Iteration 08 · 2026-06-09 · driver: native Claude Code /loop (laptop run)

Feature #7 — sevcik_fd (PR #97)

← Dashboard · PR terrylica/mql5#97 · card 10 · cut off bootstrap #90

What it computes

Sevcik (1998) fractal dimension (arXiv:1003.5266) of the per-bar mid-close curve over the 200-bar strictly-trailing window — a parameter-free roughness / space-filling measure (≈1 for a smooth path, →2 for a maximally jagged one). Causal: the window is the 200 bars BEFORE bar i (current bar excluded). None on warmup (<200 trailing closes). Map the curve into the unit square (x normalized by index xs[k]=k/(n−1), y by min/max), measure the total point-to-point Euclidean curve length L, then FD = 1 + ln(L)/ln(2(N−1)); a flat window (rng=max−min≤0) returns 1.0. Substrate = RAW mid-close level (NO log-return, NO weekend/Ouroboros gap-null — gaps are part of the level path, like hvg_clustering / mann_kendall_z). No new dep (f64 sqrt/ln); card 10 tier ε 1e-9.

The chain — one SSoT, bit-exact to the column

3-way FOSS finding (the FOSS errata caught)

The Sevcik FD is corroborated THREE independent ways: the probe-form (SSoT), a from-scratch clean-room explicit-loop unit-square curve length (no shared np.diff/vectorized-sqrt code; agrees with the probe at 4.441e-16), and neurokit2.fractal_sevcik (the real FOSS reference; agrees at 2.220e-16). ERRATA carried through the whole chain: the candidate card labels antropy@BSD-3, but antropy has NO Sevcik FD — using it would have been impossible. The bootstrap errata correctly flagged neurokit2.fractal_sevcik (MIT) as the real FOSS reference; the oracle, kernel, schema COMMENT, metadata + this FOSS leg all carry the correction. neurokit2 uses np.linspace(0,1,N) for the x-axis vs the probe's np.arange(N)/(N−1) (multiply-by-reciprocal vs divide, ≤1 ulp) — hence the 2.2e-16, well inside 1e-9.

Challenge-and-Held (inline; sub-agent dispatch skipped per the tick-3 signal)

• C1 producer integration — held + residual observe(close_mid) reads the Sevcik FD over the trailing 200 raw closes BEFORE this bar, then folds this close in; no gap-null / no reset (gaps are part of the level path the curve traces). Residual: live bigblack producer-vs-oracle smoke (no CH/MT5 on laptop — deferred to operator).
• C2 the FOSS errata (the trap) — held: the card's antropy label is impossible (antropy has no Sevcik FD); the corrected neurokit2 reference agrees with the SSoT at 2.2e-16. Errata carried through oracle/kernel/schema/metadata + FOSS.
• C3 normalization + curve-length — held: unit tests pin exact probe values (inc10 diagonal → 1.1199; smooth 200-line → 1.0579; 0/1 zigzag → 1.8842 with explicit roughness ordering zigzag>line); flat-window degenerate returns 1.0 (not None), matching the probe; clean-room explicit-loop leg agrees at 4.4e-16.
• C4 tolerance — not relaxed (kernel 1e-9, measured 6.661e-16; probe↔clean-room agrees at 4.4e-16).

Gates / scope

• mise run check:all-gates green on macOS except (a) worktree-context check:git-hook-installed and (b) the STANDING foreign research-orthogonality substrate-skip (see handoff). feature-metadata:verify-offline PASS. rust-bash-line-ref-bounds 199 refs / 0 failures.
• Operator-approved carve-out: the new CompletedBar field required a 1-line Default::default() in 9 explicit literals (2 blocklisted test files) — exact-match on the stripped literal opener, so -> CompletedBar { return-type lines are never touched. Same pattern as #91–#96.
• BarOrthogonalExtV1 field-union: created fresh off bootstrap with sevcik_fd; field-unions with #91–#96 at operator merge. Merge order: #90 → #91 → #92 → #93 → #94 → #95 → #96 → #97.
• Hygiene debt (no NEW): the research-orthogonality substrate-skip recurs as a STANDING foreign item (first surfaced #92; rust-enhancement loop follow-up; scripts/research/ untouched — 0 files).

Next

Feature #8 — sign_two_state_entropy_rate (card 83, 2-state sign-Markov entropy rate; EXACT+ε; ref impl Girardin–Sesboué + pyitlib).

Presentation only — never the SSoT.