The authoritative catalog of every column in fxview_cache.forex_bars lives in the mql5 repo at docs/feature-catalog.html (preview: eon.25u.com/mql5/). For each column it records name ยท family ยท market ยท contributor ยท meaning. Of its 229 cataloged columns, the contributor index reads @terrylica 216 ยท @nasimubd 13. This page summarizes those 13 @nasimubd columns and records the provenance verification performed against the repo (the catalog was not taken at face value).
All are Nullable(Float64), market Forex+Gold, family Feature, and CAUSAL (they consume only completed bars โค iโ1; the current bar is excluded). Each is computed over a fixed 200-bar strictly-trailing window โ parameter-free, no tuned knobs โ and is a bit-exact port of a pinned reference implementation (11 at 1e-9; the two pure-spectral kernels at 1e-6). They are internally numbered "orthogonal-extension feature #1โฆ#13" with no gaps, and span deliberately independent mathematical families so they add orthogonal information rather than restate existing columns.
| # | Column | Family group | What it measures (plain terms) | Reference port |
|---|---|---|---|---|
| 1 | anderson_darling_a2 | Distributional shape | How non-Gaussian the recent return distribution is (goodness-of-fit) | scipy.stats.anderson |
| 2 | edge_spread_bps | Microstructure | Effective bidโask spread from OHLC alone (liquidity cost) | bidask.edge (Ardia et al., JFE 2024) |
| 3 | hvg_clustering | Visibility-graph topology | Local clustering of the price visibility graph (texture) | ts2vg + networkx |
| 4 | kendall_taub_ret_spread | Microstructure | Rank dependence between per-bar return and spread | scipy.stats.kendalltau(b) |
| 5 | lziv_complexity_signs | Sign-sequence complexity | Compressibility of the up/down sign sequence | antropy.lziv_complexity |
| 6 | mann_kendall_z | Trend / drift | Rank-based monotonic drift / trend strength | MannโKendall (probe) |
| 7 | sevcik_fd | Fractal / roughness | Fractal roughness of the price curve (1โsmooth, โ2 jagged) | Sevcik 1998 (neurokit2) |
| 8 | sign_two_state_entropy_rate | Sign-sequence complexity | Predictability of the sign chain (0=predictable, 1=random) | 2-state Markov entropy rate |
| 9 | spectral_log_slope | Spectral shape | Spectral colour โ the 1/fแต tilt (white โ pink/Brownian) | real-DFT log-log OLS |
| 10 | pomeau_irreversibility_lag1 | Time-irreversibility | Time-asymmetry of returns (is the series reversible?) | trev (arXiv:0708.4022) |
| 11 | hvg_degree_assortativity_coefficient | Visibility-graph topology | Whether visibility-graph hubs link to hubs (2nd-order topology) | networkx assortativity |
| 12 | hvg_mean_shortest_path_length | Visibility-graph topology | Global small-world path geometry of the visibility graph | networkx avg shortest path |
| 13 | psd_wiener_spectral_flatness | Spectral shape | Tonality vs noise โ one dominant rhythm vs broadband | Wiener entropy (MPEG-7 lineage) |
Every claim in the catalog was cross-checked against the source of truth โ oracle test, Rust kernel, metadata, writer field, live DDL, and git authorship โ rather than trusted from the catalog text. All green.
| Check | Result |
|---|---|
| Catalog header count | @nasimubd 13 โ |
| Features attributed to @nasimubd | exactly 13/13 โ |
| Internal numbering #1โฆ#13 | sequential, no gaps/dupes โ |
Oracle test per feature (tools/fxview-backfiller/tests/oracle_*.py) | 13/13 present โ |
Rust kernel per feature (orthogonal_extension_kernels/*.rs) | 13/13 present โ |
| Git first-author of every kernel | MD NASIM <nasimubd21@gmail.com> โ 13/13 โ |
feature_metadata.yaml coverage | 13/13 โ |
BarRow writer struct (crates/fxview-core/src/writer.rs) | 13/13 โ |
Live forex_bars DDL (tools/fxview-sidecar/schema.sql) | 13/13 columns, full COMMENT โ |
Conclusion: the catalog's account of these 13 columns โ name, family, market, contributor, and meaning โ aligns with the source of truth. Every feature traces oracle โ kernel โ metadata โ writer โ live column, all first-authored by @nasimubd, and each schema COMMENT matches the catalog meaning verbatim.
sevcik_fd notes the SOURCE card mislabels antropy (real ref is neurokit2.fractal_sevcik); spectral_log_slope and psd_wiener_spectral_flatness note librosa does not compute the exact quantity, so numpy/scipy is the correct reference.docs/feature-catalog.html in the mql5 repo (all 229 columns, contributor + meaning per column)