polyjev

Live · BTC five-minute market

mode orders

Current window

closes in
--:--
slug–
market id–
entry window–
entry cutoff – before close
UP token–
DOWN token–
fee schedule–
tick · min size–
liquidity · volume–
gamma outcome px–
Settles UP when twap60(close) ≥ twap60(open). The strike is the 60s TWAP over the minute before the window, so it is fixed before trading starts and is read, never estimated.

Recent windows

UTC · outcome vs model at open

Settlement series vs strike

twap60 now
–
strike · twap60 at open
–
reference (spot)
–
freeze gap
–
•WAITING
$ · right axis
¢ / % · left axis
time
What each trace is, and where it comes from
twap60
the series the market settles on (Chainlink crypto_prices_twap_sixty), shaded green above the strike and red below.
reference (spot)
the ~1/s Chainlink reference that drives the freeze projection; dashed where it is rebuilt from the trader's journal trace (this app started mid-window).
strike
twap60 at the window's open, read from the feed or the journal, never estimated.
freeze projection
where the settlement average lands if the reference stops moving now (strike + freeze gap), ticked on the right axis.
market implied level
the settlement level the books' price implies: strike + Φ⁻¹(P) × sd of the settle average.
both sides priced
UP pays only while the freeze projection is above the green line (strike + Φ⁻¹(UP ask + fee) × sd); DOWN only while it is below the red one; between them neither side has an edge.
UP / DOWN ask
each book's best ask, in cents; fee not included.
model P(UP)
P(settles UP), this app's own model, the number the Settlement model box shows: a point about every 1.2 s while this page has a viewer, and this tick's value at the line's end. Where no viewer was here to record it (before this app started or first had one, or while every viewer was away), one point per journalled judgment, drawn faded: the trader's own P(UP), or the closed form with that judgment's gap and σ. That σ is the trader's, not this app's, so the line can step where the faded stretch meets the live one, most right after a restart, when this app's σ rests on minutes of reference and not its 30.
market P(UP)
the books' P(UP): the UP mid and 1 − the DOWN mid, averaged (a lone bid or ask where a book is one-sided); from the trader's journalled books where faded.
model − market
the region between those two lines, hatched: green where the model is above the market, red where it is below; the label is this tick's gap in percentage points, the box's "model − market".
settlement window
(T+237, T+297], the 60 seconds that enter the settlement average.
entry cutoff
the trader enters nothing after this, its ENTRY_CUTOFF_SECONDS before the close.

The two axes are independent scales: the left reads 0–100 (cents for an ask, percent for a probability - a 40¢ ask prices UP at 40%), the right reads dollars. Where a left-axis line crosses a right-axis line means nothing. Every trace from the trail (all but twap60, the reference, the strike, the freeze projection and the time markers) is faded where it was rebuilt from the trader's journal, one point per judgment, with a "journal backfill" caption. The buttons switch traces on and off; this browser remembers the choice.

per judgment · 0–1
One point per journalled judgment: the model's P(UP) against what the books implied at that moment. The gap between them is the trade.

Strategy recommendations

connecting…

paper onlyLoading the evidence…
Each chosen strategy's call right now, with its evidence verdict first, then its reason
StrategyEvidenceMarketStatusSide Prob.Price Model edge Stake WhyAge

What the statuses mean, and where each number comes from

Each row is read from the trader's journal by this app every two seconds while this page is open: an in-window strategy from its newest evaluation row (and paper fill) for the current market; a pre-open strategy from its decision for the market that just opened, with its provisional call for the next one underneath; a rule the trader does not run from its replay rule on the newest pre-open sample or judgment. Price is per share with the taker fee. Model edge is the strategy's own probability minus that price, net of fees (the router's is q's lower bound minus it): what the model claims, not evidence that it wins. A pattern rule claims no probability, so it has no edge. A stake shows only on a bet; elsewhere it is marked "would". The evidence is each strategy's headline on Strategies. Your choice of strategies is kept in this browser.

%
pp
sd
The same trail as the main chart, one panel and one scale per measure: a point about every 1.2 s while this page has a viewer, and this tick's values at each line's end, so the numbers beside the chart are the ones above it. Where no viewer was here to record it, one point per journalled judgment, faded: the trader's own P(UP) and z, from its own σ, not this app's. z starts off: P(UP) = Φ(z), so it repeats the model's line in other units, and it runs off toward ±∞ in the settlement minute as the sd shrinks to 0 (its panel is capped at ±6).

Both sides priced

Every cycle prices both tokens against their own book and their own conservative q, then takes the better net edge — buying the favoured side is not the same trade as buying the underpriced one. Gate order mirrors src/risk.js; portfolio and daily-loss gates live in the trader's ledger and are not re-checked here.

Order books · two independent tokens

UP and DOWN are separate CLOB books, not two sides of one book. A winning token pays 1, a loser 0. VWAP(20) walks the asks for the configured order size; all-in adds the taker fee.

Calibration grid

columns = model P(UP) bin · rows = minutes remaining predicted realized
A judgment at T-280s and one at T-40s answer different questions, so horizon is a cell key. A cell may be traded only once it holds 30 resolved markets; the code then uses the conservative 95% Wilson bound, never the model's raw p.

Session

Journal

READ-ONLY VIEW · places no orders · never reads or transmits the signing key