Shadow Edge — Weekly Scorecard
Week ending Friday 30 Oct 2026 · Week 1 of the frozen-rule record · Generated automatically every Friday 18:00 WITA
What happened vs. what the better rules would have done
Contents
1 · Six markets at a glance
Three assets × two products. Each row shows what we actually did and what the frozen rule for that market would have done in the same week. A market only earns real money after it has passed a hold-out week and 30+ shadow bets.
| Market | Rule | Status | Actual bets | Actual P&L | Rule bets | Rule win% | Rule P&L | Record (bets) |
|---|---|---|---|---|---|---|---|---|
| BTC range | Underdog rule | LIVE (old rule) | 5 | −$7.90 | 6 | 67% | +$19.40 | 6 |
| ETH range | Underdog rule | LIVE (old rule) | 4 | −$4.50 | 5 | 60% | +$11.80 | 5 |
| SOL range | Underdog rule | COLLECTING DATA | — | — | — | — | — | 0 · first look ~14 Nov |
| BTC above $K | Threshold rule | COLLECTING DATA | 0 | $0.00 | — | — | — | 0 · first look ~14 Nov |
| ETH above $K | Threshold rule | COLLECTING DATA | 0 | $0.00 | — | — | — | 0 · first look ~14 Nov |
| SOL above $K | Threshold rule | COLLECTING DATA | — | — | — | — | — | 0 · first look ~14 Nov |
Status ladder: COLLECTING DATA → SHADOW (rule logs bets, no money) → LIVE ($5 flat) → sizing review at 30+ live bets. Nick signs each step.
↑ Go to top2 · Cumulative P&L — actual vs rule, day by day
Both lines use the same minute data and the same fair value. Actual = fills in the Polymarket account export. Rule = the first minute each day the rule's conditions were all true, filled at the captured ask, resolved at 12:00 ET.
↑ Go to top2b · What this week's rate means per month and per year
The same two results, turned into a rate of return. Bets are a flat $5, so the honest base is the money the bets are drawn from: the $200 hot wallet the runner can reach. Shown also against the full $1,000 deposit and per dollar actually wagered. Monthly = weekly × 4.33; annual = weekly × 52, straight-line (flat bets don't compound).
| Base | Week | Month | Year | |
|---|---|---|---|---|
| $200 hot wallet runner's working capital | What happened | −6.2% | −27% | −322% |
| Underdog rule | +15.6% | +68% | +811% | |
| $1,000 deposit hot wallet + reserve | What happened | −1.2% | −5.4% | −64% |
| Underdog rule | +3.1% | +13.5% | +162% | |
| Per $1 wagered bets × $5 | What happened | −27.6¢ | — | — |
| Underdog rule | +56.7¢ | — | — |
Read with care. One week, 9–11 bets, illustrative numbers. A year's figure built from one week is a projection, not a forecast — a 64% win rate on 11 bets has a wide error band. The annual column becomes meaningful once the record passes 30 bets, and the live P&L only scales if sizing is raised, which needs Nick's sign-off. From the second Friday this section shows the cumulative rate since the record began rather than one week annualised.
↑ Go to top3 · Does the model know anything? (calibration)
The whole strategy rests on one claim: when our Deribit fair value disagrees with the Polymarket price, our number is closer to the truth. This table checks that directly, without any trading rule in the way. These rows are real — BTC+ETH range markets, 5–22 Oct, afternoon window (after 16:00 Bali), 16 days.
| Model minus market | Realised YES rate vs market-implied | Realised | Market said | Model was right by |
|---|---|---|---|---|
| +5 to +10 pp (model says YES is cheap) | 64% | 42% | +22 pp | |
| 0 to +5 pp | 55% | 47% | +8 pp | |
| −5 to 0 pp | 41% | 49% | +8 pp | |
| −10 to −5 pp (model says YES is dear) | 25% | 48% | +23 pp |
Reading it: in every bucket the outcome moved in the direction the model said, and the further the model disagreed, the more it was right. The middle two rows are illustrative pending the Friday run; the outer two are from the 24 Oct spec. From 14 Nov this section gets one table per asset and per product (six tables), because SOL and above/below each have to earn this on their own.
↑ Go to top4 · Hold-out verdicts
A rule is written down and frozen before it sees new data. Each Friday it is scored on the week it has never seen. Nothing in the rule changes between weeks unless Nick signs a new version — then the record restarts.
5 · Bet log for the week
Every bet the frozen rule would have taken, next to what the runner actually did that day. Illustrative rows.
| Day | Asset | Rule bet (Underdog) | Fair | Ask | Edge | Result | Rule P&L | Runner did | Runner P&L |
|---|---|---|---|---|---|---|---|---|---|
| Thu 23 | BTC | YES 110–112k | 0.49 | 0.41 | +8 pp | won | +$7.20 | YES 108–110k @0.66 | −$5.00 |
| Thu 23 | ETH | NO 3,900–4,000 | 0.44 | 0.38 | +6 pp | lost | −$5.00 | no bet | — |
| Fri 24 | BTC | YES 112–114k | 0.52 | 0.45 | +7 pp | won | +$6.10 | YES 112–114k @0.61 | +$3.20 |
| Sat 25 | ETH | YES 3,800–3,900 | 0.40 | 0.33 | +7 pp | won | +$10.15 | no bet (Saturday gap) | — |
| Mon 27 | BTC | NO 114–116k | 0.47 | 0.42 | +5 pp | lost | −$5.00 | YES 114–116k @0.58 | −$5.00 |
| … 6 more rows in the real page … | |||||||||
Rule P&L is $5 ÷ ask × (1 − ask) on a win, −$5 on a loss — i.e. the real payout of a $5 limit order filled at that ask. No fees are charged on Polymarket crypto daily markets at present; if that changes the formula changes here first.
↑ Go to top6 · Data health
If the capture has holes, every number above is suspect, so this is checked first. Target: ≥ 97% of minutes captured in the trading window, every feed.
| Feed | Owner | Minutes captured (week) | Coverage | Note |
|---|---|---|---|---|
| Deribit BTC chains (today / tomorrow / +1) | Minute-capture job | 99.1% | ok | |
| Deribit ETH chains | Minute-capture job | 98.8% | ok | |
| Deribit SOL chains | Gabriella | 0% | starts 28 Oct | |
| Polymarket BTC / ETH range (bid, ask, sizes) | Minute-capture job | 96.2% | gap Sun 26 02:00–03:40Z — server restart | |
| Polymarket SOL range | Gabriella | 0% | starts 28 Oct | |
| Polymarket above-on events, BTC / ETH / SOL | Gabriella | 0% | starts 28 Oct | |
| Resolutions (12:00 ET settle) | Gabriella | 7 / 7 days | ok |
7 · How this page is produced
Every Friday, 18:00 WITA, no hands
- Inputs: the minute capture files in
agents\TraderKate-Working\experiment\(Deribit chains + Polymarket books, all three assets, both products), the Polymarket account export (actual fills), and the frozen rule filekate_algo_v1_4.py(Underdog rule, Threshold rule) with its fingerprint. - Step 1 — data health: count captured minutes per feed; flag anything under 97%.
- Step 2 — actual: read the week's real fills and resolutions from the account export.
- Step 3 — rule: replay each frozen rule minute-by-minute over the same week, fill at the captured ask, score at resolution. The rule file's fingerprint is printed in the footer so anyone can confirm nothing changed.
- Step 4 — calibration: bucket every minute by model-minus-market and compare realised to implied, per asset, per product.
- Step 5 — render: write this HTML to
experiment\weekly-scorecard\SCORECARD_<date>.htmland overwriteSCORECARD_LATEST.html; optionally push to the Squarespace page.
Rules of the page
- A rule's parameters never change inside a record. A change = a new rule name, Nick's signature, and a fresh record starting at zero.
- Rule P&L is always at the ask with a size check — never at mid or last. (The +$50.65 stub result became a loss once this was enforced.)
- Nothing moves from SHADOW to LIVE on this page. The page recommends; Nick signs; Gabriella verifies the runner config matches.
- Gemini second-opinion on each new rule, with the same data pack, before it enters shadow (24 Oct ruling).
What Gabriella owns on this page
- Sections 6 (data health) and the resolution column in Section 5 — she confirms the capture ran and the settle prices are right.
- The VERIFY step whenever Section 4 recommends a status change.
Shadow Edge · GSC Nicholas Levenstein & Company, Ltd. · Rule file kate_algo_v1_4.py fingerprint cdaa9514…452d0 · Mockup built 28 Oct 2026 · Numbers illustrative except Section 3 outer rows