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

Feature #5 — lziv_complexity_signs (PR #95)

← Dashboard · PR terrylica/mql5#95 · card 40 · cut off bootstrap #90

What it computes

Normalized Lempel-Ziv 1976 complexity (DOI:10.1109/TIT.1976.1055501) of the binary up/down sign sequence of the per-bar mid log-return over the 200-bar strictly-trailing window — an entropy-rate / predictability proxy for the bar's directional sequence. Causal: the window is the 200 bars BEFORE bar i (current bar excluded). Value ≈ 1 for a near-random sign sequence. None on warmup (<16 finite trailing signs). Substrate "signs": np.sign(logret), logret = ln(close)−diff (8 h weekend/Ouroboros gap-nulled, like #1); binarization (s>0) → bit 1 iff up-bar, else 0 (a flat bar logret==0→sign 0 and a down bar→sign −1 BOTH map to bit 0); the finite-sign filter drops gap-null / first-bar signs.

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

3-way FOSS finding (it earned its keep)

antropy is the canonical Python LZ76 and there is no second mainstream independent implementation. So the independent leg is a from-scratch clean-room estimator sharing NO code with antropy: the textbook "exhaustive history" substring-membership formulation (each phrase is the shortest prefix of the remainder not yet a substring of the consumed history) vs antropy's Kaspar-Schuster exhaustive-production state machine — a different algorithm computing the same LZ76 production complexity. It is first self-validated against antropy's raw counts on 6 pinned vectors (all-up→2, alt→3, n16→6, rand32→9, doc→6, [1,0,…]→3), then agrees bit-for-bit (max|dev| 0.000e+00) across all 11200 windows. Triangulates with Rust==antropy at 2.22e-16 ⇒ antropy == clean-room LZ76 == Rust.

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

• C1 producer integration — held + residual observe() forms this bar's sign from close_mid + the prior close/close-time (gap-null), reads the normalized LZ76 over the trailing 200 signs BEFORE this bar, then folds this bar's sign in. Residual: live bigblack producer-vs-oracle smoke (no CH/MT5 on laptop — deferred to operator).
• C2 LZ76 counting — held: two independent algorithms (antropy Kaspar-Schuster; clean-room substring-membership) agree bit-for-bit over 11200 windows AND on 6 pinned raw-count vectors; the Rust kernel (exact port of antropy's machine) matched the SSoT at 2.22e-16. Three independent counters, identical phrase counts.
• C3 the sign substrate — held: logret uses numpy np.diff(np.log) (two-log form, the #94 lesson); np.sign maps up→+1 / down→−1 / flat→0 / gap→NaN; binarization (s>0) collapses down+flat→bit 0 and drops NaN — a unit test pins flat≡down and another pins the NaN-drop, so the producer's sign(logret) reconstruction matches the oracle exactly.
• C4 tolerance — not relaxed (kernel 1e-9, measured 2.22e-16; the FOSS leg agrees at exactly 0.0).

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.
• 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/#92/#93/#94.
• BarOrthogonalExtV1 field-union: created fresh off bootstrap with lziv_complexity_signs; field-unions with #91/#92/#93/#94 at operator merge. Merge order: #90 → #91 → #92 → #93 → #94 → #95.
• 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).

Next

Feature #6 — mann_kendall_z (card 20, Mann-Kendall trend test: EXACT integer S statistic, ε on the Z standardization).

Presentation only — never the SSoT.