CURVE TRADER
Treasury Spread Signal Engine
Databento —
Overnight
00:00:00
WARMING UP
NOB · 10Y vs 30Y
◆
—
Awaiting first signal…
Live verdict from the signal engine — the age chip in the footer shows how fresh.
Kyle — would you take this trade?
CLAUDE — —
COUNCIL — —
PRE-TRADE INSPECTION
8 things a trader checks before pulling the trigger
—
LIVE BOOK · TAPE FLOW
microstructure for the two legs
LONG ZN
BID
—
×—
ASK
—
×—
WIDTH
—
DAY
—
BUY VOL 60s
—
SELL VOL 60s
—
TRADES
—
AGGRESSOR
—
SHORT ZB
BID
—
×—
ASK
—
×—
WIDTH
—
DAY
—
BUY VOL 60s
—
SELL VOL 60s
—
TRADES
—
AGGRESSOR
—
STATISTICS · MACRO CONTEXT
setup quality + environment
Z-SCORE (60d)
—
—
OU HALF-LIFE
—
days to mean
STOCH K
—
—
60s Δ
—
spread ticks
5m Δ
—
last 5 min
15m Δ
—
last 15 min
DAY OPEN
—
spread ticks
DAY HIGH
—
DAY LOW
—
DAY RANGE
—
MOVE INDEX
—
—
REGIME
—
classification
CROSS-SPREAD CONTEXT
are correlated spreads confirming or diverging?
—
Awaiting evaluation…
NOB
10Y vs 30Y · ratio 2:1
ON THE DAY
—
IMPLIED
—
Z-SCORE
—
LONG · ZN
—
— × —
SHORT · ZB
—
— × —
Win rate · real trades (net)
—
— signals tracked
Wins
0
Losses
0
Best persona
—
—
Regime
—
MOVE —
CLAUDE
awaiting…
—
Direction
—
Confidence
—
COUNCIL
awaiting…
—
Council votes will populate here.
ML MODELS
XGBoost ensemble
XGBoost direction
—
XGBoost confidence
—
Raw signal
—
MARKET CONTEXT
macro + microstructure
Regime
—
MOVE
—
VIX
—
Stoch K
—
PERSONA WEIGHTS · learning loop
off — observing only until track record builds
PASSIVE
Persona weights will appear here once 10+ signals decide per persona.
AUTORESEARCH · self-tuning
Karpathy-style param mutator · evaluates & auto-reverts
Autoresearch will propose its first mutation after 15+ baseline signals.
LESSONS LEARNED · self-review
Claude analyzes its own wins/losses every hour
Lessons will appear after the first hourly self-review (or click Run Now).
SYSTEM WATCHDOG
checks pipeline health every poll
—
Loading…
Signals decided
0
All-signal rate (mostly untraded)
—
HIGH-confluence signals (scored, not trades)
0
HIGH-confluence hit rate
—
Traded W / L
0/0
MULTI-HORIZON BREAKDOWN
Same signals, scored at 4 different windows. Trader-style (TP/SL) is the most realistic.
KYLE'S VALIDATION
your YES/NO becomes the ground-truth label for retraining
No votes yet — press YES or NO on the verdict panel to start building ground truth.
PERSONA TRACK RECORD
hit rate when each persona issued a directional call
Persona stats populate after a few signals decide.
SIGNAL HISTORY
live · all signals
No signals yet.
THE TRADE LEDGER · EVERY BOT TRADE
every trade the PRIMARY strategy takes (record starts 7/21) · entered at the signal price, exited at its target/stop — or force-flat daily at 4:30 PM ET (FOMC 1:45 PM, jobs-report days 8:10 AM; Kyle's 8/6 rule) · executing TENB (z≥1.5) — the one config fold-stable at true intraday granularity (8/7 Studies D+F, 1-min replay) — other spreads signal, score and ride the ¼-size parallel track below · real $, net of costs
Building the bot's trade ledger…
P&L BY CURVE
net after fees+slippage · all-time + today, per spread
PARALLEL TRACK · LONG-HOLD (¼ SIZE, PAPER)
the multi-day strategy Kyle benched on 8/6, still measured on the same entries at quarter size (21-day window, no daily flat) · tracking only — no texts, not in the August goal
No parallel trades yet — the long-hold track books the next reversion entry the primary takes (started 8/6).