← Documentation / SQL Reference

DOI 1.19 — Four-model video evidence supplement for two discreteness claims

View DOI 1.19 on Zenodo: https://doi.org/10.5281/zenodo.22252098

New DBBuilt on the current, join-free pipeline (build_notta_tta_comp_db.py), one independent database per model.

This is a supplement, not a correction — no value in DOI 1.16 or DOI 1.18 changes. Both reports measure a quantity that depends only on the shared ground-truth images (σ_gt for DOI 1.16, μ_gt for DOI 1.18) and state that its discreteness (see 8-bit pixel quantization) must therefore be identical across all four models — but neither report backed that specific claim with this series' usual screen-recorded, hash-verified evidence. DOI 1.16's text stated it as an assumption from the GT-only dependency argument, with no cross-check at all. DOI 1.18's text claimed an improvement (an actual independent Standard/SwinIR comparison), but that check covered two of the four models rather than all four — confirmed directly here: DOI 1.18's own verification_sql_final_4model.sql (the script behind its recorded video) includes the μ_gt discreteness check only once, under Standard's .open section.

This report closes both gaps: the identical query for each quantity, run against all four models in one continuous screen recording per quantity.

σ_gt (DOI 1.16) — sigma_gt_discreteness_check_4model.sql

WITH pop AS (
    SELECT sigma_gt FROM blocks WHERE sigma_gt > 1e-6
)
SELECT 'n_total_blocks' AS metric, CAST(COUNT(*) AS VARCHAR) AS value FROM pop
UNION ALL SELECT 'n_distinct_sigma_gt_values', CAST(COUNT(DISTINCT sigma_gt) AS VARCHAR) FROM pop
UNION ALL SELECT 'avg_blocks_per_distinct_value', CAST(ROUND(COUNT(*)*1.0/COUNT(DISTINCT sigma_gt),2) AS VARCHAR) FROM pop;

-- Repeats identically for SwinIR, High-Fidelity, and Digital-Art,
-- including the top-10 most-frequent-value and bucket-breakdown
-- queries already published in DOI 1.16.

Identical across all four models, confirmed directly rather than by assumption: 79,321,389 blocks, 2,308,415 distinct σ_gt values, average 34.36 blocks per value. The top-10 most-frequent exact values (topping out at 0.0012969369171885248, shared by 53,587 blocks) and the full bucket breakdown by repeat count (1 (unique): 1,027,071 distinct values / 1,027,071 blocks; 2–10: 688,506 / 2,748,884; 11–100: 392,839 / 15,164,532; 101–1000: 194,421 / 48,961,359; 1000+: 5,578 / 11,419,543) are also byte-for-byte identical across Standard, SwinIR, Upscayl High-Fidelity, and Digital-Art.

μ_gt (DOI 1.18) — mu_gt_discreteness_check_4model.sql

WITH pop AS (
    SELECT mu_gt FROM blocks WHERE mu_gt > 1e-6
)
SELECT 'n_total_blocks' AS metric, CAST(COUNT(*) AS VARCHAR) AS value FROM pop
UNION ALL SELECT 'n_distinct_mu_gt_values', CAST(COUNT(DISTINCT mu_gt) AS VARCHAR) FROM pop
UNION ALL SELECT 'avg_blocks_per_distinct_value', CAST(ROUND(COUNT(*)*1.0/COUNT(DISTINCT mu_gt),2) AS VARCHAR) FROM pop;

-- Repeats identically for SwinIR, High-Fidelity, and Digital-Art,
-- including the top-10 most-frequent-value query already published
-- in DOI 1.18.

Identical across all four models, confirmed directly (previously only checked for Standard and SwinIR, and not on video): 80,457,746 blocks, 68,649 distinct μ_gt values, average 1,172.02 blocks per value. The top-10 most-frequent exact values (topping out at 1/255=0.00392156862745098, shared by 229,846 blocks) are also byte-for-byte identical across all four models.

Both results are exactly what σ_gt's and μ_gt's definitions predict: neither quantity ever depends on any model's reconstruction, so no cross-model difference was possible in principle. This report's contribution is closing the gap between that principle and the series' own evidentiary standard, not a new empirical finding.