The verified record supports a clear anatomy of trading edges: every documented edge had an identifiable structurally-constrained loser (index funds bound by tracking-error mandates, Nasdaq market-makers hobbled by employee agency problems, hedge funds forced to sell SPACs into redemptions, uninformed retail buying against satellite-informed shorts), and each died not from secrecy loss alone but from regulation (SOES rule changes and 1997 order-handling reform), maturation of professional liquidity provision (S&P index effect collapsing from 7.6% to 0.8% even as mechanical index flows grew), or commoditization of once-exclusive data access (satellite imagery, whose edge persisted 2011-2017 only because of an access moat). For a solo researcher in 2026, the best-documented genuinely accessible classes among those verified are prediction-market arbitrage — ~$40M of realized riskless extraction on Polymarket (peer-reviewed, on-chain measured), with combinatorial cross-market arb the least-exploited and hardest-to-detect subclass, but capacity explicitly bounded to retail scale (median 101 bps per episode, 76.9% of opportunities executable at only ~15 shares) — and structurally-floored small-capacity event trades of the SPAC-trust type, where crisis-driven forced selling periodically hands individuals 5-10% annualized near-risk-free returns. Crypto futures-spot carry averaged a documented 7-8% annualized (spikes >40%) over 2019-2024, driven by leveraged retail demand against limited arbitrage capital, though the 2024 spot-ETF launch compressed it. The consistent capacity lesson matches the individual's thesis exactly: the surviving edges are ones institutions cannot economically touch — sub-$2M tickets, eighths-and-quarters economics, and order books measured in tens of shares.
The S&P 500 index effect is a fully documented edge-death timeline: addition abnormal returns rose from 3.4% (1980s) to 7.6% (1990s), fell to 5.2% (2000s), and collapsed to a statistically-zero 0.8% in 2010-2020; deletions mirrored it (-4.6%, -16.6%, -12.3%, -0.6%). Critically, the edge died even though the mechanical demand shock GREW — index trackers went from buying ~0% of shares outstanding of additions to over 6% (nearly 8% selling for deletions) — because professional active investors now absorb the flow as liquidity providers; the demand price-multiplier fell by a factor of ~20. The 'willing loser' (mandate-bound index funds trading at any price) still exists, but the compensation for fronting them has been competed to zero at S&P 500 scale.
high | 3-0 (x3 merged claims)
Greenwood & Sammon, 'The Disappearing Index Effect' (Journal of Finance 80(2), 2025; verified verbatim from full PDF text): addition returns '3.4% in the 1980s... 7.6% in the 1990s... 5.2%... 0.8% in the most recent decade, statistically indistinguishable from zero'; deletions '-4.6%... -16.6%... -12.3%... -0.6%'; 'M has indeed declined by a factor of approximately 20 for index additions' (6.7 late-1990s to 0.30 last decade); tracker buying grew from ~0% to 6%+ of shares outstanding while total institutional ownership barely moves around changes, which the authors interpret as professional liq
A related but distinct rebalance edge remains documented at longer horizons: in the year following S&P 500 changes, discretionary deletions beat additions by an average of 22 percentage points — cap-weighted indices structurally buy richly-valued recent winners and sell cheap recent losers, with tracking-error-minimizing index funds (and their end investors) as the identifiable loser. But the exploitable alpha is thin at index scale: simple rules (trading ahead of index funds, or delaying reconstitution trades 3-12 months) add only up to ~23 bps/year to a cap-weighted portfolio — real but capacity-constrained (discretionary deletions are a small, illiquid subset), consistent with a niche rather than a free lunch.
high | 3-0 (x3 merged claims)
Arnott, Brightman, Kalesnik & Wu, 'Earning Alpha by Avoiding the Index Rebalancing Crowd' (Financial Analysts Journal 79(2), 2023, peer-reviewed): 'discretionary deletions beat additions by 22%, on average'; 'delaying reconstitution trades by 3 to 12 months, can add up to 23 basis points a year'; indices 'add stocks with high valuation multiples after persistent outperformance.' Corroborated in magnitude by Petajisto (2011, ~21-28 bps/yr index turnover drag). Merges claims 8, 9, 10 (all 3-0). Caveats verifiers flagged: multi-decade average likely front-loaded in earlier decades; authors (Resea
SOES bandits (1990s Nasdaq) are the best-documented template for an individual-scale market-structure edge. Documented profitability: at brokerage firm A (5,188 round trips, Nov 30-Dec 6, 1995), median round-trip profit $63, mean $72.48 (SE $6.13), leaving $22.48 after the $50 round-trip commission, significant at 1%. Willing loser: Nasdaq market-makers — bandits won DESPITE having less information, because dealer employees had weak incentives to update quotes (agency problems) while bandits risked their own capital. Mechanism: cross-venue quote-staleness arbitrage, not raw speed — buy via SOES at stale quotes before MOST dealers updated, hold minutes, exit inside the spread on SelectNet/Instinet; bandits who entered AND exited through SOES usually lost money. Capacity profile: 1,000-share trade cap, ~35% of trades lost money pre-commission, only ~25% of trades moved more than $250, ~50% of attempted orders executed due to bandit competition — a living made 'taking eighths and quarters.'
high | 3-0 (x4 merged claims)
Harris & Schultz, 'The Trading Profits of SOES Bandits' (Journal of Financial Economics 50, 1998, 39-62), verified against full text: 'SOES bandits' profits come primarily from market-maker losses'; 'they trade profitably with market-makers despite having less information'; 'Bandits who both initiate and close positions through SOES usually lose money'; 'the popular image of SOES bandits trading with the slowest and least alert market-maker is inaccurate'; agency-cost explanation stated explicitly by the authors. All statistics verbatim from the paper. Merges claims 3, 4, 5, 6 (all 3-0). Minor
The SOES edge died by regulation and market-structure reform, not pure crowding: NASD barred professional SOES traders in 1988 (thrown out by the D.C. Circuit in 1993 as vague/arbitrary — Timpinaro v. SEC); the max SOES trade was cut from 1,000 to 500 shares in January 1994, after which max-size volume 'dropped dramatically' (the authors infer market-maker losses to bandits fell); 1,000 shares was restored March 1995; and the 1997 Nasdaq order-handling rules forced dealers to display Instinet prices on Nasdaq, reducing (not eliminating) the cross-venue gaps bandits exploited. Lesson for the taxonomy: edges that tax a politically-connected loser (dealers) get regulated away; edges built on market-structure seams die when the seam is closed.
high | 3-0
Harris & Schultz (JFE 1998) primary text verified verbatim, independently corroborated by GAO GGD-98-194 chronology (1988 PTA rule, Timpinaro v. SEC 2 F.3d 453 remand, Jan 1994 500-share interim rules per FINRA NTM 94-1, March 1995 restoration, Jan 1997 order-handling rules). Claim 7, 3-0. Nuance: the court technically remanded and NASD withdrew the rules; practical effect identical.
SPAC trust arbitrage 2020-21 is a documented episode of a structurally-floored trade periodically handed to small players: IPO proceeds sit untouched in a trust of short-term Treasuries with a contractual right to redeem at NAV (~$10 + accrued interest) at deal vote or liquidation — a T-bill floor with equity/warrant upside. In the March 2020 COVID panic the willing losers were hedge funds (the main SPAC IPO buyers) forced to sell below NAV to meet client redemptions, letting buyers lock in 5-10% annualized near-risk-free returns before warrant optionality. Pre-boom capacity was quantified: as of May 2020, the entire universe was 98 SPACs, $27.5B market value, average -0.9% discount to NAV and 2.4% yield — i.e., the normal-regime return is thin and the crisis regime is where the edge lives.
medium | 3-0 (x3 merged claims)
Accelerate whitepaper (Klymochko, May 2020) verified verbatim by verifiers via full PDF extraction; trust structure independently corroborated by standard SEC EDGAR S-4 trust-account language (Treasuries ≤185 days, pro-rata redemption rights); hedge-fund dominance of SPAC IPOs corroborated by Klausner, Ohlrogge & Ruan 'A Sober Look at SPACs' (Yale J. on Reg.: IPO investors almost entirely hedge funds, ~97% redeem or sell pre-merger) and Forbes 'SPAC Mafia' reporting; the 5-10% annualized crisis yield independently confirmed by Progressive Capital (Nov 2020). Merges claims 11, 12, 13 (all 3-0).
Satellite parking-lot imagery is the canonical documented alternative-data edge with an ACCESS moat rather than a discovery moat: academics (Katona, Painter, Patatoukas, Zeng) analyzed 4.8M satellite observations of ~67,000 stores of 44 major US retailers (Walmart, Target, Costco, Whole Foods) over 2011-2017 using RS Metrics/Orbital Insight data; year-over-year car counts reliably predicted quarterly sales, and trading the signal yielded 4-5% abnormal returns over the three days around earnings. The willing loser was retail: individuals were net buyers of exactly the retailers satellite-informed hedge funds were shorting. The edge did NOT decay within the 2011-2017 sample — because the data remained expensive and exclusive to large funds. Lesson: data edges persist as long as the data stays expensive; they die when the vendor commoditizes it, and an individual's analogue is data they RECORD themselves rather than buy.
high | 3-0 (x4 merged claims)
Berkeley Haas newsroom summary of the peer-reviewed study (published JFQA 2024; SSRN 3222741), all figures verified verbatim: 'year-over-year changes in the number of cars... is a reliable predictor of quarterly sales'; 'yields 4% to 5% in the three days around quarterly earnings announcements'; 'individuals are net buyers of the same retailers that the hedge funds are betting against'; 'access to satellite imagery data has been so exclusive... the value of the parking lot signals hasn't yet been competed away.' Merges claims 14, 15, 16, 17 (all 3-0). Caveats: the 4-5% is a backtested academic
Prediction-market arbitrage is the best-documented LIVE edge class for individuals, in two verified layers. (a) At scale, historically: Polymarket exhibited mispriced mutually-exclusive outcome sets (full sets buyable for <$1 or sellable for >$1, guaranteeing profit), and arbitrageurs actually extracted an estimated ~$40M realized profit (measured from 86M on-chain trades, Apr 2024-Apr 2025, concentrated in 2024 US election markets; top wallet ~$2.0M). Two distinct classes exist: Market Rebalancing Arbitrage (within one market) and Combinatorial Arbitrage (across dependent markets) — the latter is computationally hard to detect (O(2^n) naive pairing; the researchers needed LLM-assisted market matching) and was exploited orders of magnitude less (~$95K vs ~$43M), making it the most plausible remaining individual niche. (b) Capacity reality-check from NBA in-game markets: executable mispricings exist but are structurally bounded by order-book depth to strictly retail scale — 290 combinatorial episodes concentrated in final minutes of games, median return ~101 bps per episode, with 76.9% of opportunities constrained to an average executable size of 14.8 shares (~$7-15 notional), and only 7 executable single-market episodes (median duration 3.6s).
high | 3-0 (x6 merged claims)
Layer (a): Saguillo, Ghafouri, Kiffer & Suarez-Tangil, 'Unravelling the Probabilistic Forest' — peer-reviewed at AFT 2025 (LIPIcs, DOI 10.4230/LIPIcs.AFT.2025.27), Flashbots-funded; '$39,587,585.02' combined extraction verified in Section 7.4; press corroboration (DL News, Cryptopolitan). Layer (b): Cheng, Yang & Zou, 'Arbitrage Analysis in Polymarket NBA Markets' (arXiv 2605.00864, Apr 2026; UCLA-affiliated), 75M order-book snapshots across 173 games; abstract verified verbatim: 'confining risk-free extraction strictly to the retail scale.' Merges claims 18-23 (all 3-0). This finding is the s
Crypto futures-spot carry (basis) for BTC and ETH averaged a documented 7-8% annualized over March 2019-July 2024, with spikes at times exceeding 40% p.a. — a persistent, large dislocation driven by leveraged retail long demand against limited arbitrage capital, not a fleeting anomaly. It is the documented existence proof for the 'crypto funding/basis dislocation' edge class, but should not be projected forward at that magnitude: the January 2024 spot-ETF launch materially compressed the basis, and carry turns negative in crashes (e.g., March 2020).
high | 3-0
Schmeling, Schrimpf & Todorov (BIS Working Paper 1087; published Management Science 2026, DOI 10.1287/mnsc.2024.05069), summarized in their CEPR VoxEU column, verified verbatim: 'Carry is persistent, shows large spikes and averages around 7-8% per year, depending on the asset and contract'; 'at times exceeding 40% per annum'; Figure 1 confirms the Mar 2019-Jul 2024 BTC/ETH sample. Claim 24, 3-0. Peer-reviewed authorship by BIS economists; the mechanism (segmented markets, retail leverage demand, constrained arb capital) is the paper's own thesis — the identifiable willing payer is the leverage
Coverage gap is the biggest caveat: the surviving 25 claims document only a subset of the requested taxonomy. Nothing survived verification on odd-lot tender arbitrage episodes, rights offerings, closed-end fund discounts, kimchi premium magnitudes/closure, weather-forecast-error trading, AIS shipping data, credit-card panels, power-market seams, sports-syndicate crossovers, WorldQuant/pod alpha-factory mechanics, alt-data vendor commoditization timelines, retail-broker seams, or — critically — verified post-2015 stories of named individuals with credible return documentation. The synthesis therefore documents the edge-anatomy PATTERN rigorously (loser identity, death mechanism, capacity bounds) from the classes that did verify, but the requested ranked hunting-ground shortlist can only be partially evidence-backed: prediction-market combinatorial arb and crisis-regime trust/NAV-floor trades are the two classes with direct documented individual-scale evidence; crypto basis is documented but compressed post-2024. Source-quality caveats: the SPAC figures come from a fund manager's marketing whitepaper (every fact independently corroborated, hence medium confidence); the Polymarket NBA capacity numbers are from an unreviewed 2026 preprint with a 173-game sample; the satellite 4-5% and the academic index/rebalance returns are backtested academic estimates, not audited live P&L. Time-sensitivity: the index effect numbers differ slightly between working-paper and published versions (shape identical); post-2024 SEC rules changed SPAC trust mechanics; the crypto carry average is explicitly a 2019-2024 sample. Finally, 'willing loser' throughout means structurally- or agency-constrained, not knowingly consenting.