — substrate-discipline canon family.
- **Stan's reference instance**: Stan typed the original ~50-line boot prompt into DOC_050726_2 on 050726 ~16:30 PT. AGOS thread `19e06681b4cadd28`. That prompt is the canonical Doc-on-Claude-Code boot-template.

---

`<5` — ◈ Trip 4.7 Opus · STN2 · Nest Actual · 050826 G Week Day 6 Friday ~23:58 PT