VIG
the house takes its cut — the research record
2026-08-02 17:49

Everything below is the actual research record — the trials, the numbers, the negative results. The paper sim on 07 PREDMKT only trades what survived this page.

Registered experiments
~70
every variant counted, none hidden
Strategies killed
15
including the flagship (0.28 → 0.05)
Survivor
0.85
walk-forward Sharpe · 77-instrument xsec · 12/15 yrs+
Original finding
0.87
2026 calibration slope — FLB reversed vs literature

The record — read in this order

The campaign, start to finish
How three days of research went: the equity work, the anomaly audit, the prediction-market falsification, and why the Desk exists.
5 sections · RESEARCH_STORY.md
Knowledge base
The condensed brain: verified edges, falsified claims, the math that matters, operating doctrine.
9 sections · KNOWLEDGE_BASE.md
The full trial log
Chronological record of every registered trial, decision, and negative result. The paper trail behind every number.
36 sections · RESEARCH_NOTES.md

Headline artifacts

Walk-forward config selection — the honest number
The harshest gate in the repo: choose configs on trailing data only, measure forward. In-sample Sharpe 0.28 became 0.05. Signal real, money gone.
cycle 9 · the one to read first
Study 2 tearsheet — cross-sectional equity
IC t=3.35 over 756 weeks, clean shuffled-label control, and why it still doesn't pay after costs.
flagship · falsified
Study 1 tearsheet — multi-asset
78 instruments, classic indicators dead after costs (best DSR 0.19), ML timing an honest failure.
baselines
The one-pager
The whole argument on one page.
pdf

Deep-research library — 11 verified literature reviews

Arbitrage across all forms and venues
Date: 2026-08-02. Scope: every arbitrage form the desk can name, across every real-money-priced venue it can reach, ranked by what is actually wireable for a patient $0, …
deep research · 2026-08-02
Arbitrage: building the best machine
Deep research report, 2026-07-31. All claims carry their source in parentheses. Items that failed or escaped verification are flagged inline and collected in the "Refuted…
deep research · 2026-07-31
Backtesting without lying to yourself
The mature betting-evaluation literature converges on a clear methodology for the Jan–Jul 2026 Polymarket backtest: evaluate edge against a sharp reference price rather t…
deep research · 2026-07-24
Data: sources and independent collectors
2026-07-31. Deep-research pass over five dimensions: on-chain mining, already-indexed datasets, official Polymarket APIs, cross-venue references, and collector engineerin…
deep research · 2026-07-31
Where edges still exist
The verified record supports a clear anatomy of trading edges: every documented edge had an identifiable structurally-constrained loser (index funds bound by tracking-err…
deep research · 2026-07-23
ML, systems and data engineering
Confirmed claims below survived 3-vote adversarial verification. Unverified claims from this run are NOT included and must not be quoted.
deep research · 2026-07-23
Factor families, replicated
100 agents; 3-vote verification. Summary:
deep research · 2026-07-23
Fees, and the arbitrage that survives them
Deep research report, 2026-07-31. All claims sourced in parentheses; items not confirmed against a primary source are marked unconfirmed or partly confirmed. Verification…
deep research · 2026-07-31
Implementing the patient games
Deep research report — 2026-08-01. Sources: Polymarket docs/APIs, Kalshi trade API, arXiv primary literature, official repos, verified press. Every claim carries its sour…
deep research · 2026-08-01
Masters of niche edges
Confirmed claims below survived 3-vote verification. Synthesis interrupted by usage limit.
deep research · 2026-07-23
The quant math that matters
Summary:
deep research · 2026-07-23
Net alpha: costs and construction
2026-07-22. 104-agent research run: 22 sources, 102 claims extracted, 25 verified by 3-vote adversarial panels → 20 confirmed, 5 refuted. Confidence labels per finding.
deep research · 2026-07-22
Novel small-market moats
Deep research, 2026-08-02. Which NEW small-market edges hold a real moat reachable by a $0-then-small solo F-1 operator. Sourced in parens; unconfirmed marked. Not legal/…
deep research · 2026-08-02
Tier 6 & 7: the penny edges (partial)
Deep research, 2026-08-02. Question: do any tier-6/7 "penny" micro-edges genuinely clear break-even for a $0-budget, automated, F-1 operator? Scope covers method-level mi…
deep research · 2026-08-02
Prediction markets: the venue
The verified evidence supports a clear picture of prediction markets as a professionally tradeable arena with documented, quantified inefficiencies. A peer-reviewed on-ch…
deep research · 2026-07-23
Who trades prediction markets
Institutional entry into prediction markets is real, named, and accelerating — but later and narrower than folklore suggests. Susquehanna (SIG) became Kalshi's first dedi…
deep research · 2026-07-23
Risk, sizing and decision theory
Confirmed claims below survived 3-vote adversarial verification. Unverified claims from this run are NOT included and must not be quoted.
deep research · 2026-07-23
The small-market cost-floor moat
Deep research, 2026-08-02. Every substantive claim sourced in parens. Unconfirmed items marked. Read the "Refuted / overhyped" section before acting on anything.
deep research · 2026-08-02
Attention signals: the WSB postmortem
Verification failed (usage limit). Claims are single-source extractions, NOT adversarially confirmed. Treat as leads, not facts.
deep research · 2026-07-23

Where it leads

Every directional strategy on the prediction-market venue pays a spread wider than the mispricing — measured both ways. The only structural winner is the maker. That hypothesis is now running live on paper, self-grading: 07 PREDMKT →