# Build handoff: T473, Scaling dimensions of the 3D Ising CFT

Draft: https://numberdb.org/T473
Canonical slug: Scaling_dimensions_of_the_3D_Ising_CFT
Family: numberdb-data#201
Build stamp: 20261025T050243Z-build

The draft is filled and remains unpublished and unoffered. The separate critique and repair stages, followed by the controller's offer action, are still pending. No published table was edited.

## Contents and choices

| Primary | Stored dimension |
|---|---|
| sigma | 0.518148806 +/- 0.000000024 |
| epsilon | 1.41262528 +/- 0.00000029 |
| T | 3 exactly |
| epsilon-prime | 3.82951 +/- 0.00061 |
| sigma-prime | 5.262 +/- 0.089 |

These are one named quantity, the scaling dimension, indexed by primary operator. All five are spatial-parity even; scalar primes denote the next primary in the same spin-flip sector, never derivatives. The scalar selection is the first two nonidentity primaries per spin-flip sector, plus the stress tensor. No artificial numerical grid or unsupported precision was introduced.

The two leading dimensions come from Chang et al. (2025), Table 1. The primed dimensions use the conditional bounds in Reehorst (2022), equations (4.1), (4.2), with each distinct spectral-gap assumption stated. This choice prefers a controlled conditional interval over the narrower nonrigorous extremal-functional estimates. Those estimates are checked for compatibility, not silently substituted or averaged. The table is measured, with T=3 identified as exact. The 100-digit general aim is impossible for these published determinations; their own uncertainties are retained.

Existing tags statistical mechanics and physics apply. T154 and T155 links use slugs read from search results. T472 is a related draft and is addressed by its supported stable T-number HREF, not a guessed slug. T472 was read through the API and its seven Ising exponents checked; it was not edited. The critique/controller should consider the publication dependency if T472 remains unpublished when T473 is offered.

## Validation actually run

- Cheap screening: already_here and already_asked, valid api/lookup ball searches at four expected values, both existing tags. No matching family found; incidental numerical matches were unrelated.
- Shared dry_run.py and check.py: exactness, ring imports, prose checks and size. Five entries; longest value 27 characters; annotated block about 0.77 KiB.
- Source transcriptions: four parenthesised uncertainties independently parsed into exact rational endpoints and matched to all four generated intervals.
- Known exact 2D Ising exponents and an intentionally inconsistent Fisher-relation control; interval arithmetic controls; float-rejection control.
- Four scalar unitarity bounds, relevant/irrelevant assignments and exact stress tensor.
- Chang Table 1 to Table 2 consistency for eta and nu.
- Five scaling relations, both with rational intervals and exactly at every corner of the two-dimensional source error box.
- Independent finite-size Monte Carlo comparison for nu, eta and omega, using Hasenbusch (2010).
- Separate EFM calculation for both primed dimensions, using Simmons-Duffin (2017); this is a different extraction within the bootstrap framework, with nonrigorous error bars.
- All seven converted Ising exponents overlap T472, with a common center-derived point in all seven stored intervals. This shares input sources and is a consistency check.
- Read-back verify(sample=None): 5/5 matched, 0 differing, 0 missing, 0 extra. All numerical checks repeated on API data.
- Site audit: clean=true, findings=[].

No conformal-bootstrap optimisation, Monte Carlo simulation, or fresh physical measurement was run. These are reproducible transcription and validation checks, not an independent 100-digit computation of the CFT.

## Files and reproducibility

The website draft has generate.py, checks.py, the unchanged shared check.py and dry_run.py helpers, source-evidence.json, convention.md, and the successful validation logs attached. Run `sage -python checks.py` for local checks and `sage -python generate.py` / `sage -python checks.py --stored` with a draft-readable NUMBERDB_API_KEY for read-back checks. In this environment, all Sage invocations used agents/sage.sh; credentials were piped through stdin.

Downloaded source PDFs, extracted text, source hashes, all scripts, initial and final numerical evidence, the draft documents and API results are retained in this run's work directory. The initial read-back type check rejected the exact wire string "3"; normalization now decodes exact integer spellings before calling the Sage-object checker. The table's stored numbers were not changed. The initial attachment attempt used a nonexistent top-level client function; the existing attachment API then succeeded.

Lessons are in lessons-proposed.md; deployment/tooling notes are in environment-notes.md. No source, prompt, skill, or published table was edited; no commit, push, deployment, review, or offer action was performed.