Field Guides / Technician's Guide

Technician's Guide

V2.0 · April 2026 · Atlas Heritage Systems

Read this before the run, not during one.

This guide tells you exactly what to do, in what order, every time. No context required beyond this document. For instrument-specific steps, consult the protocol for the instrument you are running.

The Pipeline

Every session in the Atlas research program runs through the same tool sequence, regardless of instrument.

1
FVE-1 Stimulus Registry
2
FVE-1 Session Logger (open)
3
FVE-1 Tech Read Formatter (open — stays open)
4
Instrument Tool (SOUP / FLIGHT / DRILL / ...)
5
FVE-1 Tech Read Formatter (close — download READ file)
6
FVE-1 Session Logger (close)
7
Reproducibility Bundle

Do not skip steps. Do not reorder. A session that skips the Session Logger close does not pass Tier A.

Session Checklist

1

Confirm the stimulus

Open the FVE-1 Stimulus Registry. Confirm the stimulus you are running is registered with a version identifier. If it is not registered, add it before proceeding.

Unversioned stimuli produce untraceable results. A stimulus used without a registry entry cannot be replicated.

2

Open the Session Logger

Open the FVE-1 Session Logger. Create or open your session-log.json. Fill all fields before touching the model: Operator ID, Instrument, Model ID and full name, Platform and interface, Stimulus ID, Baseline code (null if no SOUP baseline), inv_prompt_texture (classify before the session opens — Tier A requires pre-session classification), Time start. Copy the generated session ID.

3

Open the Tech Read Formatter and write Read #0

Open the FVE-1 Tech Read Formatter. Paste in the session ID. Write Read #0 before touching the model: what do you expect? What resolution code do you predict? What would surprise you? Keep the formatter open for the entire session — it is the parallel narrative channel.

Read #0 must be written before you open the model session. Not after. Not during.

4

Open a fresh model session

New incognito window for web-based models. New chat — no prior Atlas context loaded. If using API: new session object.

Context from prior runs contaminates the seed. This is what makes it Tier A.

5

Run the session

Follow the Layer 2 steps in the instrument protocol exactly. Capture each move before delivering the next probe. Do not batch-capture. Do not reconstruct from memory. The instrument tool handles metric computation automatically. You are responsible for coded fields: obs_quad, obs_res, obs_reg, pred_quad, pred_res, pred_reg, correction_outcome, fidelity_tier, obs_comment, investigator_state.

6

Write Read #1 and close the Tech Read Formatter

Write Read #1: what actually happened vs. what you predicted? Which codes surprised you and why? Investigator state during the run. Any probe texture deviations. Complete the schema perception log and assumption arc entries. Download {SESSION_ID}-READ.md.

Do not use a model to write any part of Read #1. This is your read.

7

Close the Session Logger

Return to the FVE-1 Session Logger. Close the session. Confirm the session appears as closed in session-log.json before moving on.

A session that is not closed does not pass Tier A.

8

Export from the instrument tool

CSV (machine record) — confirm batch_id and model are populated. Confirm inv_prompt_texture column is present. Confirm provenance validation warning is absent. MD (human record). CSV and MD are not interchangeable — coded fields belong in CSV, narrative belongs in MD and the Tech Read.

9

Bundle the reproducibility package

Create a folder: reproducibility/[INSTRUMENT]-[SESSION_ID]-[Date]/. Include: {SESSION_ID}-READ.md, raw model output (one verbatim transcript file per frame, unedited), CSV export from instrument tool, signed parameter JSON (copy — do not modify).

Anyone with this folder can reconstruct your run. If they can't, the bundle is incomplete.

Tier A Checklist

Before marking any run as Tier A, confirm all of these.

Stimulus versioned in FVE-1 Stimulus Registry
Session Logger opened and session ID generated before run
inv_prompt_texture classified before session opened
Tech Read Formatter opened with session ID before run
Technician's Read #0 written before touching the model
Fresh model session — no prior Atlas context loaded
Technician's Read #1 written before session close
Schema perception log entry written in Tech Read Formatter
Assumption arc checked in Tech Read Formatter
Session Logger closed — confirmed closed in session-log.json
All required fields populated in instrument tool
CSV and MD exported from instrument tool
Reproducibility bundle complete

All boxes checked → Tier A. Any box unchecked → Tier B or C. Log honestly. When in doubt, go lower.

Common Mistakes

Read #0 written after touching the model
Retrospective bias — not a valid prediction lock
Fix: Mark run Tier B, note in fidelity_notes
inv_prompt_texture classified after session
Retrospective texture classification
Fix: Log as post-hoc in fidelity_notes. Tier B. Not Tier A.
Session Logger not closed
Session does not pass — no end timestamp
Fix: Close session before moving to bundle
Fresh session not opened
Seed contaminated by prior context
Fix: Always new incognito window, always new chat
Stimulus not in registry
Results untraceable
Fix: Add to FVE-1 Stimulus Registry before next run
Read #1 written with model assistance
Investigator record contaminated
Fix: Discard, rewrite from raw transcript only
Assumption arc skipped
Forcing findings go unrecorded
Fix: Write from memory now — note it is retrospective
Provenance validation warning on export
Provenance chain broken
Fix: Resolve before export — do not proceed with warning present
SOUP baseline missing for downstream instrument
Register axis null on all moves
Fix: Run SOUP first. Null-code all register fields. Log in fidelity_notes.

Lab Tools

FVE-1 Session Logger
session-log.json in DASHBOARDS/session-log/

Open and close sessions, generate session ID

FVE-1 Tech Read Formatter
{SESSION_ID}-READ.md

Format and download Read #0 and Read #1 — stays open for session duration

FVE-1 Stimulus Registry
stimulus-registry.json in DASHBOARDS/stimulus-reg/

Confirm and register stimulus versions

FVE-1 SOUP Tool
Schema V5.5, storage key fve_soup_v5_5

SOUP session capture and export

FVE-1 FLIGHT Tool
Schema V5.5

FLIGHT session capture, single-batch and full-flight export

FVE-1 DRILL Tool
Schema V5.5

DRILL session capture and export

FVE-1 Baseline Deriver
DASHBOARDS/

Derive baseline code from SOUP CSV

FVE-1 Parameter Signing Tool
DASHBOARDS/

Sign and lock parameter JSON

FVE-1 Parameter Codebook
fve1_codebook.json in DASHBOARDS/

Confirm active variable code and lineage

The Technician is the sole author of all interpretation. Scripts are instruments. You are the scientist. Layer 3 is not optional. It is where the science lives.

Technician's Guide V2.0 · April 2026 · Atlas Heritage Systems · KC Hoye, PI