Scanner signals · five bot streams · one portfolio
System · expectancy
—
no closed signals yet
You · expectancy
—
no closed trades yet
Portfolio at a glance · Ultra Combo, last 1 year
loading…
Bot fleet
loading…
Mechanical breakdown
Side
n
Win%
Exp R
PF
MFE
MAE
T1
T2
T3
Policy: terminal exit at final target or stop. MFE/MAE and target-touch rates are tracked independently so you can model other exit styles.
⚠ These stats cover live scanner signals only — a small, recent sample (a hot streak or cold streak moves Win% a lot; quick winners also close before slow losers, flattering early numbers). The long-run record — 240 days, 200+ trades, all market regimes — lives on the Wunder 1+2 bot page under Strategy profiles; expect these numbers to drift toward those as trades accumulate.
Coil Watch — screener (feeds the Trade Watcher)
🌀 Coiled Spring Screener
What this screens for, and what a candidate is (and isn't)
Every 4 hours the whole Bitget USDT-perp universe (~740 symbols) is screened for the pre-expansion coil: a prior big run (≥1.7×), a deep crash (≥20% off the peak), a base that held (didn't give the run back), and — the core of it — a currently compressed structure: the six moving averages (SMA20/50/100/200, EMA21/55) converged to their tightest recent spread, range contracted, volume dried up, with price sitting on the MA cluster (−4%…+8%), not 30% under it. That last gate is what separates a coil from a falling knife. A candidate is a watch signal, never an entry — each one auto-enrolls into the Trade Watcher, and the entry alert (⭐ PRIME) only comes from there once a real reaction prints. Fuel score ranks candidates (funding 0–40, compression 0–25, contraction 0–20, base maturity 0–15); funding was neutral on both validation cases (BEAT +32.75%, ESPORTS +7.2%), so it's a bonus, not a requirement. Phase gate applies: no edge is claimed until ≥150 tracked signals; the fire-rate/expired split below is the honest scoreboard.
loading…
👁 Trade Watcher
The Trade Watcher has its own page — a live terminal with one channel per watched coin, the minute-by-minute play-by-play, and a standing long/short verdict for whichever coin you're viewing. CSS candidates still auto-enroll there.
👁 Trade Watcher — live tape terminal
How the Trade Watcher works — states, events, and how to read the verdict pill
A per-symbol live monitor (1-minute cadence) over a curated watchlist: CSS coil candidates, 🚀 TIER-1 pre-breakout flags and Radar arms auto-enroll, and you can add any coin manually (manual watches are never auto-evicted; watching an already-auto coin promotes it to manual). Telegram carries exactly the calls: ⭐ PRIME entries and exits (BASE_BREAK, paper closes) always ping, uncapped — the setup play-by-play lives here in the terminal and only goes to your phone if the Play-by-play toggle is on. Pick a coin in the channel list — the terminal shows that coin's play-by-play and the verdict bar keeps a standing answer to "long or short, right now": LONG when a position is on or the structure confirmed, LONG SETUP while a reaction is printing and the ⭐ PRIME checklist is running, NEUTRAL when nothing qualifies, and BROKEN after a BASE_BREAK — a 1H close through invalidation means sellers have the structure; that is a do-not-long (and the only short-side statement this module is allowed to make — it watches long structures, it has no short entry engine). Each symbol carries a rung ladder (auto-derived from the same 4H/1D pivots as the Radar) and moves through DORMANT → APPROACHING → AT_LEVEL → REACTION → CONFIRMED → TRACKING. The feed narrates only qualifying tape — sweeps, volume reclaims, stop-runs, absorption, gate clears, ignitions — never oscillator states or bare touches, so silence is information: a quiet channel means nothing tradeable has printed. Weak-volume "drift" reclaims are flagged, a third failed reclaim downgrades the coin to "range until proven", and BASE_BREAK always posts at full severity and closes any tracked position. ⭐ PRIME = location + reaction + confirmation close + RVOL ≥1.2× + risk math (stop ≤2.5×ATR, RR ≥1.5) on the same evaluation; PRIME auto-arms a simulated paper position (33/33/34 ladder, BE after t1, 72h cap) and is the entry layer for the Radar v1.5 Confirmed Runner. STARTER = location bet without the volume — simulator only. Honest degradations: no long/short positioning-book feed exists in this stack, so crowd checks report "book n/a" and CROWD_FLIP is off; the 1-minute cron is the floor, so the terminal refreshes itself every 30s but new lines can only appear once a minute. Paper only — no live execution until ≥30 closed signals show positive expectancy.
channels
trade-watcher — connecting…
Signal book — every ⭐ PRIME paper position and (sim) STARTER, all coins; the selected channel sorts first
Security watchdog
What this watches, and what it does when it finds something
Added 2026-08-05 after the site compromise. Every 10 minutes the scanner checks five things and pushes a Telegram alert on anything serious. File integrity: every executable file under the web root (PHP/INI/JS/SH/.htaccess) is hashed; new or modified files raise an alert, and files in uploads/ — where webshells usually land — are treated as higher severity. Malware signatures: every file is content-scanned each cycle for the exact patterns this attacker used (cookie-callable backdoors, the nx payload family, wp2shell, PwnKit, layered eval obfuscation, gsocket, process-name spoofing). A signature hit keeps re-alerting every 6 hours until the file is gone, so an unhandled infection can never go quiet. WordPress: the companion hardening plugin reports the administrator roster, role changes, plugin activations, failed-login bursts and admin logins — a new admin account alerts immediately, which is exactly how the last break-in persisted. Crontab: any change to the site user's cron is critical (the attacker used a 5-minute job to re-drop a webshell). Processes: anything running as the site user disguised as a bracketed kernel thread, or carrying gsocket markers — that's how the 15-day implant hid. The watchdog only ever observes and alerts; it never deletes or modifies anything, because remediation should be your decision, not a script's.
loading…
Legend
What the abbreviations and numbers mean — tap to expand
Core unit
R
One unit of risk — the distance from entry to stop. If you risk $100 on a trade, +2R means you made $200 and −1R means you lost the full $100 (your stop was hit). Every result here is measured in R so trades of different sizes and prices are comparable.
Performance stats
n
Number of closed trades the stat is based on. Small n = don't trust the number much yet.
Win%
Share of closed trades that ended in profit.
Exp R
Expectancy — the average result per trade in R. +0.30R means that over many trades you make 0.30× your risk per trade on average. Positive = profitable system.
PF
Profit factor — total R won ÷ total R lost. Above 1.0 is profitable; 2.0 means winners paid twice what losers cost.
MFE
Max Favorable Excursion — the furthest the average trade moved in your favor before it closed. If MFE is much higher than Exp R, exits are leaving profit on the table.
MAE
Max Adverse Excursion — the average worst drawdown a trade suffered before closing. Shows how much heat trades take; near −1R means many trades nearly stop out.
T1 T2 T3
The three take-profit targets. In the stats table, the percentage is how many trades reached that target. On a signal card, ✓ marks targets already hit.
Max / Min
On an open trade: the best (Max) and worst (Min) the trade has been so far, in R.
capture
Capture efficiency — how much of the profit that was available (MFE) the exits actually banked.
Lift
In Condition edge: how much a condition raises the win rate. Win% when the condition fired minus win% when it didn't — positive = that condition helps.
Signals
score /6
Confluence score — how many of the 6 indicator checks agreed (TEMA200 trend, SMA200 slope, TEMA cross, EMA21, StochRSI, RSI divergence). Minimum 4 to fire a signal.
open
Trade is still running — none of the exits below have triggered yet.
target
Closed at the final take-profit target (T3). Best outcome.
stopped
Closed at the stop-loss for −1R.
stoch
Closed early by the dynamic exit: the trade was in profit and the StochRSI turned back from an extreme (overbought for longs, oversold for shorts), so the system banked the profit rather than risk a reversal.
be
Breakeven runner exit (only under the "50% at T1 + breakeven runner" policy): half the position banked profit at T1, the rest was stopped out at entry — a small net win.
Strategy intelligence
Test
Backtests the recommendation against your recorded signal history — entry-side changes filter the log, exit-side changes are replayed bar-by-bar against real candles — and shows current vs modified results before you commit.
Apply
Writes the change into the live engine config: the 15-minute scans and the open-trade tracker start using it immediately. Revert any time from the "Live engine config" panel.
Paper mode
Risk %
How much of the current balance is put at risk on each trade. 1% of $10,000 = $100 lost if the stop is hit.
Position $
The notional size of the trade — risk amount ÷ stop distance. A tight stop means a bigger position for the same risk.
Margin $
The cash set aside to hold the position: position ÷ leverage.
LIQ
Liquidated — the trade moved against the position far enough to wipe the margin before its normal exit could trigger. Higher leverage = liquidation sits closer to entry.
◂cap
The position was capped by the leverage limit (balance × leverage), so the trade risked less than the full risk %.
Cluster cap
When several signals fire on the same 4h candle in the same direction (e.g. XRP + LINK + LTC all short), they're really one correlated bet — historically the biggest losing days were exactly these clusters. In real-world mode the cap works exactly like the live bot: entries beyond the cap are skipped. In ideal mode it scales sizes down instead so the cluster shares a combined risk budget. Empty = off.
Real-world mode
Available on both Paper mode and the Wunder pairs paper replay. Replays the history the way a live bot would actually have traded it: positions overlap in time and share the account's margin (live trades use their real open/close timestamps; replay rows hold an estimated number of days), each entry is sized WunderTrading-style as a % of the free balance × leverage, Blofin taker fees are charged both ways plus estimated funding while held, a max-position liquidity cap models what the order book can absorb, isolated-margin losses are capped at the margin, positions liquidate (⚡LIQ, full margin lost) when the worst adverse move crosses the leverage's liquidation distance (~95%/leverage) before the stop, and cluster-cap entries are skipped like the live bot does. Untick it for the frictionless ideal replay — good for comparing engine logic, not for forecasting dollars. Still not modeled: price gaps through stops, and the fact that these pairs were picked from this same period (in-sample bias) — treat even real-world numbers as the optimistic end of honest.
Loss diagnosis
Gave back
The trade reached at least T1 but then reversed all the way to the stop.
No follow-through
The trade never got going — chopped around and eventually stopped out.
Bad entry
Price moved against the trade immediately after entry.
Strategy intelligence
loading…
Bot intelligence
Ongoing review of the bot layer — strategy profile choice, cluster cap, pair health — over the full history and the trailing 90/30 days where drift shows first. Recomputed on every load as trades close; a daily 08:15 UTC review pushes any new recommendation to Telegram. Apply buttons write straight to the bot config (reversible from the bot cards).
loading…
🧠 Radar Intelligence — what the streams are telling us
How this avoids being a p-hacking machine
A tool that slices books until something looks good will always find something — that is the danger, not the feature. Three guards are built in. Every candidate runs the house stress battery (winners−40%, fees×2, drop-best-5) before it may be called a finding — the same bar a human idea faces here. Ideas already killed are cross-referenced against the DO-NOT-RE-TEST list and returned as dead ends with the date they died, so the same ground is not walked twice. Only entry-time features can become recommendations: you can gate on score or volatility because you see them before committing, but hold time and MFE are outcomes — "trades held longer did better" is survivorship (winners run for days, losers stop out in hours), so those are reported as context and can never be advice. The multiple-comparisons count is printed openly, and nothing is ever applied automatically: a finding is a hypothesis that still needs split-half validation and the full battery as a parallel A/B cohort.
loading…
Paper mode
Starting balance $
Position size % of balance
Leverage ×
Strategy
Timeframe
Fee + slippage per side %
Est. funding per trade %
Max position $ (liquidity cap)
Margin mode
loading…
Replays every closed scanner signal, compounding the balance. Real-world mode: positions overlap in time using their real open/close timestamps and share the account's margin, each sized as % of free balance × leverage; losses are capped at the isolated margin; liquidation (⚡LIQ) hits when the trade's worst drawdown crosses the leverage's liquidation threshold. Ideal mode: risk-based sizing (risk ÷ stop distance), no costs, trades one at a time.
Open · live trade plans
Log a taken trade
Coin
Side
Entry
Stop
Margin $ (opt)
Leverage × (opt)
Notes (opt)
Your trade log · every trade you've logged, open and closed
loading…
Recent signals
Strategy filter
Trade chart
Click any dot on the equity curve to open that trade's candle chart. Entry, stop, targets and the actual exit are drawn on the price action.
Live chart
live from DexScreener — streams only while this window is open, no scanner API usage
Pairs & replay · Wunder bots 1+2 on paper
What the two WunderTrading bots would have made on paper. WunderTrading caps a bot at 10 pairs, so Bot 1 carries pairs 1–10 and Bot 2 pairs 11–20; signals route to whichever list the coin is on. Pick a preset (ranked best-first from the replay + live dry-run rows), set your balance, read the result. Real-world costs are on by default — fees, funding, order-book depth, isolated-margin liquidation.
Per-coin signal stats
Paper mode — pick a preset; everything else is optional
Preset · best results first
Bot
Timeframe
Starting balance $
Deploy selected strategy
Advanced — strategy, exit, sizing & costs
Exit
Strategy
Position size % of balance
Leverage ×
Cluster cap (empty = off)
Fee + slippage per side %
Est. funding per trade %
Max position $ (liquidity cap)
Est. hold days (replay rows)
Margin mode
Strategy scoreboard — every profile on this selection
Live bot trades — positions actually sent to WunderTradingSignal history
Live cluster cap
When several pairs signal in the same direction on the same 4h candle, they're one correlated bet — the replay's worst days were 3–5 of them stopping together. This cap limits how many same-direction entries are sent to WunderTrading per 4h candle, across both bots. Extra signals are skipped and logged in Recent alerts on the Portfolio page (marked ✗ skipped: cluster cap). The scanner itself still tracks every signal, so the history above stays complete.
Max same-direction entries per 4h candle (empty = off, original behaviour)
Adaptive risk layer
Four rules that make the bots trade like a seasoned trader instead of a fixed machine — stand down when cold, trade fewer correlated bets after a rough stretch, never short a face-ripping bounce, skip shorts in chop. Each rule only ever reduces exposure and never touches exits. Validated on the full 3-year replay: Apex Refined v1.1 went from +154.5R to +167.6R with max drawdown cut from 27.4R to 10.1R, and every BTC regime (bull/bear × high/low-vol × trend/chop) came out positive. Skips appear in the alert log with the reason.
Ultra Combo · all five validated streams, one portfolio
Every validated stream — Wunder 1 & 2, TEMA, Daytrade shorts, Cipher Refined — combined as one portfolio with drawdown-parity risk weights.
How the portfolio is built — dd-parity weights, uncorrelated streams, risk units
Signals are never merged or filtered by each other (that's how edges die — see the TEMA×apex study); instead each stream gets a drawdown-parity risk weight: per-trade risk proportional to 1 ÷ the stream's own max drawdown in the window, so every stream contributes a similar worst-case dent. Weekly returns across the streams are near-zero or negatively correlated, which is why the combined curve is smoother than any component. Units: 1.00u = one base-risk unit — set the base risk % below to translate to balance terms. The curve is R-based (frictionless); for real-world dollars (margin overlap, fees, depth, liquidation) the Combo replay below prices the Wunder+TEMA core — expect a comparable haircut here. Coil Bias and TA Supreme retest stay OUT until they graduate their phase gates.
Window
Base risk % per 1u
loading…
Ultra Combo v2 · the current-version portfolio
v2 rebuilds this portfolio from today's bots — one Wunder Thunder leg, a Market Maker v1.5 leg, TEMA, Daytrade shorts and Cipher — with correlation-adjusted weights. It has its own page: paper account from a starting balance, the five bots with what each risks, and the step-by-step WunderTrading setup.
🧩 Ultra Combo v2 · five bots, one account
Five of this site's bots run side by side on one account. Each bot keeps its own signals; the portfolio only decides how much of the balance each bot risks per trade. Legs: ⚡ Wunder Thunder (WunderTrading bots 1 & 2), 🧲 Market Maker v1.5, TEMA Trend (bot 4), Daytrade 1H shorts (bot 3) and Cipher Refined (bot 5). Weights are drawdown-parity divided by how much a leg moves with the others, so nothing is counted twice. The v1 portfolio stays on the Portfolio page for comparison.
How to read this page
① Pick a starting balance and a risk per unit — the paper account below shows what the last year would have done with exactly those numbers. ② The bots table converts that into dollars at risk per trade for each bot, which is the one number you type into WunderTrading. ③ The step-by-step section walks the WunderTrading setup bot by bot, with each bot's current state (dry-run or live). Everything is replay + live paper; the “real-world haircut” counts a third of every result because live books here have run about 3× below their replays.
📒 Paper account · start from a balance
Starting balance $
Risk % per 1u
Real-world haircut
Legs
Window
loading…
🤖 The five bots · what each one risks at your balance
loading…
🛠 Step by step · setting it up in WunderTrading
loading…
👥 Customers — accounts on bot.coinrankz.com: plan, approval, activity
loading…
Customer
Plan
Status
Setup
Last seen
Paper R
Notes
Telegram pings on every new sign-up and every "I've followed the bot" press. Passwords issued here are shown once and never stored in the clear.
💳 Billing — the shared Stripe account: subscriptions, revenue, invoices
What it does. When the scanner's live Thunder evaluation opens an XAUT trade, the bot buys the Kalshi gold daily contract that settles at the second 5 PM New York close after the signal — YES on the strike nearest 50¢ for a long, NO for a short — as a taker at the ask, only if the ask is at or under the max price. It never exits early: the contract pays $1 or $0 at 5 PM. The next-5pm contract is logged as a shadow (58.7% in the study, not worth the fee). Every signal, fill and settlement is relayed to Telegram.
Why gold, why two days, what the numbers rest on
Thunder's signal direction at a fixed 48-hour horizon held on all 126 bar-covered union trades (70.6%, Wilson lower bound 62%, halves 68/73, longs and shorts); XAUT was 13/15. Buying the at-the-money contract at 52¢ is worth about +15¢ per contract at 70%, +6–8¢ at the lower bound; Kalshi's breakeven at 52¢ is 53.8%. Gold is the only underlying with a liquid daily ladder (KXGOLDD, ~$100–250k a day; crypto daily series are dead, hourly only). The signal cohort is the Thunder union with no ADX/RSI floors — the same cohort the study measured — so a row the deployed Wunder gate drops (e.g. RSI under 40) is still a gold signal here. Caveats: 15 gold replays + a handful of live joins; ~15 signals a year, so the judge (20 settled) takes most of a year; the second-5pm ladder is thin at signal time, so a real fill may need a resting order. Rehearsed dry on a 40-strike ladder on 2026-09-02.
🔑 Go-live checklist · every step, checked live
loading…
🎛 Knobs · applied by the bot within a minute
loading…
📡 Signals · every live Thunder XAUT row, and what the bot did with it
loading…
📒 Book · open contracts, settled contracts, the shadow record
loading…
🪜 Kalshi ladders right now · what a signal this minute would buy
loading…
🌤 Weather Maker — two-sided paper quotes on 7 daily-high-temperature ladders · NOTHING is sent to Kalshi
loading…
What it does. Four Kalshi studies on 2026-09-02 (BTC 15-minute, weather, MLB, a +EV replica) all found Kalshi's price to be the best forecast: taker expected value ≈ 0 after fees. What is left is the maker's side — earn the spread and the liquidity rebate, lose to adverse selection. This daemon quotes fair ± 2¢ on both sides of the seven daily-high ladders (NYC, Chicago, Miami, Austin, Denver, LA, Philadelphia), fair = a small model weight on top of the mid with hard observation constraints, and fills itself against the real trade tape: a print through our price is a certain fill ("strict"), a print at our price is counted separately (queue position unknown). Marks at +5 and +30 minutes measure adverse selection. Positions settle on Kalshi's own result.
The backtest behind the current settings, and the judge
On the 68-day tape (2,856 markets, strict fills only) the naive quoter lost in every configuration out of sample: −1.4 to −2.9¢ per contract, with the +30-minute mark −2.2¢ everywhere — adverse selection beats the 2¢ half-spread. The one positive cell was same-day markets before 7:30 local (+2.3 to +5.2¢ in every split) and day-ahead quoting was ≈ −0.2 to −1.8¢, so the paper book now quotes only day-ahead and the early-morning window (config quote_local_hours 0–7). Judge after ~2 weeks: strict-fill P&L per contract ≥ 0 after fees AND the +30-minute mark not systematically negative AND a high two-sided share (the rebate programme wants both sides ≥100 contracts). If strict P&L stays negative the edge does not exist and the bot stops; if it passes, the next step is tiny live size through the same order module the gold bot uses.
📋 Quotes now · open positions · settled
loading…
🧪 Kalshi Research · would any of our strategies work as a Kalshi bot?
Kalshi is a US-regulated prediction exchange: binary contracts that pay $1 if right and $0 if wrong. Its crypto contracts are direction bets over a fixed window (15-minute up/down, an hourly above/below ladder, daily events at 5pm New York). The taker fee is 7% × price × (1 − price), so a 50¢ contract needs a 51.75% hit rate to break even and a 52¢ ask needs 53.8%. That number is the whole question: a strategy either clears it in both halves of the history, or it is not a Kalshi bot.
verdict · no BTC rule clears the fee (only short-horizon RSI mean-reversion leans ~55%)the one live-test candidate → 🥇 Gold Bot (Thunder XAUT direction at 2 days, 70.6% n=126)🌤 Weather Maker · silent paper book
How to read this page
① Live board — what is tradeable on Kalshi right now, with a fair value per strike from realized volatility (zero-drift lognormal). Edge = fair − ask − fee; anything under ~2¢ is inside the noise of the volatility estimate. ② Strategy fit — every rule this site runs (Thunder's stoch/ADX/RSI gate, the RapidStoch reset, TEMA regime and cross, Cipher dots, the Daytrade short setup) plus plain momentum / mean-reversion baselines, transplanted onto BTC bars and asked one question at each Kalshi horizon: is the direction right more often than the fee-adjusted breakeven? Bets are non-overlapping (a bet occupies its window), outcomes use the exchange bar close, and the verdict needs the 95% interval clear of breakeven in both halves. Nothing is fitted here except the hour-of-day row, which trains on the first half and is scored on the second. ③ Paper account — pick any rule × horizon and see what compounding a fixed stake at the chosen ask would have done. ④ Kalshi's own record — its settled 15-minute markets: the base rate, the size of the moves, and how often the exchange bar agrees with the BRTI settlement (the settlement noise a bot eats on top of its hit rate). ⑤ Bot notes — what a Kalshi bot would need and the honest verdict.
📡 Live board · what is on Kalshi right now
Coin
loading…
🧪 Strategy fit · every site rule at every Kalshi horizon
Mid price ¢
Ask paid ¢
Horizon
Show
loading…
📒 Paper account · one rule, one horizon, compounding
Starting balance $
Stake % per bet
Window
click a rule in the table above
pick a rule × horizon above.
📊 Kalshi's own record · settled 15-minute markets
loading…
🛠 Bot notes · what a Kalshi bot would need, and the verdict
loading…
Combo Bot replay · both strategies as one account
The TEMA trend bot and the Wunder mean-reversion bots replayed as one account, priced the way the live account would actually trade.
How this replay works — natural hedges, risk-based sizing, real-world costs
The two bots are natural hedges: the TEMA trend bot earns riding major-coin trends (all of 2024: +139R while the scanner was flat), the Wunder mean-reversion bot earns on bear-regime alt shorts (2026 so far: +83R while TEMA longs are down). Their monthly returns are slightly negatively correlated, so one account funding both is smoother than either bot alone. Sizing here is risk-based: every trade risks the same % of current balance (P&L = risk × the trade's R) — the honest way to combine systems whose stops sit at different distances; the leverage inputs mirror what each bot runs in WunderTrading and drive the liquidation check. Validated 2026-07-09: running both bots whole beat every longs-from-one/shorts-from-the-other carve-out by ~100R at the same drawdown — the sides you'd cut earn exactly when the kept sides go flat. The side filters below let you check that yourself. Real world costs (on by default) replays the history the way the live account would actually have traded it: positions overlap in time and share the account's margin (entries downsize when free margin runs short), taker fees both ways, funding while held, ~0.05%/side alert→fill latency, depth slippage that grows with position size vs each coin's 24h volume, a book-depth cap on position size, stops that gapped past filling worse than −1R, the live cluster cap (max 2 same-direction Wunder entries per 4h candle) and adaptive risk layer (governor / streak stand-down / pump gate — same rules as the live bots, no hindsight), and isolated-margin liquidation — a trade whose worst adverse move crosses the leverage's liquidation distance (~95% ÷ leverage) before the stop loses its whole margin, even if it later recovered. TEMA trades already carry their fee/slippage/funding costs inside their R, so only latency, depth impact and liquidation are added there. Untick it for the frictionless ideal — good for comparing logic, not for forecasting dollars.
Starting balance $
Risk % per trade
Wunder profile
Wunder side
TEMA side
TEMA profile
TA Supreme
TAS lev ×
Wunder lev ×
TEMA lev ×
Timeframe
loading…
Portfolio Builder · pick your own bot combo
Tick any combination of bots and see what running them together as one account would have done — each stream from its live paper record, its backtest, or both stitched (full backtest, then live fills after it — no double-counting). Same real-world engine as the Combo replay above.
How to read this — sources, refining, and honest caveats
Source per stream: live is what the paper bot actually recorded in real time (small samples but zero hindsight), backtest is the validated replay (big samples, but the rules were tuned on that same data), stitched plays the full backtest and then continues with live fills entered after it. Risk × scales a stream's per-trade risk against the shared "Risk % per trade" (the Ultra Combo's dd-parity weights are a good starting point). Every trade risks balance × risk% × the stream's Risk ×; leverage only drives the liquidation check. Real-world costs work exactly like the Combo replay above: shared margin, taker fees + funding, depth slippage, stop gap-through, isolated liquidation, plus the Wunder side's cluster cap and adaptive layer. TEMA / Daytrade / Cipher / TA Supreme R values already carry their fees inside, so only latency + depth impact are added there. Caveats: Radar v1.2 and v1.3 take the SAME entries (only the exit differs) — running both doubles that risk, which the overlap pill will flag; radar live records are weeks old, so stitched results are mostly backtest; and combining streams that were each tuned on the same period inherits their in-sample bias — treat the stitched numbers as the optimistic end, the live-only numbers as the honest-but-noisy end.
Starting balance $
Risk % per trade
Timeframe
On
Bot
Source
Profile / variant
Filters
Lev ×
Risk ×
Wunder mean-reversion
TEMA trend
Daytrade 1H shorts
record
—
Cipher
—
TA Supreme(phase 1)
backtest
—
Radar v1 🤖📡 retest
main stream
score ≥ATR% ≥
Radar shorts 📉📡 (bull-gated)
—
—
Radar v1.2 💎📡 runner ⚠ same entries as v1.3
flat +50% target
—
Radar v1.3 🎯📡 adaptive ⚠ same entries as v1.2
ATR-scaled target
—
Radar v1.4 🌐📡 wide ⚠ superset of v1.2/v1.3
ATR-scaled target
—
Reclaim Entry 🔁📡 ⚠ sibling of v1.4
flush-low stop
—
MLS Touch 🧭📡 ⚠ same episodes as Reclaim
ATR-scaled target
—
loading…
Run sheet · the exact bots to run simultaneously
The combined account runs three WunderTrading bots, two strategies, one Blofin account — Wunder bots 1 & 2 (one mean-reversion strategy split across two bots because WunderTrading caps a bot at 10 pairs) plus the TEMA trend bot on the majors. Each bot's setup card (webhook, alert messages, profile, enable/dry-run) lives on its own page under Bots in the nav: Wunder 1+2 · TEMA.
Sizing & why all three run whole — no side filters
Run all three together, whole: the 3.5-year replay below shows the combination beats any longs-from-one/shorts-from-the-other carve-out by ~100R at the same drawdown. Position size and leverage are set per bot inside WunderTrading (standalone-validated presets: Wunder 15%/5× iso on v1.2 · TEMA 10%/3× iso). Sizing note for the combined account: at those presets the Wunder side risks ~3× more per trade than TEMA (15%×5× ≈ 4% of balance at its average stop vs TEMA's ~1.4%), and the replay shows returns stop improving past ~2% risk while drawdown keeps climbing (4%-equivalent ≈ 47% max DD) — for balanced, replay-like risk on one account, size the Wunder bots nearer 6–8% × 5× and keep TEMA at 10% × 3×. Bots 1 & 2 are currently a live A/B (reset vs Apex Refined v1.1) in dry-run; the replay defaults to Apex Refined v1.2 — deploy it to both via the profile selects on their setup cards when you're ready to end the A/B.
The scanner only evaluates coins tradeable on Blofin (USDT perpetuals). When a signal opens or closes, its alert message is POSTed to the WunderTrading webhook and WunderTrading executes it on Blofin.
Setup guide — connecting WunderTrading & Blofin, step by step
Connect Blofin to WunderTrading. In WunderTrading go to Exchanges → Add exchange → Blofin and paste API keys created in Blofin (API → Create API, with read + trade permissions — withdrawal never needed).
Create the bot. In WunderTrading: Trading bots → Create bot → TradingView / Signal bot, pick your Blofin account. Set position size, and choose multi-pair if offered — this scanner signals ~53 different coins, so one multi-pair bot beats one bot per coin.
Copy the four alert messages. The bot's settings page shows its unique messages for Enter Long, Exit Long, Enter Short, Exit Short. Copy each one and paste it into the matching field below, exactly as shown — the codes are unique to your bot.
Check the webhook URL. The bot's page also shows the webhook address the messages must be sent to. If it differs from the default below (wtalerts.com/bot/custom), update the field.
Multi-pair bots need the symbol in the message. If your alert message contains a symbol (or WunderTrading asks for one), use placeholders — they're filled in per signal before sending: {{coin}} = BTC · {{symbol}} = BTCUSDT · {{pair}} = BTC-USDT · plus {{side}} {{entry}} {{stop}} {{tp1}} {{tp2}} {{tp3}} for price fields. A single-pair bot ignores these.
Save, then rehearse in dry-run. Tick Bot enabled, keep Dry-run ticked, and Save. Alerts now appear in the log below without being sent — let a signal or two flow through and confirm the rendered messages look exactly like WunderTrading expects.
Test delivery. Pause the bot inside WunderTrading (so no real trade opens), untick Dry-run, Save, and hit Test enter-long. A ✓ in the log means WunderTrading accepted it. Re-enable the bot in WunderTrading when you're happy.
Go live. With Dry-run off and the bot enabled in both places, every new scanner signal opens a Blofin position via WunderTrading, and the scanner's exit (stop / target / stoch) closes it. Exits are only sent for positions this bot opened.
Bot pairs (comma-separated coins — must match the pairs selected in the WunderTrading bot; empty = all)
💡 Position size (% of balance, leverage, margin mode) is set in WunderTrading, not here — open your bot in WunderTrading → Edit → Position size and pick "% of available balance". The scanner never sees your Blofin balance; it only sends entry/exit signals. The paper-mode and replay inputs on this dashboard are simulation-only and don't change live sizing.
Strategy profile
Exit profile
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (optional — manual kill switch, never sent automatically)
Test sends a sample BTC alert and honors dry-run. Before the first live test, pause the bot inside WunderTrading so it doesn't open a real position.
A second WunderTrading bot for the next-10 pairs by replay performance (WunderTrading caps each bot's pair list at 10). Set it up exactly like bot 1: create another signal bot in WunderTrading on the same Blofin account, select these 10 pairs there, then paste its unique alert messages below. Keep the two pair lists different — a coin routes to whichever bot's list it's on.
Webhook URL
Bot pairs (comma-separated coins — must match the pairs selected in the WunderTrading bot)
Strategy profile
Exit profile
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (optional — manual kill switch, never sent automatically)
Test sends a sample BTC alert and honors dry-run. Before the first live test, pause the bot inside WunderTrading so it doesn't open a real position.
⚡ Wunder Thunder · one strategy, two bots
Both WunderTrading bots run the identical gate — the union of the two rules that were previously split between them (reset-macro ∪ apex-bear-v12), plus the ADX floor nothing in the system ever had. They differ only in which 10 coins route to them, because WunderTrading caps a signal bot at 10 pairs.
Why the floors exist — and what is NOT proven about them
Every profile in the scanner asks only for adx > 20 (and Apex adds a < 45 ceiling). Nobody ever set a floor. The 20–25 bucket turned out to hold the entire book's drawdown while contributing ~3% of its R, and on the live forward trades it is the whole story: the trades the floor removes lost −16.2R at an 11% win rate, while the ones it keeps made +14.5R at 75%. Adding an RSI floor trades total R for win rate. What is not proven: the floor was found partly on this same book. On the independent 470-row Bitget replay it only confirms directionally (+25% average R, not +42%), and the long side does not replicate there at all — the effect that reproduces is on shorts. The low-ADX bucket was also positive in 2026 H1 and only became catastrophic in H2, which is a 19-trade sample. Treat the book below as an audition: judge it on live fills, not on the history. Full working: REFINEMENT.md §2026-08-21.
ADX floor
RSI floor (0 = off)
Saving the floors changes both the paper book and the live gate at once — they are the same function. Arming points both bots' strategy profile at ⚡ Wunder Thunder; it does not enable or un-dry-run anything.
Why these floors · variant ladder
Every configuration recomputed from the same merged book, right now. Battery = split-half, winners−40%, last-60, drop-best-5 and a −0.05R/trade friction charge — a config has to keep every cell positive.
loading…
📒 Paper mode · both bots as one account
Start from a balance and the page shows what the Thunder book would have done with it, trade by trade, compounding. Defaults are auto-picked for the highest profit this book supports: the configuration (which floors) that ends highest under the real-world haircut, and the largest risk per trade whose worst dent stays under 25%. Change anything and it recomputes.
Starting balance $
Risk % per trade
Configuration
Custom ADX / RSI floors
Real-world haircut
Bot
Source
Window
Deploy · bot 1 · pairs 1–10
Webhook URL
Bot pairs — must match the pairs selected inside the WunderTrading bot
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (manual kill switch — never sent automatically)
Position size, leverage and margin mode are set in WunderTrading, not here. Before the first live test, pause the bot inside WunderTrading so it can't open a real position.
Deploy · bot 2 · pairs 11–20
Webhook URL
Bot pairs — must match the pairs selected inside the WunderTrading bot
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (manual kill switch — never sent automatically)
Trades & alert log
loading…
Historical tester · totals by timeframe
The low-timeframe study (2026-07-08), interactive: the live engine replayed bar-by-bar on the 20 bot pairs per timeframe, with costs charged against each trade's own stop distance. Change the strategy gate, side, or cost model and the totals recompute — no trade-by-trade noise, just what each timeframe would have paid. Deployed gate = exactly what the live Daytrade Bot trades (1h shorts · Stoch+ADX · BTC pump gate).
Strategy gate
Side
Cost model
Starting balance $
Position size %
Leverage ×
From
To
Chart / monthly TF
loading…
Bot 3 setup · Daytrade 1h shorts
The strategy — why 1h shorts only, and how the bot is gated
Quick trades on the 1-hour timeframe — shorts only, median hold ~13 hours, force-closed after 48h. Built from the 2026-07-08 low-timeframe study: 1h Stoch+ADX shorts netted +0.31R per trade after real taker costs (positive 7 of 10 months, 12+ coins); 1h longs and everything at 15m/5m/1m LOST money net of fees, so this bot never sends longs and lower timeframes were rejected. Entries pass the BTC pump gate, its own cluster cap (separate from the 4h bots), and the adaptive streak/governor as its trade history builds. Set up a third WunderTrading bot on the same Blofin account with these pairs and paste its Enter/Exit Short messages below. Start in dry-run — let it build a live shadow record first.
Webhook URL
Bot pairs (comma-separated — must match the WunderTrading bot's pairs; ranked by the 270-day 1h replay)
Enter Short message
Exit Short message
Exit ALL message (optional — manual kill switch, never sent automatically)
Cluster cap
Hold cap (hours)
Its signals appear in the signal log with 1h timeframe and in the alert log as bot 3. In WunderTrading set position size % and leverage as usual — sizing guidance pending the live shadow record.
Bot 4 setup · TEMA 4h trend · ETH BTC SOL DOGE
The strategy — TEMA 20/50 + SMA200 band, audited & rebuilt with real Blofin costs (+219R / 3.5y)
Port of the TradingView TEMA Cross Bot after the 2026-07-09 audit & rework: TEMA 20/50 regime + close beyond the SMA200 ± 0.8·ATR band enters, the opposite band edge is the stop and re-trails every 4h close, 40% takes profit once at 0.5R, an opposite signal reverses the position. Rebuilt with real Blofin costs after finding the original's divergence entries were broken churn (they faked the 74% win rate) and its ATR trail never existed in Pine. Validated 2023→now: +219R across 296 trades, max drawdown 17.8R, every year positive — expect ~33% winners with the profit in a few big trend rides, the opposite shape to the mean-reversion bots. ETH entries are additionally gated by BTC vs its daily 200-SMA (long above / short below); BTC, SOL and DOGE tested better ungated. This is a completely separate stream from bots 1–3: its trades live in the tables below, not the signal log.
Webhook URL
Bot pairs
Macro-gated pairs
Pump gate %/24h
Exit profile
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (optional — manual kill switch, never sent automatically)
⚠️ The 40% partial at 0.5R has no WunderTrading message yet — live, the full position rides until the band stop or reversal closes it (slightly more aggressive than the backtest). It's recorded here either way, so the dry-run R matches the validated numbers.
Paper mode · TEMA strategy
How this replay sizes, compounds and liquidates
WunderTrading-style replay of the TEMA strategy's trades: margin = balance × size%, position = margin × leverage (isolated), compounding trade by trade in exit order. Costs are already inside every R (the backtest charged taker/maker fees, slippage and funding), so what you see is net. A trade whose worst drawdown would have eaten the whole isolated margin at your leverage counts as a liquidation even if the trade later recovered — that's what actually happens on Blofin. Trend trades hold for days–weeks with stops 2–6% away, so this system tolerates less leverage than the mean-reversion bots: at 10%/3× the full backtest never liquidates; push leverage up and watch the liquidation counter before dreaming big.
Starting balance $
Position size %
Leverage ×
Side
Exit profile
Trades
Timeframe
From
To
loading…
Live state & positions
loading…
Cipher Bot · Market Cipher B waves · 1h paper signals
The strategy — VuManChu Cipher B confluence, scoring, and the validated Refined gate
A faithful rebuild of the open-source VuManChu Cipher B oscillator (the free Market Cipher B) with multi-timeframe confluence, running as its own strategy stream: a 1h green dot (WaveTrend cross up in oversold) or blood diamond (cross down in overbought) only becomes a trade when money flow agrees, the 4h wave is positioned on the same side and curling over, the daily wave has room, and the EMA-ribbon trend allows it — counter-trend entries need the A+ combo (extreme zone + WaveTrend divergence). Each signal is scored (max ≈ 10.5): ≥ 6 trades, ≥ 8 is an A+ setup. The 365-day backtest showed the raw system fails the graduation bar (+0.03R avg, PF 1.05 — fee churn), so the Refined gate now trades only where the edge actually lives: score ≥ 6.5 with an extreme-zone dot, the daily wave stretched past ±40 on the signal side, taken against the BTC regime (longs only while BTC is under its daily 50-SMA, shorts only above), max 3 entries a day — validated at +0.45R avg / PF 1.74 net of costs, both halves positive. Trades are paper-tracked with a structural/ATR stop, 1R/2R/3R targets, an opposite-extreme signal exit and a 5-day time stop. Phase 1 sends no orders anywhere — it must first earn ≥ +0.15R per signal net of costs with PF ≥ 1.3 over 150+ signals (and A+ must beat standard, or the scoring isn't discriminating). The Telegram feed and the tables below are the audit.
Ship order — dry-run until graduation, and why only a 10-pair slice executes
Execution wiring for the Cipher strategy — set up exactly like the other bots: create a signal bot in WunderTrading on your Blofin account, select its pairs there, and paste its unique alert messages below. Ship order: this bot should stay in dry-run until the live paper record passes the graduation gate (≥ 150 signals or 60 days live, ≥ +0.15R per signal net of costs, PF ≥ 1.3, A+ beating standard — the pills on the settings card above track it). The refined gate signals the whole universe (~5 trades a week spread across hundreds of coins) while a WunderTrading bot holds at most 10 pairs, so the pair list below chooses which coins' signals actually execute — in the 365-day backtest the top 10 coins carried only ~37% of the total R, so a 10-pair slice capturing part of the edge is expected, not a bug. Every signal keeps paper-tracking regardless of this list.
Webhook URL
Bot pairs (executed subset)
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (optional — manual kill switch, never sent automatically)
Alerts (sent and dry-run) appear in Recent alerts on the Portfolio page alongside the other bots, tagged bot 5.
Paper mode · Cipher strategy
How this replay sizes, charges fees and liquidates
WunderTrading-style replay of the Cipher trades: margin = balance × size%, position = margin × leverage (isolated), compounding in exit order. Cipher R is raw engine R, so the fee % knob charges a round trip (taker + slippage) against each trade through its stop distance before compounding — 0.1% is the spec's cost assumption, and with 1h-sized stops it matters. A trade whose worst drawdown would have eaten the whole isolated margin counts as a liquidation. Backtest rows are the all-pairs historical replay of this exact engine; live rows are what the paper tracker has recorded since it switched on. Refined gate (on by default) replays only the trades the deployed gate would take — untick it to see the raw unfiltered system, which is the configuration that failed the graduation bar.
Starting balance $
Position size %
Leverage ×
Fee %
Side
Grade
Trades
From
To
loading…
Live state & paper signals
loading…
TA Supreme · TEMA-200 retest · daily & 4h paper cohorts
The hypothesis — TEMA-200 retest from the ONDO case study (Phase 1, not validated)
The TEMA-200 retest hypothesis from the July-2026 ONDO falling-wedge case study, encoded exactly as specified and run as its own paper stream: a coin qualifies when its TEMA 200 turns up after at least 20 flat/falling bars (a fresh trend birth, no more than 40 bars old) and price then comes back to touch the rising line for the first time and holds — the bar's low dips within half an ATR of the TEMA and the same bar (or the next) closes back above it. Entry is that close; the stop is dynamic — a bar CLOSING below the rising TEMA 200 invalidates the trade at that close, so risk shrinks as the line climbs; the target is the first structural level (S/R pivots + volume nodes), with a 60-bar time stop. Long only. Each entry also logs its confirmations — apex compression, higher-low above the 0.786, pre-breakout volume anomaly — and the counterfactual result under a static 0.786-fib stop, so every sub-hypothesis of the source document accumulates its own evidence. The 1D and 4h cohorts are tracked separately and never pooled (daily TEMA-200 births are rare; the 4h cohort accumulates statistics faster). This is an n=1-derived hypothesis, not validated edge — Phase 1 sends no orders anywhere: it must earn ≥ +0.15R per signal net of costs with PF ≥ 1.3 over 150+ signals before any execution wiring.
Paper mode · TA Supreme strategy
How this replay works + the confirmation filters
WunderTrading-style replay of the TA Supreme trades — identical math to the other strategies' paper modes: margin = balance × size%, position = margin × leverage (isolated), compounding in exit order, and the fee % knob charges a round trip against each trade through its stop distance (0.1% is the house cost assumption). Backtest rows are the all-pairs Crypto.com historical replay of this exact engine; live rows are the paper tracker's own record. Pick the cohort — 1D and 4h results are separate experiments and are never added together. The confirmation filters subset to entries that carried the case study's supporting evidence (apex compression / higher-low above the 0.786 / a ≥3× volume anomaly in the prior week) — that's how each sub-hypothesis gets audited against the unfiltered stream.
Cohort
Starting balance $
Position size %
Leverage ×
Fee %
Trades
From
To
loading…
Live paper signals
loading…
Coil Bias · post-impulse coil scoring · 4h paper cohorts
How Coil Bias works — impulse, coil, tiered entries (Phase 1, not validated)
A deterministic 4h state machine per coin: an impulse (a move ≥ 2.5 ATR inside 6 bars) is followed by a coil — at least 8 bars trading in a tightening range over the impulse's fib retracements. While a coin coils, a bull and a bear bias score (0–100) accumulate from sub-signals: which fib levels get defended, micro higher-lows/lower-highs, divergence, wick pressure, volume drying up, and RVOL anomalies. Entries come in tiers — Tier A (score ≥ 60 + stochastic reset at a defended fib, early), Tier B (a confirmed 4h close beyond the coil boundary in the biased direction), Tier C (score ≥ 75 + RVOL ≥ 3× inside a tight coil, anticipatory). Stops sit beyond the opposite coil boundary; T1/T2/T3 are fib extensions of the coil range. Universe = the bots' 26 pairs, scanned hourly on confirmed bars only. Phase 1 — paper only, no orders anywhere: each tier must independently earn ≥ +0.15R per signal with PF ≥ 1.3 over 150+ signals before any execution wiring; the tiers are never pooled.
Coil radar · where each coin is in the cycle
How to read the radar
COIL rows are the live setups — the bull/bear scores show which way the coil is leaning while it builds (a signal fires only when a tier's trigger conditions are met, not on score alone). IMPULSE means the coin just made an outsized move and the engine is waiting to see if a coil forms over it. IDLE means no active structure. Key level = the highest-scoring defended fib on the biased side; fib pos = where price sits inside the impulse's retracement (0 = full giveback, 1 = at the extreme); the ⚡ flag marks a volume anomaly during the coil — the candidate accumulation tell.
loading…
Live paper signals
loading…
Pattern Cohorts · chart patterns graded at scale · research layer
What this is — detection grammar, break events, cohort grading (no orders)
A server-side engine detects classical chart patterns (H&S / inverse H&S, ascending / descending / symmetrical triangles, rising / falling wedges) from confirmed-bar zigzag pivots on 1H / 4H / 1D, fully anti-lookahead — a pattern exists only from the bar its final pivot confirmed. A break is the first confirmed close beyond a boundary (triangles must also clear the last touch pivot, so apex drift doesn't count), tagged qualified when break-bar RVOL ≥ 2×. Every break is graded mechanically: HIT50 (half the measured move), HIT100, INVALIDATED (full-height adverse close or stop), or TIMEOUT — with MFE/MAE and R (stop = opposite boundary). Cohorts key on (type · timeframe · direction · qualified · BTC regime · volume tier). No stat is trusted below n=30, and nothing is "validated" without split-half + the stress battery. A validated "this pattern is noise" result is a success of the module, not a failure. Research/telemetry only — no orders, no writes to any other module's statistics.
Priors under test · hypotheses, not truths
How to read this
These are the claims pattern lore makes; the cohort ledger tests them. Qualified > unqualified — do RVOL-backed breaks outperform quiet ones? Triangle direction by regime — do triangles resolve with the BTC regime? Wedge reversal — do wedges actually break against their slope like the textbooks say? H&S after costs — does the most famous pattern survive fees at all? Numbers update as the ledger grows; nothing here is a recommendation.
loading…
Cohort table
loading…
Chart · patterns rendered on live candles
How to read the chart
Candles come from the scanner's own data feed (Bitget primary with failover) — the same bars the engine detected on. Solid lines are the selected pattern's boundaries with ● pivot markers; ▲/▼ marks its break bar — green when qualified (RVOL ≥ 2×), amber when not. The badge under the chart shows the pattern's exact cohort record — the ✓ appears only when that cohort passed split-half + the stress battery. Containment overlays the other detected patterns on the symbol (dimmed) so a setup is read inside its larger structure; trade ladder overlays the entry/stop/target lines of an open Radar trade card on the same coin, read-only. Click a row in the gallery to jump to that historical instance's own chart window.
Instance browser
choose filters and hit Load
🚀 Launch Sniper · pump.fun graduates paper — buying the graduate, honestly measured
How the Launch Sniper works (and what it will never do in this phase)
A separate service listens to Solana in real time for new pump.fun mints and Raydium pools, enriches each launch with on-chain structure at T0 / T+2 / T+10 / T+30 minutes (authorities, bundle share of supply, deployer history & funding origin, holder distribution, LP burn/lock), and scores it in two layers: hard disqualifiers (any one → REJECT — e.g. bundle ≥25% of supply; the RAKO launch at 74.61% is the canonical fail) and a composite 0–100 on survivors (bands A ≥75 / B 60–74 / C 45–59). Only A/B alert, at the T+10 snapshot — never at T0, that's the anti-FOMO gate. Everything is fail-closed: a launch whose data didn't fully resolve is INCOMPLETE and can't alert or paper-enter — missing data is treated as adverse, not neutral. A/B/C launches are then paper-tracked on 1-minute confirmed bars with three mechanical long entries — E1 graduation retrace, E2 sniper-flush reclaim, E3 LP-burn retest — each with a pre-committed stop, +2R/+4R partial ladder, trailing runner, 4h time stop, and 1% modeled costs per side. Phase 0 contains no execution code of any kind; the owner gate for even discussing live capital is 150+ paper fills at ≥ +0.15R net, per rule. Score weights are priors (n=0) — this page exists to find out whether they discriminate.
loading…
📒 Paper account — Launch Sniper v1: buy the graduate at its first 5-min close, 5% of stack, trail 30% / stop −30% / 24h, 1.5%/side · instant graduates (wash-pump cohort) excluded
—
Paper validation scoreboard (net of 1%/side modeled costs)
—
Launch feed
—
—
🐋 Whale Snipa Phase 0 — research only, no alerts, no orders
What this study is, and the three ways it would fool you
A backfill service enumerates every pump.fun graduation in a fixed window, replays each token's swaps in SOL terms to grade its outcome, and collects the wallets that bought early in the ones that ran. Prices need no oracle: memecoins are SOL-paired, so the SOL leg of the swap is the price, read from balance deltas — which works identically on pump.fun, PumpSwap, Raydium, Meteora and any Jupiter route across them.
Three failure modes decide whether any of this means anything. Survivorship — search winners for their early buyers and you will always find wallets with great records, so every candidate is re-scored against its whole trading history, not just its hits. Latency — many consistently profitable meme wallets are block-0 sniper bots whose edge is uncopyable, so every entry is re-priced as though you bought 60s after their fill, net of costs; that column, not realized PnL, decides whether alerts are ever built. Sybils — operators run dozens of wallets and the survivor looks like a genius; the cluster gate for this is not yet implemented, so treat a listed wallet as a candidate, not a conclusion.
The outcome distribution below is the honesty check. Calibration on 2026-08-16 found five separate ways the winner metric was inflated — bonding-curve baselines, fee dust misread as trades, dust clustering in time, no maturity requirement, and a dust filter that disarmed itself — which moved the measured 5x rate from 80% to 4.3%. If the bands ever drift back toward "most tokens win", the metric has broken again.
loading…
📈 $1,000 paper account — balance curve, drawdown and the trades that mattered (PASS wallets only, mirrored at the copy price +60 s, 1%/side)
—
🔴 Live watch & forward test — frozen wallets, mirrored on paper at the copy price
Wallets are frozen into the list the moment they qualify (battery PASS, or a habitual 3× seller); everything after each wallet's freeze is out-of-sample. Their buys and sells arrive by Helius webhook (poll fallback every few minutes), each is re-priced 60s after their fill from the coin's own tape, and a 1-unit paper position mirrors the wallet at that price, net of 1% a side. The avg× column on the sold part is the number that will decide whether copying is real. Alerts go to Telegram + the app; touch sniper.paused on the VPS silences them without stopping the record.
—
—
Latest moves
—
🐳 Big prints — whale discovery on the liquid pools — the lens the graduate study is blind to
The pump.fun-graduate lens can only surface small fast operators (an 85-SOL pool cannot absorb a whale). This stage watches the liquid SOL pairs — pump.fun's top coins by market cap + Raydium's busiest, resolved to their most liquid pool on any DEX — by polling vault balances; a jump that stands out from the pool's normal flow pulls that window's transactions and records every swap ≥ the print floor with the wallet behind it. Those wallets enter the same profile → battery → copy-lag → freeze pipeline as everyone else, tagged bigprint. Prints ≥ the alert floor ping the Snipa bot as a discovery ping, not a call.
—
—
Latest big prints (24h)
—
Study progress
—
Outcome distribution (from the graduation price)
—
Qualified wallets — offline battery, NOT a forward test
—
—
Consistent 2× · 3× · 5× sellers — second lens, not a gate
Not "was this wallet early" but "does it actually sell at 2×/3×/5×+, position after position, week after week". Realized on the part of each position actually sold (≥50% of it — a 1% scalp at 5× while holding the bag does not count). Ranked by distinct weeks with a 3× exit: a habit, not one lucky week. A wallet can lead this table and still fail the battery — the verdict column says why.
—
Why candidates failed
Gates are pass/fail, and the first failure is recorded. Knowing which gate kills a cohort is the most reusable output of the study.
—
💰 Server Wallet on-chain state of the SOL bot keypairs
What this page is, and what it deliberately never shows
Each SOL bot owns a keypair on this server. The dashboard only ever receives the public address, the balance, the signature list and any live (non-dry) orders — the secret keys never leave the box and the backend rejects a payload that looks like it carries one. A zero balance with zero signatures means the wallet was never funded, which reads very differently from a zero balance after trading, so both numbers are shown side by side. Armed is read from each bot's own config: a funded wallet that is not armed will not trade, and that is the normal state right now — every SOL bot is paper until its judge point is met.
—
loading…
📐 S/R Chart — TAS-SR v1.3 levels, pattern boundaries and scalp signals, your MA stack, and the 🧲 book on the coin
What is drawn, where it comes from, and what the 3-year grade says
The indicator is James's TAS-SR v1.3 Pine, ported line-for-line to the server (engine.js tasSr) and run on completed bars only — the HTF context comes from completed higher-timeframe buckets, so nothing here repaints the way request.security does on TradingView. Levels: the historical pivot map (touches · holds · breaks → strength 0–100, brightest = strongest, up to 3 each side within 8 ATR). Orange/aqua lines: the pattern boundaries (3 aligned pivot highs / lows). ▲▼ markers: scalp buy/sell signals exactly as the Pine fires them (threshold, cooldown, per-level lock, chop guard, knife + fee-gap vetoes). MA stack: EMA 21 green · TEMA 20 light blue · TEMA 50 yellow · TEMA 200 dark purple · SMA 200 red · SMA 50 blue. 🧲 book: every open MM position on this coin (entry solid, stop dashed red, target dashed green, labelled by stream), armed shorts and harvest watches (base / breakout high, dotted) as the potential trades, recent closes as ⬆/⬇ pairs. Potential read (right panel): the live buy/sell score and exactly which rule blocks a signal right now. Grade (2026-08-25, 119 coins × 3y 4h + 118 coins × 1y 1h, mechanical exits, 10 bps): the signals beat random entries by +0.05–0.10R on every exit rule tested (the levels carry information) but are net negative on all 26 exit rules (4h −0.05R/trade at best, WR 49% on a 1R target). Higher scores are worse, not better. One context passes split-half: sells at resistance with TEMA20 < TEMA50 and price < SMA200 (+0.08R, both halves positive) — small, and shown as a tag on the signal, not a rule change. Treat the chart as a map: where the levels are, what the book is doing, what would have to be true for a signal — not as an entry trigger.
🧪 Silent paper cohort — the one lead (4h sells at resistance with TEMA20 < TEMA50 & close < SMA200 vs every other signal as control · trail exit · judge ≥150 closed, ≥+0.15R · no alerts, no orders)
loading…
🔮 Predictor — what similar setups did next, drawn as a probability cone
How this works, and how to read the cone honestly
Type a coin and the tool takes the last W bars of its chart, normalizes the shape, reads the feature "boxes" (RSI, StochRSI, which side of the SMA200, 20-bar drift), and searches the whole corpus — every coin, years of bars — for the K most similar historical situations (shape correlation minus a feature-distance penalty, capped per coin so one token can't dominate). What price did over the next H bars in each of those situations becomes the cone: the solid line is the median path, the inner band holds the middle 50% of outcomes, the outer band the middle 80%. The cone is a distribution, not a call — when the bands straddle zero, the honest reading is "no edge either way". If a tracked chart pattern (HS, wedge, triangle…) is active on this chart, its graded cohort from the pattern engine is shown beside the cone. The calibration card is the part that keeps this honest: a nightly walk-forward test predicts at random historical moments (analogs restricted to what was knowable then) and reports the median path's directional hit rate and whether the bands contain reality as often as they claim (50% / 80% targets). If those numbers are near coin-flip, the tool is a visualization, not an edge — and it will say so itself.
Probability checkpoints
—
🎯 Calibration — the tool's own measured accuracy, walk-forward
—
⚡ Breakout precursor scan — does NOW resemble the eve of a graded break?
—
🧲 Live MM state — what the MM system itself detects on this coin right now
—
✅ Pre-break checklist — the factors our break study VALIDATED, evaluated on this chart live
—
Closest analogs
—
📐 Active chart pattern
—
Corpus
—
backfill fetches ~3y of 4h + ~1y of 1h bars for the top-volume Bitget coins (one-time, ~15 min); the corpus then tops itself up hourly
⚡ Zeus — MM v1.5 + Daytrade 1H shorts + Wunder shorts (v1.1), weighted as one account
What Zeus is, and why daytrade shorts and not TEMA
The ask was MM v1.5 + TEMA shorts. Measured, that combination is a downgrade: TEMA shorts earn well (+761R) but carry a 468R drawdown against MM's 42R, and at every weight from 0.05 to 1.0 they LOWER return-per-drawdown — 17.8 → 3.5 at full weight. The optimum allocation to them is zero. Daytrade 1H shorts are the opposite: correlation −0.17, and at ~25% weight they add R and remove drawdown. A hedge is a leg that cuts the drawdown — a second earner is not a hedge, however profitable it is. The short leg is gated: it takes 1H shorts only while BTC's 7-day return is at or below −10%. Ungated the same book is +232R but averages +0.031R against 0.05R of friction, so it loses on execution alone and fails winners−40% at −896R; gated it is +378.6R at +0.299R average and passes all five battery tests. That gate is why it sits out most months — five of the last ten it took zero trades, including April 2026, when all 58 candidates were blocked and the ungated book would have lost 17R.
This is a view, not a bot: no cron, no Telegram, no orders. The daytrade leg has 167 replayed trades and zero live fills, and it is the leg delivering the lower drawdown — so the multiplier is also a dial on how much unverified strategy you are holding.
Daytrade weight0.25Wunder weight0.5
loading…
💵 Dollar paper account — capital-constrained: margin is HELD for the life of each position and a signal is refused when the account is full
🎯 Scalp Snipa — optimal settings — what the bot does, and how it would have grown $10,000
loading…
Advanced — change balance, risk, target
Starting balance $
Risk % per trade
Target
loading…
loading…
What this is, and what the numbers mean
The scalp watcher (/home/coder/scalp-watcher) stamps 5-minute sweep/climax bars on ~160 Bitget perps. v1.5.1 pinged every A+ stamp instantly: 367 pings in 5 days graded at 49% — a coin flip, and negative after fees. v1.6 (08-30) pings only LONG stamps on coins whose 5m ATR ≥ 0.5% of price, and only after a 15-minute confirmation (+0.3 ATR beyond the stamp); the card targets +0.6 ATR against a −1.5 ATR invalidation. That cell graded 84% / 76% in both halves (n=37) — small target, wide stop, only pays with maker fills. Every stamp it refuses (shorts, thin-ATR coins, unconfirmed) is still written to the ledger and walked as a paper position, so the gate is judged against what it turned down. Live = pings actually sent · Paper = all arms walked mechanically on 5m candles (stop-first when both touch, 4h time-stop, fee 0.10%) · Backtest = the 5-day study the gate came from. Judge: ≥50 confirmed pings closed.
Details — live pings, paper arms, shadow ledger
More views — paper account by arm · longs vs shorts · backtest charts
💵 Paper account — risk-% sizing (a −1R trade loses exactly risk% of the balance; 1R = the card's invalidation distance)
Starting balance $
Risk % per trade
Arm
—
🏆 Longs vs shorts — top-winning coins per side and how each side's book has run
Source
Min trades
Rank by
—
🎛 Gate knobs — applied by the grader within 5 min, the watcher picks them up on its next cycle; saved to scalp.json on the box
loading…
📚 Backtest charts — the 5-day study behind v1.6 (in-sample; look-ahead cells excluded)
A watcher on the box (/home/coder/deal-watch) polls hiddenclearances.com every 2 minutes — that is where the two X accounts' deals land — and checks every new deal title against the keyword list below. A hit is sent to every Telegram chat that has pressed Start on @DealGlitchesBot. X itself is polled every 10 min as a backup (X rate-limits unauthenticated reads, so expect it to say "429" most of the time). Keywords match whole words, case-insensitive: pc matches "Gaming PC" but not "PCS" or "UPC". Phrases are allowed ("gaming pc", "rtx 4070").
🔑 Keywords — saved to the watcher within one poll (~2 min)
📣 Alerts sent — deals that matched a keyword and went to Telegram
—
🧾 Latest deals seen — everything the watcher parsed, newest first (matches highlighted)
—
🎯 Migration Sniper loading…
🔴 LIVE NOW — real money, read from the chain · the paper twin of the same decisions beside it, never mixed in
loading…
📒 Paper accounts — one per migration stream, compounding through chain-audited closed trades
—
📈 Account view — full run since the stream's cutover (not 24h) · in $ at the live SOL price · paper model wallet, not the live wallet · chain-audited rows only
—
📊 Paper results (24h) — last 24 hours only · raw SOL at each row's fixed paper clip · chain-audited rows. Real money is only in the 🔴 LIVE card at the top.
—
—
🩺 Daemon
loading…
What this bot does, and the one number that decides it
The Whale Snipa study's biggest earner (FcvvJv…, +736 SOL in 15 days, 78% WR) turned out to be a graduation-slot sniper: the first buy on the PumpSwap pool in the very slot a pump.fun token migrates, sold 2–4 s later into the crowd that piles in. Copied one second late it is 1.005× — the edge is pure latency. So this bot exists to answer one question honestly before any SOL is risked: how many slots behind the migration are we, and how much SOL lands ahead of us?
It listens at processed commitment, decodes the migration's CreatePoolEvent and every pool swap from logs (no RPC in the hot path), and simulates a fixed clip against the real reserves at our assumed landing slot, exiting on take-profit / stop / hold / no-follow-through. Because a subscription opened after detection cannot see same-slot buys, every paper trade is audited 20 s later against the chain — that is what killed the 20× and 88× "wins" that were really 300–750 SOL whale snipes landing before us. The live executor fires on the same decision object as the paper stream, at the same instant, and mirrors its clip and exits; what it really got is settled per position from the wallet's on-chain SOL change.
🎛 What-if replay — the whole book re-walked with REAL execution priced in: slippage, take-profit and stop are yours to set
Where these numbers come from, and what each knob means
The live test found real entries cost ~20% more SOL than the paper model. Chasing that through 29,166 real swaps recorded in our own traces showed the pool's true curve prices ~25% ABOVE the vault model in both directions — buys pay it, but sells also receive it — decaying to ~20% by 5 minutes as pump.fun's boost escrow is bought back. So a round trip mostly cancels it; what a trade really loses is the premium decay during the hold, the exit latency (measured on 160 probed exits: tp exits realize −1.9%, sl +0.7%), and fixed costs (Jito tips + PumpPortal ≈ 0.008 SOL/trade). Every cell here re-walks the audited traces through the position engine under those rules. Slippage = the entry premium (25.5% is the measured value — the exit side scales with it). Exit model: measured decay is the honest default; flat (no decay) is the optimistic bound; entry-only charges the premium one way (the pessimistic bound — this was the old "-11 SOL everywhere" model); cap at trigger additionally refuses every favourable gap past the tp trigger (harshest). Grid precomputed by run.mjs whatif — the reduced cohorts only carry the measured slippage.
—
📏 Balance → clip → P&L — where a bigger wallet stops helping and starts hurting
How balance turns into clip size, and exactly where it goes negative
1. Balance → clip. Live sizes every trade as clip_pct (3.77%) of the confirmed wallet balance (lib/sizing.mjs, mode balance_pct). 3.77% is simply 0.4 SOL ÷ the 10.6 SOL the $1,000 paper account started with, so a 10.6 SOL wallet clips 0.4, a 26.5 SOL wallet clips 1.0, a 106 SOL wallet clips 4.0. Paper stays at a fixed 0.4 SOL clip. Nothing else in the sizing path knows about the pool — that is the problem. 2. Why the clip is not free. Every pump.fun graduation lands in a PumpSwap pool holding the same ~67.4 SOL of quote (min = max in this cohort). It is a constant-product AMM: our own buy pushes the price up by ≈(1+clip/67)²−1 — 1.8% at 0.6 SOL, 4.4% at 1.5, 12% at 4, 18% at 6. The mark the bot trades on is realizable value (what selling the whole position back into the pool returns), so that push is not ours to keep: reaching the +15% take-profit needs the pool to move 15% plus our own push, while the −30% stop is reached 30% minus it. The edge we ride — the boost buy-back and the crowd behind us — is a percentage of the clip; the cost of getting out is a percentage that itself grows with the clip. Linear gain, quadratic cost: they cross. 3. Where it goes negative. Each v4-pass row below is re-walked through the same position engine at 14 clip sizes with the evc≤1.10 price cap removed (so a bigger clip is priced, not refused) and the execution model on (venue premium, exit latency, tips). Net peaks near 0.6 SOL (≈16 SOL wallet), is flat-to-falling to 1.25, and is negative from 1.5 SOL (≈40 SOL wallet): take-profit exits collapse and stop-outs + timed-out holds take over. Beyond that every extra SOL of balance loses money. Hover the chart for the exact numbers at each clip; drag the wallet slider to see what YOUR balance would clip and earn. Read honestly: in-sample v4 rows (n small), replay == booked at 0.4 SOL. Drop-best-5 is negative at nearly every clip — the whole curve leans on a handful of winners. The shape (edge is capacity-bound near 1 SOL) is robust; the exact peak is not. Recomputed hourly by run.mjs clipcurve.
—
🎯 Realistic paper (24h) — the same fills, execution priced in
Paste any token's contract address (any chain DexScreener covers — Solana, Ethereum, Base, BSC…) and the tracker polls its highest-liquidity pool on your chosen interval (reading one pool matters: taking the max across all pools picks up stale prices from dead pools). Set alert levels — price or market cap, above or below a value — and a Telegram alert fires the first time the level is crossed. A level with ↻ re-arm resets itself once the market recovers 3% past the level, so repeated dips alert again; without it a level fires once, ever (you can re-arm it manually from the card). The optional status ping sends a periodic price / mcap / 24h summary regardless of levels. Both intervals are per-token and editable any time from the settings row at the bottom of each card — changes save instantly. 💰 Track position records your entry — as the price or the market cap you bought at, whichever you know — plus an optional $ invested, and the card then shows live PnL % (and $ value when the invested amount is set); the position line also rides along on level alerts and status pings. 🧮 What if is a target calculator: type any price or market cap and it shows the move from here, the implied other metric, and — with a position recorded — what your money would be worth at that target. 📈 Chart opens DexScreener's own live chart in a fullscreen overlay — it streams straight from DexScreener in your browser (it never touches the tracker's polling budget) and the stream stops the moment you close the overlay, so leaving it closed costs nothing. Alert-only — nothing here feeds the scanner's signal logic.
Contract address
Check every
Status ping
Tracked tokens
Level values are number × unit — 500 with K selected = $500,000 (the = preview shows exactly what will be saved). Typing shorthand like 500k or 1.5m also works. MC = market cap, PX = price.
Two independent layers. Buy/sell pressure needs no API key and always runs: DexScreener's trade counts and volume per window, so when sells outnumber buys ≥1.8× on a real sample (≥25 trades in the hour) you get a selling pressure alert — crowd-level distribution, no wallet named. Top holders and large trades need a Helius key (Solana only): the 20 largest token accounts are polled every 15 minutes and diffed, so a specific top-10 holder cutting ≥10% of their own stake (while holding ≥0.5% of supply) alerts as "top holder is selling", and ≥60% gone reads as "sold out". Individual swaps above 2% of pool liquidity (floor $500) alert as large buys or sells. Limits worth knowing: a falling token-account balance can be a transfer rather than a sale — it's a red flag to check, not proof; liquidity-pool accounts are excluded once flagged, since their balance moves on every trade; and the whole holder layer is Solana-only because that's what the provider covers. Alert volume is deliberately conservative — significant moves only, the same discipline as the liquidity-sweep tiers.
Once a day the top-65 perps by volume are screened on volume precursors only (range compression was tested and is anti-predictive at this horizon, so it's not used): rv1 = yesterday's volume vs the 20-day median, creep = last-5-day vs prior-20-day average volume, drift = the 5-day price drift that disambiguates direction. TIER-1 (rv1 ≥ 2, creep ≥ 1.5, drift up) pushed individually — backtest odds ~29% of a +10% move within 3 days (2.2× base rate). TIER-2 is digest-only (1.3× lift, on probation). The breakdown mirror (same volumes, drift down) flags risk on open longs. Honest coverage: these tiers catch about 1 in 5 explosive moves — the rest are news/listing-driven with no daily precursor; the Market Watcher's intraday RVOL alerts are the net for those. A flag promotes a coin to the watchlist conversation — it is never an entry. Tier-1 flags also auto-register a machine breakout trigger at the 30-day high (volume-gated, 5-day expiry) whose fire quality is being graded against human-registered levels.
loading…
Machine trigger registry
Live grading — flags vs what price then did
MM Liquidity Context · pools, sweeps & fakeout anatomy (shadow)
What this layer is — and is not
No villain is being modeled: stops cluster at obvious places, price gets drawn to resting liquidity, thin books amplify sweeps — mechanical phenomena regardless of who drives them. The layer maps liquidity pools (equal highs/lows, prior day/week extremes, round numbers, watched MAs, and our own registered levels — which are by definition obvious), detects sweep → reclaim events through them (the BEAT 07-24 / ONDO 0.3985 shape; ⚡ = full pierce and reclaim inside one 1H bar), and scores every breakout-class trigger for fakeout risk (obvious level + thin volume = the 0.3866 anatomy; strong volume into already-swept liquidity = the 07-21 anatomy). Everything is shadow-mode: annotations and advisories are recorded and graded against realized outcomes; no module consumes any of it until its cohort clears n≥150 with a ≥ +0.15R delta. A graded "no" kills that advisory — that's a successful result.
Alert policy (reworked 2026-08-10) — what pings, what digests, what stays silent
The old behavior alerted every ⚡ flash sweep — ~295 eligible alerts/day, unreadable. The module's own graded ledger (9,000+ events) showed why that was wrong: sweeps of round numbers and moving averages grade at a coin flip, so they now never alert (still recorded + graded). What's left is tiered: ⚡ MAJOR = a sweep of a structural level (equal highs/lows, prior day/week extremes, daily open, or a registered level) at least 1×ATR deep on 3×+ volume (or 2×ATR any volume) — the slice that grades best (51–54% marks-the-extreme, +3–4.5% avg follow-through). Majors ping in real time only on coins you're engaged with (an open trade card or a Trade-Watcher coin), max 1 per coin per 4h and 3 per 6h globally — and only ▼ down-sweeps (52% / +5.2% avg on 1,204 graded): ▲ up-sweep majors grade at a coin flip (50.1% / +0.6%), so they ride the digest instead of pinging. Notable sweeps (≥0.75×ATR on 2×+ volume of a structural level) and any majors over budget fold into one 🧭 digest message per 4h, one line per coin, silent when empty. Every alert states its direction lean in plain words and carries the live-graded hit rate for its cohort — the claim is measured, not asserted.
Alert scoreboard — the graded evidence behind the tiers
loading…
Recent sweep / reclaim events
loading…
Trigger annotations — fakeout risk before resolution
Cohort grades (nothing is consumed until a cohort clears its gate)
🔴 Live trade · stats & health
🔔 While a trade is live, every confirmed health-band change alerts on Telegram immediately (no rate limit), and in-band score drifts of ≥15 points send a 📊 health alert too.
Rank list · strongest setups on the radar right now
How the radar ranks coins
Every hour the radar scores the whole universe (every live USDT perp on the exchange that clears the Min 24h volume floor — thin, untradeable pairs are skipped entirely, which also makes the scan much faster; bot pairs and your watchlist are always included) in both directions using the same watch-mode gauge as the trade cards: auto-derived structure levels + Structure/Levels/Volume/Momentum over a 3-day window. A high LONG score means the coin's structure, volume and momentum currently favor a long — the trigger column shows the entry level that would confirm it and how far away it is. New setups reaching the alert band push one 📡 Telegram digest per scan. The exchange's tokenized stock/commodity/index perps (TSLA, NVDA, SPY, oil…) are scored too but hidden by default — tick "Include stock pairs" to see them (this also lets them into the digest). Ranking is decision support only — click 👁 to arm full entry/exit alerts for a coin.
loading…
🏦 Prop Op · MM v1.4 engineered for a prop-firm account
Strategy — every candidate is its own book filtered to the majors FundingPips actually lists, re-simulated as a $100k account under the firm's rules
Risk % per trade
What this is — and how it differs from the Market Maker book
The same harvest → base touch → volume-reclaim idea as the 🧲 Market Maker, transposed onto the 8 majors a
prop firm actually offers (BTC ETH SOL XRP ADA DOGE LTC BNB) and re-engineered for FundingPips' rules. The alt detector's
fixed-% thresholds are re-expressed as ATR multiples (calibrated once from the alt v1.4 cohort, median ATR 1.16% — a
transposition, not a curve fit), because a 10% flush that is routine on an alt is a black-swan candle on BTC. The alt v1.4
depth/close-pos entry gate does NOT transpose (non-monotone on majors) — the Prop Op gate is the system's own pull-profile
concept instead: slow walk-down ≥ 12h from harvest to base touch. Exits are v1.4's (disaster −2R, target → 1.5R trail,
break-even arms at +0.5R and locks +0.1R, 120h time stop). Sizing is the whole point: 0.30% of equity per trade keeps the full
3-year replay's drawdown at 7.5% — under the 8% design budget and well under the firm's 10% kill line, with the worst single
day at −3.85% against the 5% daily limit. The cost of that safety is pace (~+0.5%/mo — the phase-1 target takes months, which
only works because FundingPips has no time limit). Judge the live book at 30 fills before pointing a funded evaluation at it:
the replay's 2026 slice is negative (−5.6R), so this must earn its live proof first. Long-only; shorts stay parked.
loading…
💧 Skimmer — weekly income: keep a fixed amount on each exchange, move the rest off every payday
loading…
Keep on each exchange $
Simulation start $
Skim fraction
Min transfer $
Payday
Record
Max notional / trade $
Money on the exchange vs moved off — the simulation at the policy above, one exchange. All three exchanges run the same book, so multiply by the exchanges you fund.
Paydays
Payday
Income
Week R
Trades
Left on exchange
Moved off to date
Exchanges now — real balances, read through the executor
Exchange
On exchange
vs cap
Due now
Moved off to date
Last move
What I actually moved — log each withdrawal; the transfer itself stays manual
Exchange
Amount $
Date
Note
Date
Exchange
Amount
Note
🔴 Live Account
Real money only. Everything on this page is read straight from the exchange selected above (Bitget = bot 7, MEXC = the second balance) — your balance, your open positions, your filled trades and their realised P&L. No paper book feeds this page; if a number appears here, it happened on the exchange. Order attempts that the exchange rejected are shown too, so a failure can never hide as silence.
loading…
🪙 Token Launch preparation only
A place to plan a just-for-fun launch cleanly and, if it ever happens, watch it with the same eyes our sniper uses. Saving here does not launch, buy, sell or promote anything — the launch itself is done by hand in a wallet you control.
loading…
📣 Call Sniper
loading…
Positions
—
Alerted calls
—
Open orders — limit buys waiting for their price
—
🧪 Sol Combo paper leg
What this is — four auto-entry variants racing on paper
Every captured alert-channel call is paper-entered with a fixed 0.5 SOL at the listener's Jupiter quote and exited on the grader's sampled sell quote — so spread and impact are already inside every multiple. The four variants differ only in exit rule (take-profit × stop), 24h max hold. The 08-31 sweep on executable paths found no profitable no-stop TP on the current channels (most calls rug toward ~0.1× inside 24h; a rug can gap through the stop in one sample) — which is exactly why this runs on paper until a variant earns it at ≥100 calls. Nothing here can spend.
—
⚙ Settings — alert channels · disaster stop · top-holder sell alerts · what this is
What this is — and what it is not
The listener watches shiller channels and captures every contract address the moment it is posted, with an instant Jupiter quote as the honest "what a bot would have paid" mark. Channels ticked under Alerts ping your phone and appear below with an Enter button. Nothing enters on its own: pressing Enter queues a buy that the on-box trader executes from the Call Sniper wallet through Jupiter; Sell does the reverse for 100% of the holding. The measured record (deletion rates 33–84%, insydercasino calls = the top, BLACKBULLS −85% inside one 3-minute sample) says these calls are exit liquidity more often than not — size accordingly. Three gates on the box: wallet exists · armed file · balance ≥ size + reserve.
Alerts — channels that ping Telegram and appear above
—
🛑 Disaster stop — always-on floor under every position
—
🐋 Top-holder sell alerts — Telegram while you are in a position
—
📈 Channel scoreboard — entry (our quote at the post) → exit-able peak, every captured call, medians
—
Recent commands
—
🤖 Telegram Copy Bot
What it copies, and the one number that decides it
The source channel posts fully structured trades — pair, LONG/SHORT, a limit entry,
three take-profits, a stop loss and a confidence rating. This book mirrors them exactly: it only counts a trade if price actually
traded to the limit entry (their own "SIGNAL INVALID" rule — TP1 printing before the entry fills voids the signal), exits at
TP1, and honours the posted stop. When a single candle contains both the target and the stop, the
stop is assumed to hit first. Fees 0.06%/side.
The decisive arithmetic: TP1 sits closer to entry than the stop does, so a win is worth
roughly 0.66R while a loss costs a full −1R. That puts breakeven near a 60% win rate.
A high win rate is therefore not good news by itself — it is the minimum entry fee. What matters is how far above 60% it stays once
real fills, slippage and unfilled limits are counted. This is a paper book precisely so that question gets answered before any money does.
⛔ BACKTEST VERDICT: NO EDGE — do not fund.
All 992 signals (Apr 17 → Aug 20, 29 pairs) walked on 4.75M Bitget 1-minute candles: 915 fillable, win rate 58%, avg −0.116R, PF 0.75 — $1,000 at 2% risk becomes $109. Every exit variant is negative (TP1/TP2/TP3, break-even-after-TP1, shorter entry windows), both directions lose (long −0.152R, short −0.087R), the channel's own confidence rating does not separate winners, and the result is identical whether ambiguous candles are resolved for or against us. The reason is arithmetic: TP1 sits nearer than the stop, so a win pays ≈+0.66R against a −1R loss, which needs ~60% accuracy to break even — and the source delivers 58%. The earlier 69% read was a single symbol (ADA, n=35). The paper book below keeps running as a measuring instrument only.
loading…
🧲📣 MM v1.5 · WunderTrading version
Why this is not simply "MM v1.5 with fewer coins"
A WunderTrading signal bot caps at 10 pairs. MM v1.5's replay trades
469 coins — breadth is the strategy — so one WT bot can only ever see a slice of it.
The 2026-08-22 walk-forward (choose pairs on the first half's activity, score the held-out half) found the split that matters:
the reclaim leg keeps just 5% of its R under a 10-pair cap — it is a scan-the-whole-market
strategy and cannot be squeezed into one bot. The MA leg does the opposite and IMPROVES:
average R 0.195 → 0.692 and win rate 59% → 72%, because concentrating on the coins it fires on most selects for the ones with
reliable structure. So this bot is all v1.5 legs on the pairs the MA leg is busiest on — a
different product from MM v1.5, not a subset of it. Pairs are picked by activity, never by
profit: past per-pair profit does not persist, activity does (all 10 chosen pairs were still trading in the held-out half).
Runs beside the Bitget bridge, never instead of it, and a WunderTrading outage cannot affect it.
loading…
Paper mode · the bot's book as a $ account
The 10-pair book compounded at a fixed risk % of the running balance.
Frictionless and risk-based — P&L is risk$ × R, with no fees, funding or slippage.
It has to be: the MA replay rows carry no exit timestamps and the reclaim replay rows no exit price, so a
depth-aware $ replay is not possible on this book. Treat it as the shape of the equity curve, not a forecast.
Starting balance $
Risk % per trade
Legs
Deploy · the WunderTrading bot
Create a signal bot in WunderTrading on exactly the pairs listed above, then paste its four alert messages here. Fills on those pairs are mirrored to WunderTrading alongside the Bitget bridge — this never replaces it, and a WunderTrading outage cannot affect it.
Webhook URL
Pair source
Pairs (max 10 — must match the WunderTrading bot's selection)
Legs to trade
Enter Long message
Exit Long message
Enter Short message
Exit Short message
Exit ALL message (manual kill switch — never sent automatically)
Position size, leverage and margin mode are set in WunderTrading. Before the first live test, pause the bot inside WunderTrading so it cannot open a real position.
Live trades & alert log
loading…
🧲 Market Maker
Bridge
loading…
The doctrine, the pipeline, and how the book grades itself (2026-08-14)
A SEPARATE system from the scanner (own 5-min + 1-min crons in /home/coder/trade-engine), built on the market-maker read of breakouts: not every breakout gets pulled, but the pulled ones are the trade. Pipeline: every qualified lid-break arms a harvest watch → a harvest candle (≥3×ATR down bar on ≥5× volume after a ≥10% run) fires the 🩸 mass-liquidity-pull alert → the cascade is tracked minute-by-minute → the 🧲 entry fires on the volume RECLAIM of the pre-break base (1-minute close confirm — VELVET 08-14: entry 0.669 vs 0.926 on 15m bars). Pull-vs-dump discriminator, measured on 2,102 episodes: genuine pulls overshoot the base ~1.3% and walk down slowly; dumps overshoot ~32% and crash through. Entry tiers: FULL (slow walk-down — sim 270tr +179R avg +0.66R, full battery pass) and SQUEEZE (violent flush that still reclaims — VELVET class, 32tr +15.9R, half size). Deep-book ignition breaks trade the flip-retest (🚨). Exits are mechanical (15m-close stop, BE@+1R, pre-pull-high target, 5d cap) with a 📐 multi-timeframe exit-intel layer for the human. Self-review: the 🚨 ignition and 🧲 reclaim classes run a dump-day cluster rule (2026-09-03, on the v1.3 twin each actually trades — every v1.4 reclaim fill is one of the v1.3 twin's rows): net R over the trailing 48h at or below the bench line (ignition −15R, reclaim −6R) benches the class to HALF size on the bridge with a ⚠ tag on alerts, lifted once that net is back above the lift line (−5R / −2R; hysteresis — a cluster rolls out of the window on its own within two days). The report classes keep the trailing-30 rule — negative at n≥30 demotes them to ledger-only. Pre-registered rules; gates never move. Crypto-only universe (tokenized stock/ETF perps excluded). The book lives in trade-engine.db and refreshes here every 5 minutes.
loading…
📐 Trade chart — any coin the book traded or watched, TAS-SR levels + MA stack, and every strategy's honest entries / exits overlaid (2026-09-09)
Execution bridge — Bitget (bot 7 — the book's fills placed as real orders, direct Bitget API)
What fires, how it's sized, and the arm-up path — read before turning anything on
What fires: every 🧲 OPERATOR RE-ENTRY confirm (both the 5-min and the 1-minute fast-confirm crons) opens a long at market, and the selected book's exit closes it at market — v1.0 follows the live book's exits (15m-close stop, BE after +1R, pre-pull-high target, 5-day cap) while v1.1 follows the refined shadow book (⛔ −2R disaster stop + ride past the target with the 1.5R trail — battery-validated on replay, but its LIVE record is days old; the A/B judge at ~30 shared fills decides which one earns this seat). Entries are identical in both books, so the stream choice only changes when you get OUT. 🚨 ignition breaks and 15m-confirm SQUEEZE tiers still trade: squeeze entries are sent at half risk, exactly like the paper book — but ignition (🚨) trades are NOT bridged, this card is the reclaim book only. The brx.paused kill-switch file and a self-review class demotion both silence entries here the same way they silence the alert. Sizing is risk-based like paper: each entry risks Risk % per trade of futures equity over the entry→stop distance, isolated margin at your leverage, capped at Max margin % per position — tight-stop trades get downsized, never oversized. Exits market-close the whole position, so no size bookkeeping can orphan a remainder. Keys: the Bitget API key is shared with the Radar page's bridge (one key, two bots) — create it with futures trade permission only, NO withdrawal. Arm-up path: save with dry-run ON → watch the log below for a few days (dry-run logs the exact order it would place, with real size from your real equity) → house phase gate before real orders: the LIVE paper book (not the replay) needs a meaningful positive record — it opened 08-14, so this stays dry-run for weeks, not days. One honest caveat: the backend mirrors the paper book's timing, so fills lag the paper entry by up to a cron cycle and stops book at the 15m close like paper — live R will track the paper book closely but not exactly.
—
Followswhich book's fills and exits the live account copies
Exit stream
Sizingrisk per trade · leverage per trade · margin ceilings
Risk % per trade
Sizing
Leverage mode
Leverage (fixed mode)
Leverage cap (adaptive)
Margin % per position
Max margin % / position
Legswhich of the book's legs may open real positions
Guards0 = off on every one
Max open positions (0 = off)
Cluster cap / 4h (0 = off)
Slip guard % (skip if chased)
Max chase (stops)
Venue & API keys
Venue (bot 7)
Bitget API key
Secret / passphrase
MEXC API key
MEXC secret
Hyperliquid API wallet key
Hyperliquid account address
KuCoin API key
KuCoin secret / passphrase
BingX API key
BingX secret
🟦 Second balance (bot 10) — a second copy of this bridge on its own venue and keys
Fed by the same events, with its own venue, keys, risk and ledger. Runs BESIDE bot 7, never instead of it. Ships off + dry.
Paper mode (the book priced like a real perp account — same engine as the app's Paper screen; defaults = the live bridge)
How this replay works — the backtest seed, sizing, costs, liquidations
Backtest = the harvest-cycle study tape replayed under the EXACT live rules (15m volume-reclaim entry, stop under the cascade low, BE at +1R, run-high target, 5-day cap, both tiers): pull-profile full size + squeeze tier at half size — the validated book the alert copy quotes. Live = trades the 🧲/🚨 crons have actually closed since 08-14; combined stitches backtest → live with no double-counting. v1.1 = the SAME entries under the refined exits validated 08-14 (split-half + full stress battery pass): an intrabar ⛔ disaster stop at −2R caps the loss tail (the outsized losses were tight-stop trades whose 15m close blew far through the stop), and on target touch the trade RIDES instead of banking — stop floors at the target, then trails 1.5R under the highest close (🏄). Replay: +195.8R → +223.6R, worst loss −4.55R → −2.19R, same win rate. The v1.1 live book is a silent shadow A/B on the same fills — judge at ~30 closed. Pricing mirrors the Trade Engine replay: % of FREE balance committed as isolated margin × leverage, overlapping positions share the account, taker fees both ways + funding by hold time, squeeze-tier entries auto-sized at half, and liquidation when a position's worst excursion crosses ~95%÷leverage (isolated) or the account drains (cross). Untick real-world mode for the frictionless risk-based ideal.
Bookwhich trade stream to price
Trade stream
3-year cohort
Accountsizing exactly as the live bridge: 0.5% risk · adaptive leverage ≤50× · cluster cap off
Starting balance $
Sizing
Position size % of free balance
Leverage ×
Leverage mode
Cluster cap (0 = off)
Max margin % of equity
Max concurrent (0 = off)
Costs & marginreal-world mode only
Fee + slippage per side %
Est. funding per trade %
Max position $ (liquidity cap)
Margin mode
loading the book…
Candidate legs and parity twins — silent paper books that earn their way onto the bridge. Tap a row to open it.
🚀 Ignition v2 — breakout research + silent paper (research only · not earned · no orders)
▸
What this card is
The ignition-v2 candidate James asked for on 2026-09-13 (earlier entries, hold more of the move; LSK as the case study). The backtest record and the silent live paper twin load from the password-protected store with their rules, costs and verdicts. Research numbers only: selected after seeing out-of-sample data, on a survivorship-biased tape, with zero real fills — the next gates are an unseen-holdout test and the paper twin's live record.
loading…
🧲 MM v1.4 — the entry gate (same reclaim fills as v1.3, judged at the reclaim bar: depth + conviction + listing age; walked under v1.3 exits with a +0.5R/+0.1R breakeven)
▸
What the gate is, where it came from, and how v1.4 is judged
James, 2026-08-19: "is there any refinement that could get us closer to 70% WR with even higher profit than v1.3?" The 3-year continuous-tape study (471 coins, 7,217 reclaim episodes, mm_v14_study.py) found that on the honest tape no exit change fixes v1.3 (avg +0.22R, PF 1.37, fails winners−40%) — entry quality is the lever. Three entry-time facts recur in every gate that passes the full battery: depth — the fill is still ≥3% below the 96-bar (24h) 15m mean, so the cascade was real and the entry is at a discount (entries near the mean have no edge); conviction — the 15m reclaim bar closes in the top 30% of its range but is not a full marubozu (close-pos ≥0.97 is worse: you are chasing the bar); listing age — under two years (old perps do not harvest→reclaim). Gated: n=1,664 of 6,999, WR 56%, avg +0.50R (2.2×), PF 1.87, halves +456/+370, winners−40% passes, every year positive — a plateau (every neighbouring threshold passes; looser fails). The WR lever on top: breakeven arms at +0.5R and locks +0.1R → WR 60%, avg +0.46R, PF 2.08. 70% WR only exists in the 2023 cohort — ~60% is the honest ceiling that keeps the edge. Total replay R falls (fewer trades) but live is slot-capped, and R per slot-hour is 2.3× — at fixed slots the gated book earns more. v1.4 = this gated 🧲 leg + the 📐 MA leg + the 🔥 ignition leg (both v1.3 exits); the 📉 short leg is parked (negative on the 3-year tape). Live: every 🧲 fill enters this twin book — pass → walked, fail → booked "skipped" with the reason, and the v1.3 twin's result on the skipped fills is shown so you can always see what the gate cost or saved. Silent (no alerts, no orders) until ~30 shared live fills.
loading…
🔥 Ignition leg — MM v1.4 (paired A/B on identical fills — the 🚨 ignition entries under the v1.0 exits vs the v1.3 exits; v1.4 = the v1.3 legs + this leg)
▸
What this candidate is, how it is judged, and what it is not
The Market Maker's 🚨 ignition entries (deep-book lid-break, flip-retest fill) have always been walked under the v1.0 exits in the live book. Since 2026-08-19 a silent twin (brx_ign_v13) walks the same fills under the v1.3 exits — the exit set the 🧲 reclaim and 📐 MA legs already use. Rows filled before the twin existed were re-walked from their own fill on the same 15m tape (marked "history re-walk"); rows after it are walked live. Because both books hold identical entries, every difference in the table is exit-only. MM v1.4 candidate = the v1.3 legs (🧲 reclaim + 📉 base-break short + 📐 MA-reaction) plus this 🔥 leg. It is priced in Paper mode below as "v1.4 (candidate)" but stays a candidate — no alerts, no execution — until ~30 live (non-seeded) fills have closed and the paired A/B still favours the v1.3 exits.
loading…
🧲 MM v1.6 — silent 4h twins: 🪃 Catapult long + 🐻 Bear-rally short (candidate, judge @30 fills each)
▸
What the two 4h twins are, where they came from, and how they are judged
The 2026-08-26 v1.6 study (TEMA200 / catapult, 3-year 4h tape) found two things that PASS the battery and were not selection noise: the 🪃 catapult long (a base retest with relative volume ≥1.5 and a stacked TEMA slope, 4h exits) and the 🐻 bear-rally short (a StochRSI turn while BTC is under its 4h SMA300, the coin is under its TEMA200 and TEMA20 < TEMA50) — the first short leg that LOWERS the portfolio drawdown instead of adding to it. Both run as silent paper twins: no alerts, no exec bridge, their own book, seed=0 rows are live, seed=1 rows are the 3-year replay on the same rules. The replay is in-sample by construction; only live fills count toward the judge (≈30 each). The short leg physically cannot fire while BTC is above its 4h SMA300, so it may sit empty for weeks — that is the rule working, not the twin failing.
loading…
🩸 MM v1.6.1 — the DUMP leg: 🔥 ignition mirrored to the short side (silent live-only twin, judge @30 fills — no backtest can exist)
▸
What the dump leg is, what it deliberately does not mirror, and how it is judged
James (2026-09-03): "form a dump leg for shorts tracking the opposite of what triggers a long for ignition … see if maybe that is our winning short strategy". Stage for stage the 🚨 ignition pipeline flipped: a floor break (lowest low of the 60-bar 15m window, forming bar, ≤3.5% under it, volume ≥1.2× the median baseline, highs stepping down) scored by an upper stop-run sweep (15), a MARKDOWN daily regime (falling highs, close in the bottom 40% or under D-EMA21; 15 — counter-regime never trades), room below to the lowest prior-20-day low (≥6%, 10) and a 5m event bar (RVOL ≥2 closing in its bottom 40%, next bar holds under the midpoint; 20). Entry = limit at the floor within 3h, stop = floor + 1×ATR15, T1 = the break low, then the v1.3 exits inverted (−2R intrabar disaster, target touch → 1.5R close trail under the lowest close, BE lock −0.25R at +1R, 120h cap, fees both sides, no funding). Two ignition rules are refused, not mirrored: the ≥10× climax override (on the short side that bar is the capitulation low) and post-capitulation charts (a ≥40% day in the last 5) — both booked as "skipped" with the reason, like thin books. Ungated on BTC by design: every row stores whether BTC was under its 4h SMA300 when booked, so the bear cohort below IS the gated book and the leg gathers evidence in either regime. A forming-bar 5m/15m detector with no order-book history cannot be replayed — the live twin is the only evidence it will ever have. Silent: no alerts, no exec bridge. Priced in Paper mode as "v1.6.1".
loading…
🧲 Replay parity — is the live reclaim leg running the rules that passed the battery? (silent twin, 2026-09-03; judge @30 live v1.4 fills with parity exact)
▸
Why this exists and what it proves
09-03 finding: over 08-17→09-03 the live reclaim engine (harvest watches from lid-break + hourly tape seeds, per-pass caps) and the 3-year replay traded mostly DIFFERENT episodes — only ~60 of 116 live fills had a replay twin, 42 live-only fills lost −20.6R, and 31 replay episodes in coins the live book trades had no harvest watch at all. The battery was passed by the replay's rules, so the live leg must run the replay's rules: this twin runs mm_backtest_hist's own episode detection and walk_long live, on a per-coin 15m tape cache (par_tape.db), booking every fill the replay would book (entry = reclaim bar CLOSE, stop = cascade low × 0.998, target = max(run high, +2%), tiers, the v1.4 gate) for both detectors (tape / break) and both streams (v13 ungated / v14 gated). Seed rows (from 08-13) are compared row for row with brx_bt3y — entry, stop, target, exit and R must be identical; live rows accrue from deploy. Silent: no alerts, no exec bridge — the bridge still follows the old v1.4 twin until this book earns the judge, then every reclaim leg (v1.4 → v1.6.1, Zeus, Colossus) and the bridge move onto it.
loading…
🏛 Colossus Bot (the two systems run together on ONE account — Radar v1.4 Wide Runner + Market Maker v1.3, all legs)
▸
What Colossus is, why the two books diversify each other, and what it is not
James, 2026-08-18: "if I ran the 1.4 wide runner bot alongside the 1.3 MM bot, would that be realistic?" Colossus is the honest answer: one paper account holding both bots' positions at once — System A's 🌐📡 v1.4 Wide Runner (newly-healthy longs, score≥73 + ATR≥2%, atrK20 target) and System B's Market Maker v1.3 exactly as the exec bridge runs it (🧲 reclaim v1.3 + 📉 base-break short + 📐 MA-reaction v1.3). Nothing is re-optimised: every row keeps its own book's entry, stop, exit and R, and the account simply carries both. Why it is realistic: on the 365-day replays the two legs' monthly R correlate at about −0.1, only 1 of 78 runner trades ever overlapped an MM trade on the same coin at the same time, and the combined R-drawdown is no worse than the MM book alone — the runner rides multi-day trend expansions, the maker trades the intraday liquidity pull, so they mostly pay in different weeks. Both are still paper-only, phase-gated books (v1.4 n=78 replay / handful live; MM v1.3 live since 08-16) — Colossus inherits every caveat of its parts and adds none of its own edge. What it is NOT: a new strategy, an alert stream, or an execution target — no cron, no Telegram, no orders. Same-coin overlaps are kept as two positions (that is what two bots on one account would do) and counted in the portfolio read below.
loading…
🔁 Bridge reconciliation — paper rows ↔ bridge orders (every 30 min; real mode auto-holds the bridge on drift)
▸
What is measured and what holds the bridge
Link 2 of the parity chain (2026-09-03): for every live paper fill of the legs the bridge follows (🧲 reclaim v1.4 · 🔥 ignition · 📐 MA · 📉 short v1.5) — was an entry REQUESTED (brx.py's own ledger, bridge_calls), was it ACCEPTED (a plan, dry or real) or REFUSED (and why: cluster cap, min contract size, slip guard, leg off…), was the paper EXIT followed by a close order, how late was the request (entry lag), how far had the market moved from the paper entry when the plan was made (chase — recorded on every plan since today, dry-run included), and an estimated bridge R per closed pair = paper R − chase ÷ risk%. Missed R = paper R of fills the bridge did not take. Drift hold: in REAL mode, drift > 4R over the last 20 closed pairs, entry acceptance under 60% over ≥10 requests, or a real position without a close order 20 min after its paper exit flips mm_exec_dry_run back to 1 and pings Telegram; in dry mode it only warns. History: the bridge ran REAL from 08-20 23:11 to 08-23 02:26 — 46 entries, 46 closes, every coin balanced (no orphans); dry since.
loading…
🪜 Parity ladder — every leg: replay battery · live battery · replay↔live parity · paper↔bridge (battery_live.py, nightly 03:12 UTC)
▸
How to read a row
One row per leg. Replay battery = the house battery (both halves > 0 · worst 40-trade stretch > −40R · total minus top-5 winners > 0 · every year with ≥30 trades > 0) on the continuous 3-year replay of that leg's rules. Live battery = the same battery on the leg's live paper closes (under 30 closes = accruing). Parity = does the live engine run the replay's rules (the parity twin's row-for-row check; "by construction" where one script produces both). Bridge = the reconciliation for legs the bridge follows. GREEN = replay passes, parity exact, live not failing, bridge clean · AMBER = something is unmeasured (no replay exists, no parity twin) · RED = a battery fails, parity mismatches, or the bridge drifts / drops exits. A leg the bridge trades on real funds should be GREEN — anything else is an open question, not a strategy.
loading…
Radar Bot · what if the Rank list traded itself
Radar Bot · main stream (v1 rules — your 🤖📡 Telegram signals come from here)
How the main bot works — entries, exits, and why v1 carries the signals again
A fully automated paper stream over the radar's own signals. Arm: a coin/side whose hourly radar band reaches the arm band gets an armed setup carrying the auto-derived levels (re-armed with fresh structure while it keeps qualifying, dropped 24h after it stops). Entry: a confirmed 1H close beyond the trigger level, filled at that close — capped at the max-open limit, best score first, one position per coin+side. Exits: the stop level's 1H close (stopped), the target touch (target), the strength gauge collapsing to Invalidated (faded), or a 10-day max hold (time). All R figures are risk multiples against the initial stop distance; net R bakes in taker fees both sides (0.06%/side). Every entry is recorded in the Call Accuracy ledger with the price at the moment the alert was sent. Why v1 is the main stream again (2026-08-03): in the live A/B that started 07-24, v1 ran +5.75R over 20 trades (75% WR) while the 365-day-validated v2 retest variant ran −6.38R over 12 — breaks kept running without pulling back, so v2's retest limits either missed the move or filled only the reversals. Honest caveat — read this before sizing anything on v1: the full 365-day replay (2026-08-04, today's 242-coin universe, the exact engine gauge, live 10-slot config) says v1 has no year-long edge: 992 trades, −55.1R, 60% win rate, PF 0.86 — and 60% is almost exactly this geometry's breakeven, so the machine is a coin flip minus fees in every regime tested (bear/bull, longs/shorts all PF 0.77–1.01, 11 of 13 months negative). The same replay reproduces the current hot streak (its Jul-24→now slice: +2.8R, 71% WR vs live's +5.75R, 75%) — meaning the live form is real but borrowed from a friendly trending tape, not proof of edge. v1 carries the alert stream because its signal texture has been useful; the validated edges live in the v3 challenger and the shorts stream below.
arm band, hold cap and fee set here are shared by all three streams (v1.1 runs its own max-open cap of 15)
loading…
The tables here are the bot's raw R-multiple record. For what this stream does to an actual perp account — margin, leverage, fees, funding, slippage and liquidations — open Trade Engine → Paper Mode.
Radar Shorts Bot (the validated short side — own stream, own dials, 📉📡 alerts)
Why shorts get their own bot and their own gate — the 365-day study behind it
The main stream is longs-only because raw radar shorts lost money in every variant of the original sweep. The 2026-07-24 follow-up found out why, and fixed it: breakdown shorts only work against a firm tape. Gating shorts to bear regimes (which works for the Wunder bots' rally-fade shorts) made radar shorts WORSE — in a bear tape everything "breaks down" and then squeezes back through the level. The winning rule is the opposite: shorts trade only while BTC's last completed daily close is at/above its 50-day average — a coin breaking support while the tide is rising is genuinely sick (think AVAX and ONDO flagging ahead of their dumps while BTC held up). Validated over 365 days: 70 trades, +16.8R, avg +0.24R, PF 2.02, max drawdown 4.5R, both halves of the year positive; the combined long+short portfolio passed the full stress battery (fees ×2, winners −40%, drop-best-5, Monte Carlo). Everything else mirrors the v2 retest rules with the signs flipped: 1H breakdown close places a retest limit at the trigger for 24h, touch-based deep stop above, 50% banked at the first target with the stop moved to entry, runner rides the gauge, fade exit at Impaired. When the gate is closed the radar digest still shows every short call — tagged 🚫 so you never lose visibility — and a break during bear tape simply drops the setup (it re-arms as soon as conditions qualify again). Honest caveats: 70 backtest trades is under the house 150-signal phase gate, so this ships as its own A/B stream and earns trust (or doesn't) in live paper; and like the long side, ~55% of trades are small gauge-fade losses with the profit concentrated in partial+runner trades.
gate: checking…arm band, max open, hold cap and fee are shared — set them on the main Radar Bot card above
loading…
Radar v1.2 · the Refined Runner (v1.1's entry, gated to score ≥75 + ATR ≥4% — 💎📡 alerts, the designated Bitget-direct stream)
What the refined gate is, why it exists, and why this one passed the full battery (2026-08-06)
The discriminator study asked: what separates the New-Healthy alerts that run 30%+ from the ones that go nowhere? Answer, measured at alert time across 780 episodes: the runners are already hot (median +33% over the prior 3 days vs +12% for failures), volatile (1H ATR 5.0% of price vs 2.7%), near their 30-day highs, on surging volume — and the radar score does NOT separate them (median 72 for both). Score separates profitability, ATR separates movers, so v1.2 requires both: armed score ≥75 AND 1H ATR ≥4% of entry price, always full size. Everything else is identical to v1.1 (buy the newly-healthy transition at the last confirmed 1H close, +50% full-close target, structural card stop on 1H closes, 10-day cap, longs/crypto only, 24h re-entry cooldown). Why it earned its own card: it is the first config in the runner family to pass the FULL validation battery including winners−40% — 74 trades/365d, +34.2R, PF 2.00, both halves balanced (+16.9R / +17.3R), last-60d PF 2.19, fees×2 and drop-best-5 pass, max drawdown 5.7R, and the ATR threshold is robust from 3% to 6% (PF rises monotonically). The mechanism is physical, not curve-fit: a coin moving 2.5% a day cannot travel +50% inside 10 days — the ATR floor removes trades that can't reach the target. Expect ~1–2 entries a week; v1.3 (the Adaptive Runner below) takes the SAME entries with an ATR-scaled target, which is exactly what makes the three-way judgment clean: v1 vs v1.2 vs v1.3 at ~30 shared fills. This is the stream the execution bridge should trade in Direct Bitget mode when it goes live. (v1.1, the untiered runner this gate came from, was retired 2026-08-06 with 0 live fills — its 780-trade book fails winners−40% and the adaptive-exit sweep confirmed no exit fixes it.)
▶ Check now on the main card runs all streams · entries fire from the hourly radar scan itself
loading…
Radar v1.3 · the Adaptive Runner (v1.2's entries, ATR-scaled target instead of flat +50% — 🎯📡 alerts, A/B vs v1.2)
What the adaptive target is, the sweep behind it, and the honest caveats (2026-08-06)
James asked: instead of a flat +50%, capture anywhere from 10% to 100% if the coin keeps running. The adaptive-exit sweep tested four ways to do that against the full battery. Everything that reacts to the price path lost, again — wide step-ratchets and extend-on-strength floors joined the already-rejected trails. What won is set at ENTRY and never moves: target% = 20 × the coin's 1H ATR as % of price, clamped to 10–100%. A coin moving 4% an hour gets an 80% target; 5%+ gets 100%; on quieter books it would shrink toward 10% (v1.2's ATR≥4% entry gate means live targets span 80–100%). On the identical 74-trade/365d entry book this beats v1.2's flat 50: +43.6R PF 2.22 vs +34.2R PF 2.00, winners−40% stress +11.9R vs +6.8R, drop-best-5 +13.1R, last-60d PF 2.26 — and the k parameter is a plateau (15–20 both work), not a spike. The controls matter: fixed +60/75/100% targets do NOT beat flat 50 (+32.9/+34.7/+33.8R) — scaling to each coin's volatility is the mechanism, "aim higher" alone is not. Honest caveats: n=74; the gain over a fixed +100% comes from the 22 trades with ATR 4–5% getting 80–99% targets; win rate drops 46%→42% and more trades ride to the 10-day time cap (36/74) — expect longer holds and streakier equity than v1.2. Same entries as v1.2 by design, so the three-way judge is clean: v1 vs v1.2 vs v1.3 at ~30 shared fills — flat-50 stays the designated execution stream until v1.3 earns it. (v1.1 retired 2026-08-06: 0 live fills, its 780-trade book fails winners−40%, and no exit design fixes it — third confirmation.) 🕶 EMA21 ride shadow (2026-08-06, from James's EMA-adherence study): after a trade banks its target we keep watching the same position on paper — floor locked at the banked target, ride until the first 1H close below the 1h EMA21, 10-day cap. The backtest says riding pays +67.3R vs +43.6R, but every extra R came from ONE VELVET rally, so nothing changes until ~10 live shadow rides resolve; the 🕶 tags in the closed table and the shadow line above are that experiment accruing. Nothing ever trades off the shadow.
▶ Check now on the main card runs all streams · entries fire from the hourly radar scan itself
loading…
Radar v1.4 · the Wide Runner (wider gate: score≥73 + ATR≥2%, ATR-scaled target — 🌐📡 alerts, own judge clock)
Where the 73/2 gate came from, what it caught that 75/4 missed, and the honest caveats (2026-08-08)
Born from James's "80–90% of New-Healthy alerts win" observation (2026-08-07). The graded ledger says the raw feeling is repeat-alert bias — ungated, ALL New-Healthy entries with the adaptive exit lose (−5.9R over 666 trades; winners re-alert 4–6×, one-shot losers alert once) — but sweeping the gate frontier between "everything" and v1.2/v1.3's score≥75+ATR≥4% found a battery-PASS plateau at score ≥73 + 1H ATR ≥2% (long/crypto only, atrK20 adaptive target): 78 trades/365d, +46.8R, avg +0.60R, PF 2.18, WR 41%, max DD 9.4R — every stress cell green, and neighboring cells 73/2.5 and 73/3 pass too, so it's a plateau, not a spike. The score-73 floor is load-bearing: any-score + ATR≥2 fails winners−40% (the 70–71 band is where the one-shot losers live). What it buys over v1.2/v1.3: roughly 2× the entry rate, and it would have caught SKYAI's 08-05 alert (score 75, ATR 5.2%) that the 75-gate scans missed. Honest caveats: n=78 < the 150-signal phase gate; the sim book came from the replay harness (deployed code path is the same entry machinery as v1.2/v1.3 with the two thresholds widened, but live is the real test); superset of v1.2/v1.3's entries — when all three fire the same coin those are the SAME bet, which the portfolio overlap pill flags. Judge: own clock at ~30 fills vs v1.2/v1.3 on shared entries. This stream also feeds the Reclaim Entry card below: every gated alert (score≥73+ATR≥2) is recorded as "context" whether or not this bot is on.
▶ Check now on the main card runs all streams · entries fire from the hourly radar scan itself
loading…
Radar v1.5 · the Confirmed Runner (v1.4 trigger + LWN ⭐ PRIME as the second layer for entries AND exits — 👁📡 alerts)
How the two layers compose, what the paired judge measures, and why there is no backtest (2026-08-09)
James's design: "we have our entry trigger, but before entering the LWN is the second layer of confirmation for entries and exits." The trigger is the SAME validated v1.4 wide gate (newly-healthy long, crypto only, score ≥73, 1H ATR ≥2%) — but instead of buying the alert close, the coin is ARMED and enrolled in the Trade Watcher. The buy happens only if the watcher's ⭐ PRIME checklist completes within 48h: price reaches a real level, prints a reaction (sweep / stop-run / absorption / volume reclaim), and confirms with a close back over the level on RVOL ≥1.2 with sane risk. Entry = the confirm close, stop = the reaction structure (sweep low, or rung −1×ATR) instead of the radar's card stop; target stays the validated atrK20 adaptive ladder, 1H-close stop basis, 10-day cap. Exits get the same treatment: an LWN BASE_BREAK (1H close through invalidation) force-closes the position on the 1-minute watcher cron — hours before the hourly card stop would see it — and cancels arms that never entered. Arms with no PRIME inside the window expire as an explicit PASS. The judge is the point: v1.4 keeps trading every trigger, v1.5 trades the LWN-confirmed subset of the same triggers — a paired, controlled test of whether tape confirmation adds R, filters churn, or just skips winners (the score-gate studies died on exactly that sword, 3×). Decision at ~30 v1.5 fills; phase gate ≥150 / ≥+0.15R before any edge claim. Why no backtest: PRIME needs 1m/15m tape and there is no such history — this book is live-only and starts EMPTY; with the current dry tape (zero healthy crypto longs), expect the first arm to take days. Both layers already run — this stream adds zero new detectors, only the composition.
arms fire from the hourly radar scan · entries + LWN exits fire from the 1-min watcher cron
loading…
Reclaim Entry (big-run flush to the 1H SMA50/TEMA200, volume-surge reclaim — 🔁📡 alerts, sibling of v1.4)
James's pattern, the detector, why it only works on gated coins, and the honest caveats (2026-08-08)
The pattern James watches on the big runners: a large run, a dump back to the 50 SMA or 200 TEMA on the 1-hour, then a further push up with a surge in volume from that point. Mechanized as a state machine on confirmed 1H bars: RUN = close ≥30% above the lowest low of the prior 10 days and above the MA → FLUSH = a bar's low tags the SMA50 or TEMA200 (whichever price reaches first; the setup dies on a 1H close >7% below the line, or after 10 days) → RECLAIM = a close back above the MA that takes out the prior bar's high on volume ≥2× the 20-bar average. Entry at that close, stop = the flush low (1H-close basis), ATR-scaled target (same atrK20 as v1.3/v1.4), 10-day cap, 24h per-coin cooldown, longs/crypto only. The filter IS the edge: universe-wide this pattern fails all 27 sweep cells (WR 22–28%, PF ≤1.08 — pullback-reclaims happen constantly on coins going nowhere). Restricted to coins with a gated (score≥73 + ATR≥2%) New-Healthy alert in the prior 14 days, 18 of 27 cells pass the full battery; the deployed cell is James's original spec: 105 trades/365d, +93.8R, avg +0.89R, PF 2.31, WR 39%. It's complementary in coin space — UAI, an alert-time one-shot loser (−62%), is this book's top coin (+21.8R on the SMA50 cell): the flush-reclaim entry monetizes coins where buying the alert itself loses, because it waits out the dump. EMA21 as an extra flush target was tested 08-08 and REJECTED (76% duplicate entries at shallower tags; the genuine increment fails the battery; the edge is the DEEP flush). Honest caveats: n=105 < the 150-signal phase gate; 98/121 sim entries (SMA50 cell) sat inside a v1.4 alert's 10-day window — running both is a deliberate second bite at the same coins, NOT diversification (overlap pill will flag it); the sim used full-history indicators while live recomputes from the exchange's most-recent-1000-bar window (TEMA200 convergence — episodes that armed further back can be missed); and the live 14-day context starts accruing at deploy, so expect ZERO entries for the first days until fresh gated alerts build context.
▶ Check now on the main card runs all streams · reclaim scans fire hourly after each 1H close
loading…
MLS Touch Runner (MA-location entry: deep EMA21 touch while the 1H trend-stack holds — 🧭📡 alerts, paired challenger vs Reclaim)
The MLS spec, the state machine, what the 08-14 study validated (and killed), and the A/B vs Reclaim
From James's MLS v0.1 spec (2026-08-14): instead of waiting for a bar-close event pattern, the trend STATE decides when a moving-average touch IS the entry. State = the previous confirmed 1H bar closed with the stack aligned (close > EMA9 > EMA21 > SMA50) and the EMA21 rising vs 3 bars ago. Trigger = a bar whose low tags the EMA21, recovers ≥0.3×ATR off its low and closes back above the line — with a chase veto (no entry if the close sits >2.5×ATR over the EMA21) and a dead-tape floor (1H ATR ≥2% of price, the study's best single refinement). Stop = min(SMA50 − 0.5×ATR, touch low) on a 1H close; atrK20 ATR-scaled target; BE arm on a 1H close ≥ +1.5R; 10-day cap. What the study validated: on coins with a gated (score≥73 + ATR≥2%) alert in the prior 14 days — the same context feed as Reclaim — this cell runs 72 trades/430d, +71.3R, avg +0.99R, PF 2.20, WR 22%, every battery cell green (deployed close-based machine; the study's intrabar-stop sim was 78tr/+80.9R/PF 2.63, also a full pass). What the study killed (5th confirmation MA state alone has no edge): the raw universe, BTC-regime-only and TEMA200-side cohorts, ALL reclaim-cross variants (crossing the EMA21 from below — the spec's own BEAT trap), shallow EMA9-only touches, and EMA9/EMA21 trail exits (again). Why it's paired with Reclaim: 57/108 sim trades sit within ±48h of a Reclaim-family episode on the same coin and carry ~92% of the R — this is NOT extra coverage, it's a different trigger on the same episodes. On the 36 shared sim episodes MLS collected +74.5R vs the reclaim trigger's +47.3R, entering on average 1.37% cheaper (the touch fires on the way down/at the line; the reclaim waits for the volume-surge push). The judge is therefore paired on shared episodes at ~30 fills — same coins, same windows, trigger vs trigger. Honest caveats: n=72 < the 150-signal phase gate; WR ~22% by geometry (lottery-shaped: most entries scratch near −1R, winners run to the ATR target) — a 53%-WR partial-at-+1R variant also passed but with a weak first half, queued not shipped; running this AND Reclaim doubles exposure to the same episodes by design (overlap pill flags it).
▶ Check now on the main card runs all streams · MLS scans fire hourly after each 1H close
loading…
🏛 Colossus Bot (v1.4 Wide Runner + Market Maker v1.3 on ONE account — the two systems as a portfolio; same card as the Market Maker page)
Why the runner and the maker belong on the same account — and why this changes nothing about either bot
The 🌐📡 v1.4 Wide Runner (System A, this page) and the 🧲 Market Maker v1.3 (System B, its own page) never see each other: different triggers, different timeframes, different crons. Colossus is the paper account that holds both books' positions at the same time — the runner's multi-day trend rides plus the maker's reclaim longs, base-break shorts and MA-reaction longs — with every row keeping its own book's rules and R. On the merged 365-day replays their monthly R correlate ≈ −0.1, same-coin overlap is 1 of 78 runner trades, and the combined R-drawdown is no worse than the maker alone: the two edges pay in different weeks, so one account can carry both without the drawdowns stacking. It is a view, not a bot: no toggle, no Telegram, no bridge routing, and each leg's phase gate (~150 live signals at ≥ +0.15R before real money) still applies on its own. Price it in dollars under Trade Engine → Paper Mode → 🏛 Colossus (radar costs: depth-aware, 2-day replay hold) or Market Maker → Paper mode → 🏛 Colossus (maker costs: 1-day hold).
loading…
Execution bridge (bot 6 — real-order wiring for the radar streams, via WunderTrading or direct Bitget; ships OFF + dry-run)
How to connect this to your WunderTrading account on Bitget — step by step
On Bitget: create an API key (API Management → Create API Key) with futures trade permission only — no withdrawal, and paste WunderTrading's IP whitelist into the key. In WunderTrading: Exchanges → connect that Bitget key, then create a custom signal bot on the Bitget account (USDT-perps; set position size % and leverage there — that preset is your real sizing). WT generates JSON alert messages for enter/exit; copy each one into the matching box below (they carry the bot's UUID, which is how signals route to YOUR bot). Here: pick the stream, list the pairs your WT bot trades, keep dry-run ON — alerts are logged but not sent — and watch the log for a few days before unticking it. What gets sent: the stream's retest fills (entry), the final close (exit), and — only if you fill in the optional partial message and your WT plan supports partial closes — the 50% target booking; without it the live position simply skips the partial and closes in full at the final exit (slightly different R than paper, logged honestly either way). v3 half-size fills (loud breaks) are never sent: the WT preset can't halve, and the validated edge carrier is the full-size quiet cohort. Picking v1 — read this first: v1 is selectable but ⚠ NOT recommended for real money: the 365-day replay (2026-08-04) showed zero year-long edge (992 trades, −55.1R, PF 0.86, breakeven-minus-fees in every regime) — its strong live stretch was friendly tape, not edge. If you run it anyway: v1 trades BOTH sides (fill in the short templates too), it takes no partials (the partial boxes are ignored on this stream), and the dry-run + phase-gate advice applies double. House phase gate still applies: no stream should leave dry-run until its LIVE paper record clears ~150 signals at ≥ +0.15R. Direct Bitget mode (execution = Direct Bitget API): the scanner skips WunderTrading entirely and places orders itself, so bot 6 trades the same full universe as the paper stream — the only filters are the radar's own min-volume floor, the stock-pairs toggle, the band/trigger rules, and the max-open cap; the pair list and alert JSONs above are ignored. Create a Bitget API key with futures trade permission only, NO withdrawal, paste it here, and use Test connection. Sizing is risk-based like paper: each entry risks Risk % per trade of futures equity over the entry→stop distance (isolated margin at your leverage, capped at Max margin % per position — positions with tight stops get downsized instead of oversized). Exits market-close the whole position; the streams' 50% partials are skipped in this mode (v1 takes none anyway). Dry-run logs the exact order it would place — with real size from your real equity — to the bot alert log without sending; the phase gate applies exactly the same before unticking it.
Stream
Execution
WT pair list
Webhook URL
Bitget API key
Bitget API secret
Bitget passphrase
Risk % per trade
Leverage (isolated)
Max margin % / position
Enter long (WT JSON)
Exit long (WT JSON)
Partial long — optional 50% close
Enter short (WT JSON)
Exit short (WT JSON)
Partial short — optional 50% close
Paper Mode · the Trade Engine stream priced like a real perp account
How this replay works — sizing, margin overlap, costs, liquidations
Every closed trade the Radar Bot has taken (live paper + the backtest replay) is priced the way a real Blofin account would have traded it — the same real-world engine the Combo Bot replay uses. Sizing: each entry commits size% of the FREE balance as isolated margin × leverage, WunderTrading-style, compounding as the balance moves. Margin overlap: positions open and close on their real timestamps, so concurrent trades share the account — when free margin runs out, entries are skipped, exactly like a real account that's fully deployed. Costs: taker fees both ways, funding charged by actual hold time (0.01%/8h), and with depth-aware costs on, impact slippage that grows with position size vs each coin's own 24h volume (captured when the setup armed), a book-depth cap on position notional (~2% of daily volume), and stops that were swept past filling worse than −1R. Liquidation: isolated — a position whose worst adverse excursion crosses the leverage's liquidation distance (~95% ÷ leverage) before its stop loses its entire margin (⚡LIQ), even if the trade later recovered; cross — the whole account backs every position, deep wicks survive while equity holds, but stops pay their full distance and one runaway move can wipe the account (a live comparison of both modes at your settings is shown with the results). Cluster cap: radar entries often trigger together on the same 1H candle in the same direction — one correlated bet; entries beyond the cap are skipped like the live combo bots do. Untick real-world mode for the frictionless risk-based ideal — good for judging the engine's logic, useless for forecasting dollars.
Trade stream
Starting balance $
Position size % of free balance
Leverage ×
Cluster cap (0 = off)
Fee + slippage per side %
Est. funding per trade %
Max position $ (liquidity cap)
Margin mode
loading…
Radar · the levels you're watching, checked live
How to read a trade card — levels, the strength gauge, provisional values
Each card is one active trade with its watch levels defined ONCE, each fully qualified: price, timeframe, and whether it needs a close (confirmed bar of that timeframe only) or just a touch (wicks count). The engine enforces the wick-vs-close discipline — a level that needs a 1H close does not fire on a wick, ever. The Next decision strip is the answer to "what am I watching again?": the single nearest pending level in each direction. The Trade Strength gauge (0–100) combines five live components — Structure Integrity (30), Level State (25), Volume Posture (20), Momentum Posture (15), R-State (10); the gauge without its parts is a vibe, so every component shows its score and one line of live evidence. Bands: 70+ Healthy · 40–69 Contested · 15–39 Impaired · <15 Invalidated-pending-confirmation. Band changes (held for two checks) push to Telegram tagged [TRADE] — on a live position every confirmed band change alerts immediately (the 30-min rate limiter only applies to watch cards), and an in-band score drift of ≥15 points sends a 📊 health alert so a slide never goes quiet just because it hasn't crossed a band edge yet. Values in amber italics come from the in-progress bar — provisional, never used for level status. ⏳ Cold start: a card under 12h old has almost no rung/level evidence yet, so Structure and Level State sit near their neutral defaults and the score leans on Momentum — treat early bands cautiously; the gauge measures the position since its entry, not the chart's bigger trend. Near-live monitoring: cards are checked every minute (multi-source candles — Bitget, Blofin and Crypto.com failover — so one exchange's rate limit never blinds the loop). Close-based levels still fire only on their confirmed bar close, alerted within ~a minute of it; touch levels fire intrabar off completed 1-minute wicks, so a wick through your level alerts in ≤~60s instead of at the next hourly close. Decision support only: nothing here places orders or feeds any other module.
no trade needed — levels auto-derived from the coin's 4H/1D structure, entry/exit alerts armed; edit the levels aftersort
loading…
New trade card
Coin
Direction
Entry price
Entry time (blank = now)
Structural stop
Margin $ (opt)
Leverage × (opt)
Size note (opt)
Thesis (one line)
Templateclone a template, fill in prices — 60 seconds, not schema design
Price
TF
Condition
Role
Label
on-fire
Closed trade cards
—
Call Accuracy · how this page's calls have graded out
How calls are graded
Every call this page makes — a confidence band change (⬆/⬇), a level fire, a 🎯 entry signal, a 📡 radar new-setup digest, a 🤖 Radar Bot break or entry — is recorded with the price at the moment it was made, then graded automatically against what price actually did next. The primary test is the 24h forward move in the direction the call implied, normalized by the trade's initial risk so every call is judged in R: ≥ +0.25R = correct, ≤ −0.25R = wrong, in between = flat. The 4h and 72h moves plus the best/worst excursion inside 72h are stored too, so a call that was "early but right" is visible. Hit % counts correct vs wrong only (flats excluded); avg R is the mean risk-normalized 24h move across all graded calls of that type — for DOWN-calls a falling price counts as correct, so a well-calibrated gauge should be positive on every row. If any call type goes statistically bad (n≥20, hit <45% or negative avg R) a weekly 🧪 Telegram flags it — the trigger to recalibrate the gauge weights.
loading…
Call ledger
—
Live blotter · every position the combo bots take
How to read the blotter — dry-run, live and history rows
One blotter for the whole combo, newest first — every position the two Wunder bots and the TEMA bot enter. While a bot audits itself in dry-run its rows are flagged: the entry was generated and logged but never sent to WunderTrading, so this is the honest preview of what the live account would be holding. Once a bot goes live its rows show a green live pill. R fills in when the tracker closes the position. Grey history rows are each bot's historical record — the trades its current strategy profile took across the replayed history (Wunder bots: live scanner + ~3y replay on their pairs; TEMA: the 3.5y backtest) — so the blotter shows the full trade-by-trade story, not just what's happened since the bots switched on. The Daytrade bot is a separate system and isn't shown here.