Monolith Research
News.
Working papers, research notes, and measured conclusions — filed into a numbered reading order.
The file is organised as a short curriculum. Each entry is a dated working paper or note; where a figure is referenced, it is reconstructible from a versioned notebook and a dated data snapshot, and where a snapshot is no longer available the entry is withdrawn rather than re-cut. New here? Begin with the method, then read where you like.
- 1 The Gate & Method How the firm decides a result is real — the validation gate, deflation, stressed-cost screening, and the discipline behind every pass. 5entries
- 2 Early Systems (2024–2026) The first systematic strategies — OU, Gold, NAS — their walk-forward records and what the research taught. In preparation
- 3 The Graveyard Strategies retired on out-of-sample evidence. The kill-record is the asset; a back-test is worth more next to the failures it survived. In preparation
- 4 The Alpha Hunts The standing search for a new edge — real phenomena, and the walls of cost, capacity, and data access that keep them out of reach. In preparation
- 5 Microstructure & Flow Order flow, latency, and the mechanical effects that live closer to the tape. In preparation
- 6 Risk, Ops & The Record Risk policy, prop-firm constraints, infrastructure, and the quarterly file. 4entries
- 7 What Is Actually Live The honest current status — what carries a real record, and what does not. In preparation
Rejected Hypotheses — Three approaches tested, three rejected before promotion
Three literature-grounded approaches were built, backtested, and formally rejected before any capital was allocated. Distance-based pairs, VIX-regime filtering, and ARIMA signal generation all failed to meet the promotion threshold on corrected, stressed-cost OOS evaluation.
Read the note-
The Trump Trader — a political signal, built as a joke
A keyword-based signal on Trump's Truth Social posts. OOS Sharpe −0.31. This was built as a joke. An expensive one, intellectually. It is filed here because even jokes go through the pipeline.
-
Quarterly - 2025 year-end
A short ledger for the year that built the file. What was opened, what was kept, and what remains under observation.
-
Note - Prop-firm constraints, translated into rules
Daily and total drawdown limits expressed as deterministic locks before any strategy code runs. The challenge is a constraint problem first, a strategy problem second.
-
Working paper - The promotion pipeline
A research candidate becomes a live candidate by passing through four named gates: walk-forward, paper, shadow, and small size. The order is the work.
-
Note - The risk shell goes first
Before any new signal, regime classifier, or learned policy, a deterministic wrapper the strategy cannot bypass. The order matters more than the choice of model.
-
Working paper - Stressed-cost validation as the first gate
A back-test that flatters itself under generous assumptions is a hypothesis. A back-test that survives a pessimistic cost model is a candidate. The two are not the same thing.
-
Note - On academic Sharpe above three
Each of the published papers we have attempted to replicate with the largest headline Sharpe figures behaved differently on real data. The replication budget exists for a reason.
-
Note - On opening the working file
A short statement on what these entries are and what they are not. Records for the desk, not signals for the reader.
Notes are research records, not recommendations. Strategy logic stays private; method, data, and evidence are made legible.