1# PHASE 8 — Paper (2026-11-03)23## Manuscript4`paper/main.tex` → `paper/main.pdf` (latexmk, pdflatex + bibtex, style `aer`, UQO template: `newtxtext`/`newtxmath`,5title page macros). Sections: title page + abstract, introduction, hypotheses (5 formal statements), related work,6data, model, experimental design, results (pilot, development test 2024, calibration, subgroups, economics, zero-shot7and untouched 2025), ablations and falsification (incl. Stage-B paths), limitations, reproducibility and release,8conclusion, bibliography, appendices (baseline specifications, additional tables, universe). 36 pages.910## Provenance of every number11- `paper/results_macros.tex`: 3 468 macros generated by `scripts/evaluate.py` from the saved leaderboards, comparisons12 and pairwise files (`\crps<exp><model>H<h>`, `\rel…`, `\pw<exp><model>vs<ref>H<h>`, `\tpw…`, `\ppw…`, coverage macros);13 digits in macro names are spelt out (`Hsixzero` = H = 60).14- `paper/tables/*.tex` (53 files): leaderboards (scaled CRPS, WIS, QLIKE, calibration) and relative-change tables for15 pilot / main / main held-out / ablations / final / final held-out, universe strata, groups and ticker lists, baseline16 specifications (from `configs/baselines.yaml`), measured benchmark (from `results/metrics/benchmark_M3U96b.json`).17- `paper/figures/*.pdf` (27): scores by horizon, coverage, PIT histograms, forest plots, training curves, drawn only18 from result files by `src/chronos_x/eval/figures.py` (Times/STIX, monochrome + one accent, panel labels).19- Hand-typed numbers in the prose (subgroups, ablation ranges, copula scores, economic results) are copied from20 `results/metrics/*/pairwise_*.csv`, `subgroups_*.csv`, `stress_calibration_*.csv`, `path_scores_*.json`, `economic_test.csv`21 and listed in `docs/PHASE_5..7_REPORT.md`.22- Bibliography: 54 entries, every DOI/arXiv id fetched and title-matched (`scripts/build_references.py`,23 `results/metrics/references_verification.json`); four entries without DOI are typed by hand (LightGBM, Holm, Sklar, López de Prado).2425## Claims → evidence (central ones)26| Claim | Evidence |27|---|---|28| All models beat the random walk by 3.5–4.9 % | `main/compare_crps_vs_naive_rw_test.csv`, `final/…_final.csv` |29| H1 effect ≤ 0.1 % in sample, absent in 2025 | `main/pairwise_test.csv`, `final/pairwise_final.csv`, `ablations/pairwise_test.csv` |30| HAR-RV better from 15 min, neural better at 5 min | same files, `tables/main_crps_scaled.tex`, `final_crps_scaled.tex` |31| Transfer without penalty | `main/pairwise_test_heldout.csv`, `final/pairwise_final_heldout.csv` |32| H3, H4, H5 nulls | `ablations/pairwise_test.csv`, `stress_calibration_test.csv`, `path_scores_mrpt_small_copula_test.json`, `main/economic_test.csv` |33| Time-only ≈ 90 % of the skill | `ablations/pairwise_test.csv` (mrpt_small_time_only vs mrpt_small_b4k; b4k vs naive_rw) |3435## Checks36- `latexmk` exit 0; 0 undefined references/citations after the final pass; 1 remaining overfull box (appendix table, cosmetic).37- Visual check of the results pages (figures and tables render; legends below panels).38- No proprietary bars, grids or per-origin returns appear in the manuscript, tables or figures (scores and tickers only).39