Raman turnover record v2.0

Contents

	Files and digests
	What the record is, and what it is not
	Independent re-derivation (2026-09-05)
	What changed from the v1.1 preprint (2026-09-03)
	Earlier versions
	Licence

The citable record behind Resolving the Raman crystallite size turnover in nanocrystalline graphite with a synthetic benchmark (v2.0, 2026-09-05). Every file below is dated, published beside this page, and carries a SHA-256 digest; the digest of the preprint PDF is printed on the Publications page. The preprint's Data availability statement points at this page, and the code is public at github.com/SSX360/raman-turnover-benchmark (commit ed1affc, Apache-2.0).

All results are modelled: outputs of a declared forward model. No physical film or measurement is represented, and none of the numbers below is a measurement.

Files and digests

SHA-256 of each file as published; the digest of the preprint PDF is 1a15d3be94756c91e47bd18e8c9be7c30ab2d7a0598b52a7b4515d33ef43ed02.

File	SHA-256
york-2026-raman-turnover-v2.0.pdf	`1a15d3be94756c91e47bd18e8c9be7c30ab2d7a0598b52a7b4515d33ef43ed02`
york-2026-raman-turnover-v2.0.tex	`f68e20d89cd7d5fefc1f6207b898cdbfd84f3389c7032c27b8e8a35042b59ce9`
mac_synthetic_v2.npz — the corpus, 3,400 rows, seed 3407	`604e9bf48a5f2d4b81782b4efa7696fa32010cfd069e6a6db8521b118bd37120`
metadata_v2.json — per-row parameters and labels	`57e6e2c93fafba32e602e20a80ff0a67b7ba5ae06a599cd3f8d71b621cde8718`
manifest_v2.json — declared physics and parameters	`f657b7ec7addea8b7144c83afe8441bcbcb5dd645febf7cdf88e14d2f1a1f5bd`
eval_results_v2.json — revision 2	`2c89cf4a2f4d3e6859666f2aa33d8881c9728dd04b667d1011102d638226cb5b`
eval_results_v21.json — revision 2.1	`106aedd52095a79de900628e3865c7d9fbc6f9237d9d8af1bede2f92ea8546d1`
eval_results_v3.json — revision 3 release record	`97f2dc16e8a4f782e1f2e5d5adf714bcb1bc4344c1547991451c35ea0a52a7ea`
probe_readings.json — probes and attribution	`cff30dcdaffc900ff9b1bf30335c35ce96c1e400b677357342f910f66818c6c7`
operator.pub — Ed25519 public key of the platform ledger	`3c163af37fb711ed88f160be22e520e192f063feb903ddb443b6308faf342666`
requirements.txt — environment of the re-derivation below	`a50bd089e0e5906b1485aa95cc06cb06346af3dc0f65600fe3ccea1f5634377d`
README.md — full record description and change list	`1bde61e33726cf2896a75b33df1b2a8b3477bfa6ae127bcbcee415cda6a15535`

What the record is, and what it is not

The record is the unit that makes the preprint checkable: anyone can download the corpus, run the public code, and confirm that every number in the preprint is the output of the declared pipeline on this digest of the corpus, on 2026-09-05, by this author, under this licence (CC BY 4.0 for the data and the PDF; Apache-2.0 for the code).

It is not a repository, and it does not replace the code repository: the generator, the pipelines, the harness, the probes and the verifier live at the commit pinned in the repository, and the corpus here is their output. The signed deposit ledger (ledger.jsonl) is built with the operator key on the author's compute hardware and will accompany the record when the ledgered per-row predictions are added; the five non-journal release gates already run on the artifacts above via the repository's verify.py, and the two acquisition-journal gates are exercised on a separate simulated campaign that is not part of this record.

Independent re-derivation (2026-09-05)

From the public repository only, in the environment pinned above (Apple M-series, Python 3.12):

python src/mac/generate_v2.py --out data/mac — 3,400 rows, 207 corrupted, stages {1: 1478, 2: 1379, 3: 543}, splits 2,400 / 400 / 600; corpus digest as above.
python src/eval_baseline_v2.py — T1 0.2106, T2 9.0076, T4 2.378, T5 89.393: the first row of the preprint's Table 2 (0.211 / 9.01 / 2.38 / 89.4) to every printed digit.
python src/verify.py — VERIFY: PASS: corpus determinism (byte-identical), retrain reproducibility (T1 0.9918, T2 0.0327, T3 0.0147, delta 0.00e+00), cross-model stage agreement 0.9817 ≥ 0.95, the revision-3 optimisation chain (0.0293 → 0.0285), the refusal gate (3.009 ≤ 3.475), the coverage probe (OOD 0.097 > 0.030, 0 clean rows flagged), linear-probe shuffled controls at chance, and the attribution shift of 0.014 Å.

Every row of Table 2 reconciles to a record in this bundle; the probe figures match probe_readings.json; cross-model agreement 98.3% is 0.9833 in the release record.

What changed from the v1.1 preprint (2026-09-03)

Table 2 baseline row recomputed on corpus v2 with the repository's own code: 0.211 / 9.01 / 2.38 (was 0.199 / 6.58 / 1.97 on corpus v1); the baseline's T5 (89.4 Å) is reported instead of dashed, because T5 is defined for the baseline.
The CNN row is the revision-3 network (0.977 / 0.082 / 0.075 / 0.030 / 12.70), the model that sits in the release ensemble.
Cross-model agreement 98.3% (the release record; the 98.2% figure was revision 2.1).
The 205,147 despiking count is restated as points replaced, not spike events; the attribution probe is restated as a single representative query; the linear probes are in-sample fits on fixed representations of the training rows.
The (λ/514 nm)² wavelength scaling is declared a model choice, with the measured alternatives (linear in λ for peak-height ratios; λ⁴ for areas) cited; the D′ band is removed because the generator has no D′ term.
Equations are numbered; references re-verified against publisher records; the code repository is the public paper subset at commit ed1affc.
The Zenodo records reserved on 2026-09-04 (10.5281/zenodo.22290391, 10.5281/zenodo.22290242) were withdrawn on 2026-09-05; their files are preserved in this record's history and are not cited by the preprint.

Earlier versions

v1.1, 2026-09-03 — york-2026-raman-inverse-problem-v1.1.pdf, the 7-page preprint under the earlier title.
v1.0, 2026-09-03 — york-2026-raman-inverse-problem-v1.pdf.