NOSTRADAMUS · Position Analytics Engine
SIMULATOR Will Anthropic have the best AI model at the end of June 2026?
← Back to live dashboardEmbed cardOG previewTop moversArb
A live, interactive instrument for dissecting a single binary position. Sweep the inputs and watch every indicator recompute — payoff geometry, Kelly growth, Bayesian posterior, KL divergence, cost waterfall, Monte-Carlo equity fan, forecast calibration. Companion to the live /feed/pm-will-anthropic-have-the-best-ai-model-at-the-end-of-june-2026 page.
▲ YES EDGE · +0.002 · f★ 2.0% · deploy 1.0% · net -0.52pp
§1 · Position economics
YES · Expected P/L per share +0.0023@ model P(YES) = 0.887
P/L per sharemarket pricemodel Pprofit zoneloss zone
Profit is linear in the eventual settlement price.
f★ = 2.00% · g(f★) = 0.003%deploy 1.00% · g = 0.002%
g(f)f★ optimumdeployed fgrowth zone
Underbet leaves growth on the table; overbet destroys capital. The interior maximum is f★.
§2 · The trade ticket
YES @ 0.884 · EV +$1stake $250 · 1.00% of bankroll
Deployed stakestake
$250
1.00% of bankroll
Sharesunits
282
each pays $1 if YES
Max payoutwin
$282
gross, if win
Max profitwin
+$33
net of cost
Max losslose
-$250
binary settles to $0
Payout multiple×
×1.13
$1 → $1.13
Risk:RewardR:R
0.13 : 1
win $0.13 per $1
Expected P/LE[P/L]
+$1
probability-weighted
| Outcome | P(model) | P/L | Contribution |
|---|---|---|---|
| Resolves YES (win) | 88.7% | +$33 | +$29 |
| Resolves against (lose) | 11.3% | -$250 | -$28 |
| Expected value | 100.0% | — | +$1 |
What you actually win and lose. The bottom table tabulates probability-weighted P/L by outcome.
§3 · Break-even & cushion
Cushion +0.2 pprelative edge +0.3%
Required win ratebreak-even
88.4%
price = implied probability
Model win rateP(win)
88.7%
what you forecast
Cushionedge
+0.2 pp
margin of safety
Fair pricemodel
0.887
where you think it should trade
The market price equals the win rate you must beat to make money.
§4 · Odds conversion
Implied probabilityP
88.4%
= price
Decimal oddsEU
1.131
total return per $1
AmericanUS
-766
risk $766 to win $100
FractionalUK
0.13 / 1
profit per $1 risked
Profit per $100stake
+$13.06
clean dollar framing
underdog (+)favorite (-)your price
Five views of the same number.
§4b · Time & annualized return
APR 5% · APY 5%ROI 0.3% over 21d · 17.4 turns/yr
Time to resolvehorizon
21.0 d
504h capital lockup
Raw ROIper resolve
+0.3%
APR (simple)scaled
+5%
ROI × 365/days
APY (compounded)if redeployed
+5%
(1+ROI)^(365/d) − 1
Daily expectedper day
+0.01%
geometric, per day held
Capital turns/yrvelocity
×17.4
how often this slot recycles
simple APRcompounded APYyour horizon
Rank positions by APR, not raw ROI. A thin edge tomorrow beats a fat edge next year.
§5 · Costs & net edge
Net edge -0.52 pperosion 325% · break-even w/ fees 89.2%
gross edgefrictionnet edgefee 0 bps · spread 1.50¢
The number that decides whether to trade.
§6 · Sizing menu
Full Kellyf★
$500
2.00% · g = 0.003%
Half Kelly½ f★
$250
1.00% · g = 0.002%
Quarter Kelly¼ f★
$125
0.50% · g = 0.001%
Flat 1%1%
$250
1.00% · g = 0.002%
Flat 2%2%
$500
2.00% · g = 0.003%
Flat 5%5%
$1,250
5.00% · g = -0.003%
Recommended¼ f★
$125
survives model error
Quarter-Kelly is the industry default — survives model error far better than full Kelly.
§7 · Information theory
Market entropyH(p)
0.516 bit
max 1.0 at p = 0.5
Your entropyH(q)
0.509 bit
Δ -0.007 bit vs market
Surprise · YES−log₂ p
0.18 bit
self-information
Surprise · NO−log₂(1−p)
3.11 bit
self-information
H(p) peaks at p = 0.5 (one bit of irreducible doubt).
NOISE · D_KL(q ‖ p) = 0.0000 nat (0.0000 bit)belief ≈ market — stand down
YES contributionNO contributionbelief ‖ marketnoise
Zero KL ⇒ you know nothing the crowd doesn't.
§8 · Bayesian inference
MARKET PRICE INSIDE 95% CIposterior μ 0.887 · CI [0.75, 0.97] · κ 26.9
Posterior meanE[θ]
0.887
Beta(23.8, 3.0)
95% credible intervalHDI
[0.75, 0.97]
price INSIDE → weak edge
Concentrationκ
26.9
pseudo-obs behind belief
Disagreementvs crowd
+0.2 pp
posterior − price
market prior (dashed)model posterior95% credible bandmarket price
When the market price falls outside the 95% credible interval, your disagreement is statistically meaningful.
§9 · Tail risk · Monte-Carlo (mode A · single position to resolution)
E[P/L] +2.0% · P(YES) 90.3% · VaR₉₅ 100.0%400 paths · 504 bars to resolution
Expected P/Lper $1
+2.04%
P(YES) empiricalq
90.3%
Best pathmax
+13.1%
Worst pathmin
-100.0%
VaR 95%5%
100.0%
CVaR 95%ES
100.0%
median path25/75 + 5/95 bandsentry pricemodel q
Logit-space mean-reverting walk + terminal flip with probability q. Answers: 'what happens to THIS one position'. Distinct from the repeated-edge fan below.
§9b · Tail risk · Monte-Carlo (mode B · repeated independent edges)
Median CAGR/bet -0.00% · ruin rate 0.0%400 paths × 120 bets · f deploy 1.00%
Sharpe / betμ/σ
0.008
μ 0.00% · σ 0.4%
Sortino / betμ/σ↓
0.003
downside-only denominator
VaR 95%5%
-1.0%
per-bet worst-case
CVaR 95%ES
-1.0%
mean tail loss
Max drawdownMDD
-1.4%
Calmar -0.00
Ruin rate≤50%
0.0%
P(equity ever ≤ 50%)
median25/75 band5/95 bandruin line
Answers a different question: 'if I could find this exact edge forever, what is the bankroll trajectory'. Compounds 120 sequential resolutions which is NOT what happens to a single position.
§10 · Base-rate & macro context
ANCHORED · supported by convictionanchor gap +44.5pp · crowd gap +44.2pp
Anchor gapmodel − base
+44.5 pp
Crowd gapprice − base
+44.2 pp
Verdictdiscipline
ANCHORED
Reference-class anchoring prevents narrative-driven blowups.
§11 · Forecast quality (synthetic ledger)
SKILL POSITIVE · in-sample BSS 20.5% · AUC 0.768out-of-sample BSS (5-fold) 20.5% ± 2.7% · Brier 0.1988 · log-loss 0.5946 · n 1600✓ n = 1600
BrierBS
0.1988
lower = better · ō 0.51
BSSvs base
20.5%
improvement over base rate
ReliabilityREL
0.0042
miscalibration · want ↓
ResolutionRES
0.0556
decisiveness · want ↑
Log lossLL
0.5946
cross-entropy
AUCROC
0.768
0.5 coin · 1.0 oracle
calibration curveROCUNC (irreducible)RES (skill, ↑)REL (miscalib, ↓)
Computed on a seeded synthetic forecast ledger. Reseed (⟳) to redraw.
§12 · Journal vitals (synthetic ledger)
PROFITABLE · PF 1.06 · expectancy +0.027R180 trades · win 50.6% · Sharpe 0.024
Total P/Lnet
+$1,226
on $45,000 cycled
Win ratehit %
50.6%
91 W / 89 L
Profit factorPF
1.06
$ won / $ lost
Expectancyper trade
+$6.81
avg $ per position
R-expectancyper risk
+0.027R
in units of risk taken
Avg win / losspayoff
$257.97 / -$250.00
ratio 1.03 : 1
Sharpe / traderisk-adj
0.024
μR / σR
Closing line valueCLV
+3.36 pp
avg edge vs close
cumulative P/Lprofitable zonered zonesynthetic · seeded from asset
The scorecard every trader checks. Synthetic ledger seeded from the asset slug — recomputes against your real fill history once wired.