Route within fee-free allowances across venues — rail 14 is the boundary, never a budget to relax

[eaitbrahim]

Rail 14 caps monthly BUY notional at the venue's fee-free volume allowance — the cross-verification showed it is the profitability boundary, not a budget: inside the allowance the reconstruction sits at break-even (14.9% vs 14.88%); outside it, at the 120 bp taker fee, break-even demands 29.18%. And per #427, at one venue's $500/month the 258 pooled trades the power analysis wants is about 8 years away.

What

• Route and schedule BUY notional to stay inside each venue's monthly allowance, across venues (Coinbase today; Alpaca via Phase 12 — keel-broker-alpaca Phase B: equities paper profile, attestation semantics, runbook #370, keel-broker-alpaca Phase C: cost-fidelity restatement + DCA benchmark on equities #371)
• Maker-vs-taker behavior where the venue rewards it
• Multi-venue allocation chosen to maximize in-allowance throughput

Guardrail (non-negotiable)

Rail 14's cap is never relaxed to trade more. This work moves trades INTO the allowance; it never enlarges it. Anything that trades through the cap is a compliance regression, full stop.

Success metric

Stated in n_eff, not trade count — signals herd (about 8 assets/day, ICC 0.21, DEFF 2.58), so pooled trades grow faster than evidence. Deliverable includes an honest time-to-detection estimate per venue mix.

[eaitbrahim]

First slice landed in #481: keel/research/throughput.py — per-venue in-allowance throughput (min(signals, allowance/mean notional); at the measured numbers, 0.1187 trades/month against $500), the n_eff machinery (DEFF 2.57516 from the measured k=8.43/ICC=0.212; every time-to-detection claim divides by it, per #427), and the greedy never-breach allocator with a rendered report that states the rail 14 guardrail outright. 14 tests reproduce the published values.

Remaining for this issue: maker-vs-taker routing (gated on #333) and runtime wiring into the venue profiles (Phase 12 #370/#371).