spb/anomaly-atlas Public License

1---2project: anomaly-atlas3document: expG_cost_frontier/analysis4author: Simon-Pierre Boucher5contact: contact@spboucher.ai6data_source: hfmarketdata.io7created: 2026-08-128modified: 2026-08-129status: reviewed10---1112# Analysis — expG_cost_frontier1314Run: `results/expG_cost_frontier/20260812T072128Z/results.json` (pool rebuilt15mechanically from committed expC/expD/expF outputs — no hand-picking; cost16model and κ sweep pre-declared; cache-served).1718## The frontier1920| κ (× half-spread paid per trade) | survivors / 31 |21|---|---|22| 0 (gross) | 31 |23| 0.1 | 3 (AXDX 30min, CKX 1day, HTD 5min — all sparse names) |24| 0.25 | **1** (CKX 1day, κ* = 0.28) |25| 0.5 | 0 |26| 1.0 | 0 |2728Median breakeven **κ\* = 0.0114**: the median double-filtered rule captures29~1 % of one half-spread per trade. The intraday reversion cells sit at30κ\* 0.004–0.04; the 2014-2015 lead-lag residuals at 0.003–0.03; ES→SPY at310.0028 (identical across all three splices). Zero intraday rules survive32κ = 1 — the pre-registered falsification clause did not trigger.3334## Reading35361. **The three-layer doctrine closes.** expF showed statistical correction37   cannot detect mechanism; expG shows the mechanism's price: the gross38   survivors were harvesting exactly the thing they would have to pay.39   Charter Q5 answered on this pool: the cost frontier sits an order of40   magnitude below the most optimistic execution assumptions41   (Frazzini-style κ ≈ 0.1–0.25).422. **The lone κ=0.25 survivor is the skeptic's case study.** CKX (ultra43   sparse; wide, noisy EDGE spread; daily contrarian) has the classic44   profile of estimation artifact rather than economics. It is NOT45   discarded by hand — it goes to expH's validation split carrying the46   skeptical prior, which is what the protocol is for.47## Hand-off4849expH evaluates on the untouched validation split: (i) CKX 1day (the lone50cost survivor), (ii) the daily 2008-2015 reversal family (gross,51cost-marginal — evaluated for the decay/negative record), and (iii) the52NEGATIVE finding itself ("nothing intraday survives costs") — which, if it53replicates out-of-sample, becomes the atlas's first confidence-labeled54entries (charter result-types C and E).55