#60 vg_time_directed_clustering_kld

In plain terms

In the visibility network, compares how tightly each point connects to its past neighbors versus its future neighbors. An imbalance means the series looks different when played backward — a time-arrow signature captured through triangle structure rather than simple connectivity.

What this iteration will do

1. worktree, branch feat/odb-ext/vg_time_directed_clustering_kld; draw/refresh the dashboard flowchart FIRST (R6)
2. freeze the gap-aware oracle oracle_vg_time_directed_clustering_kld.py over the frozen fixture → forex_orthogonal_extension_vg_time_directed_clustering_kld_oracle-v1.jsonl + .sha256 (transcribed from probe kernel k_vg_time_directed_clustering_kld — NEVER the discovery harness)
3. port the causal Rust kernel + rolling state (libm-only transcendentals, NaN/Inf-total, compute-then-update)
4. wire the producer in the SAME PR: BarOrthogonalExtV1 field · writer.rs BarRow/COLUMN_NAMES · schema.sql TEXT-only DDL · feature_metadata.yaml · v4 cohort slot (R4 no split-brain)
5. robustness suite ALL GREEN: non-vacuous parity (batch ≡ streaming) at the locked tolerance · ≥2-symbol replication · crash-stress · golden snapshot · hardening witnesses (RG)
6. mise run rust:check + mise run check:all-gates — red gate ⇒ fix-or-halt (R7), never --no-verify
7. Challenge-and-Held (1 Attacker + 1 Defender) → committed challenge-and-held-*.md; fix conceded gaps before PR-open
8. open the PR (never merge) · overwrite this page with full results · append the iteration-ledger row · Pushover briefing