ODB × time-bar: which next step?

The situation, in one paragraph. Our time-bar feature set is saturated (a 77-indicator baseline had ~1 non-redundant feature) — more features is not the lever; turnover/cost is. ODB (event-clock) intra-bar order-flow features are the project's first temporally-stable signal (same sign 2022→2026) — genuinely new information — but they do not survive realistic trading costs. A strong gross signal (per-bar Sharpe ~0.9) collapses to negative at ~7 bps round-trip. So the question is not "add more data" — it is how to combine the two bar-clocks into something that survives cost, and which minimal experiment tells us most, fastest.

What the research says (adversarially verified). Across equities, options and crypto, strong-gross/high-turnover signals reproducibly die at realistic cost — cost is the binding constraint, not signal quality. Orthogonality between the two clocks is real in information but theoretically under-determined (classical mutual information can't cleanly separate "synergy" from "redundancy" without an extra axiom) — so it must be measured carefully, not assumed. The convergent recommendation is a small honest pilot before any fusion: establish single-clock baselines → measure the true added information → then test fusion under real cost. Options below are graded by Effort and strength of Evidence.