The Idea ↑ top
Polymarket's daily "will Bitcoin be above $X" and "between $X and $Y" contracts are priced by a retail crowd. The listed options market on Deribit is priced by professionals, and its deltas, implied volatility and skew already say where those professionals think price will land. Shadow Edge reads the second market to price the first. It is poker where you can see the other professionals' hands.
The gap is real and it has now been measured independently: a June 2026 academic paper (arXiv 2606.19517) puts the average wedge between Polymarket and options-implied probability at 5.6–6.3 percentage points, decaying with a half-life of roughly four hours. That is the entire opportunity, and it says two things at once: the edge exists, and it belongs to whoever is fastest and most disciplined.
How It Works ↑ top
| Universe | Polymarket daily BTC and ETH price contracts resolving 16:00 UTC; range ("between") contracts only as of 25 Aug |
| Fair value | Deribit call deltas interpolated across the 08:00 UTC expiries to the 16:00 UTC resolution; the professionals' implied probability, on our clock |
| Entry | Ask price 0.55–0.75; our probability ≥ 51%; net edge after fees ≥ 5 pp; one position per asset; $5 maximum ticket |
| Exit | Hold to resolution (current). Edge-captured take-profit and thesis-broken exits are specified but not yet live |
| Hedge | None since 12 Aug; Deribit contract minimums dwarf pilot tickets |
| Stack | Python runner on a dedicated server, Polymarket CLOB via proxy wallet, live positions published at levenstein.net/polymarketinfo |
| Oversight | Daily loss-analysis run at 07:00 Bali; daily sign-off call; independent verification of the trade ledger; Polymarket is the position record of truth |
Honest History ↑ top
| Mar – Apr 2026 | Manual trading from Nicholas's own wallet. 54 markets, 26 won / 28 lost, +$35. Proved the wedge exists; proved a human cannot watch it. |
| 29 – 30 Apr | Two manual losses (−$23 from a five-minute delay, −$68 from an unmonitored position). Decision: full automation. |
| 19 May | First fills through the automated pipeline. Two positions, two wins. |
| 25 Jun | Worst day. A sign inversion on a newly added feed bought four deep out-of-the-money contracts; the small account went to zero. Refunded to ~$40; hard price band written into the spec. |
| 30 Jun – 8 Jul | 52 resolved trades, net −$1.69. The 16 in-band trades went 16 for 16 (+$26); the losses were all out-of-band fills the runner should have refused. |
| 16 – 20 Jul | Wallet rotation broke the signer / API-key pairing; 162 orders rejected. Fixed 20 Jul; current wallet funded. |
| 28 Jul – 11 Aug | Sixteen days of silence: the runner was not running. Then 190,795 decisions and zero approvals in a low-volatility regime. |
| 12 Aug | Restart, unhedged, hold-to-resolution. Model reverted to v1 without sign-off. |
| 17 – 24 Aug | One-per-asset breaches; a 74-order retry storm into a Polymarket outage (19 Aug); wrong-expiry pricing (−$10, 20 Aug); near-expiry pricing off a dead endpoint; a kill order on "above" markets executed seven hours late (24 Aug). |
| 25 – 26 Aug | Range-only in code; 60-second cadence confirmed; daily sign-off instituted; expiry-bracket formula v1.2 drafted. |
Results ↑ top
On-chain wallet activity, not internal ledgers. Dollar figures are pilot-sized by design.
| Period | Markets | Won / Lost | Deployed | Net |
|---|---|---|---|---|
| Manual · 24 Mar – 26 Apr | 54 | 26 / 28 | $1,624 | +$34.86 |
| Automated · 20 Jul – 26 Aug | 40 | 25 / 15 | $176 | −$7.36 |
| of which range contracts, 12 – 23 Aug | 8 | 6 / 2 | — | +$3.65 |
| of which "above" contracts, 12 – 23 Aug | 12 | 6 / 6 | — | −$14.21 |
| In-band cohort · 30 Jun – 8 Jul | 16 | 16 / 0 | — | +$26.11 |
What Changes Next ↑ top
Every loss on this page has a name and a fix. The programme below takes the fixes in the order that removes the most risk soonest, and only then scales.
Formula v1.2: the two Deribit expiries must bracket the Polymarket close; no entries after 08:00 UTC; no entries inside the final 60–90 minutes; fall back to the far chain when the near one goes dead. This alone would have blocked the wrong-expiry losses of 20 Aug and every dead-endpoint fill of 23–24 Aug.
Price band 0.55–0.75 enforced in code, the 0.85 confidence haircut actually wired, one position per asset read from the live wallet, a portfolio-level overlap check, and range-only until the "above" branch is explained line by line. The July in-band cohort went 16 for 16; this makes in-band the only kind of fill there is.
Redemption write-back so the ledger reconciles to the chain; then edge-captured take-profit and thesis-broken exits go live. With a four-hour edge half-life, capturing 3–5 points early and recycling capital is worth more than any single resolution.
Attempt cap, exponential backoff, order de-duplication, automatic halt on venue errors, and a kill switch the runner itself observes, with a market-type filter. Nicholas's stop authority becomes a control, not a convention.
Staging server before production; every strategy change carries a commit hash and a signature; a one-tap daily approval on Nicholas's phone; independent ledger verification; regression tests R1–R5 run on every release. Silent reversions and seven-hour kill delays end here.
After 30 consecutive clean resolved signals, limits step from $5 to $25 per ticket and a $1,000 bankroll; after 100, the sleeve-fund bar. Solana dailies are built and waiting on staging; Kalshi oil and gold ladders and XRP weeklies follow. Directional inputs (funding, basis, skew drift) run in shadow mode first. The wedge is public now; speed, breadth and discipline are the moat.
Status & Structure ↑ top
| Stage | Validation pilot; not open to investors |
| Capital at risk | ~$200 hot wallet; $1,000 bankroll planned ($200 hot / $800 reserve) |
| Ticket size | $5 maximum, by design, until 30 clean signals |
| Team | Nicholas Levenstein (principal); Aneep Tandel (engineering); independent ledger verification; external advisor |
| Transparency | Positions published live; every fill and every loss analysed daily |
| Fund status | Sleeve-fund structure follows a 100-signal validated record and legal entity; no solicitation until then |
Why We Publish This ↑ top
Most strategies show you the backtest. We are showing the pilot, including the day it went to zero, because the point of a validation phase is to find out what breaks before it matters. What broke was never the idea. Every one of the fixes above is specified, owned and dated, and the page will be updated as each goes live.