Crypto Trader

Pipeline and Gates

IdeaDaily Scout ideas are deduped into GitLab tickets before they become backtest specs.
Test SpecTicketed ideas get a testable strategy spec, market scope, venue assumptions and risk premise.
Strategy SearchBacktest/search must pass robustness, worst-market and cost-model checks before promotion.
Shadow AdmissionCandidate enters simulated live tracking only after solid gates and runtime spec are present.
Shadow SoakFresh market data updates positions, MtM, drawdown, fills and retirement checks without real orders.
Live Review GateShadow-proven strategies enter this approval gate before capital, portfolio and family-capacity allocation.
Live TradingLive activation still requires explicit capital allocation, backend gate acceptance, kill switches and monitoring.

Risk rule: Real orders stay blocked until Plak explicitly approves capital and activation. The current exception is the bounded Microtest window: 25 EUR je Market for BTC and ETH with automatic safe-off conditions.
Required order: Idea -> Test Spec -> Strategy Search -> Shadow Admission -> Shadow Soak -> Live Review Gate -> Live Trading.
Fail behavior: Any failed gate stops forward movement. A new fix must restart at the relevant earlier gate and re-prove the chain.

Movement Rules

Idea to Test Spec

Moves forward when the idea is deduped into a ticket and has a testable hypothesis, market scope, timeframe, venue assumption and risk premise.

Stays blocked when the idea is only a narrative, has no measurable signal, or cannot be tied to a realistic venue.

Search gate Strategy families Venue rules

Test Spec to Strategy Search

Moves forward when the spec can be run repeatably with fixed inputs, cost model assumptions and measurable pass/fail metrics.

Stays blocked when the data source, symbol set, timeframe, cost model or expected trade path is ambiguous.

Strategy Search Cost model Metric columns

Strategy Search to Shadow Admission

Moves forward when the search artifact marks the candidate solid_valid, worst-market return, drawdown and profit factor survive the configured gates, and the promoted candidate ledger plus runtime evidence exist.

Stays blocked when a single market carries the result, the candidate is exploratory, the cost model is too optimistic, or required runtime metadata is missing.

Valid strategies All contexts Cost model

Shadow Admission to Shadow Soak

Moves forward when the candidate is paper-only active, has a runtime spec, simulated fills/positions can be tracked, and a fresh Shadow artifact reports its position and status.

Stays blocked when runtime spec, shadow state or data integrity is missing. Those cases fail closed instead of inventing a position.

Running shadow Position Status

Shadow Soak to Live Review Gate

Moves forward when the active Shadow candidate passes the exact live-readiness gate below and the backend marks it as review eligible.

Stays blocked when it is retired, stale, not venue-executable, below score/observation thresholds, or has a Shadow/readiness blocker. Missing capital or family-capacity limits block live trading later, not this review gate.

Exact 5m/15m Shadow to Review Rule

5m and 15m currently use the default live-readiness minimum: 30 Shadow days and 15 closed Shadow trades. There is no separate faster 5m shortcut in the current backend.

minimum_shadow_observation_days = 30 and minimum_closed_shadow_trades = 15.
shadow_observation_days >= 30 and closed_shadow_trades >= 15.
live_readiness_score >= 75.
eligible_for_plak_review = true: observation/trades pass, score passes, no readiness blockers.
plak_review_candidate = true: mirrors the strict Review Gate. Open Shadow-soak time keeps the candidate out of Review.
Retention must be active: no pending_retirement_at, no retired_at, and shadow_retention_state is active or live_now.
Live path must be approved for Plak: live_executable_for_plak = true.
For the Review page specifically: approval_eligible = true.
live_gate_bucket explains the later live-allocation state. Missing capital, portfolio exposure and family-correlation capacity do not keep it out of Review.
not_live_ready does not appear as a Plak Review card.

What the retention rule means

This is a health check for the Shadow candidate itself. A candidate may only move to Plak Review when Shadow is not already trying to remove, quarantine or pause it.

pending_retirement_at means an exit rule already fired, but Shadow is waiting for an open simulated position to close before final retirement. That is not reviewable.
retired_at means the candidate already left the active Shadow set. That is not reviewable.
active means normal Shadow tracking. live_now means already promoted/live-tracked. Both are allowed retention states.
watch, degraded, paused_new_entries or retired_final mean Shadow still has a quality, drift, evidence or retirement concern. Those states must stay out of Plak Review even when the score looks high.

Why Review happens before live allocation

The Plak Review page is the Shadow-to-live decision page. It shows candidates where research, Shadow evidence, retention and venue path are good enough for a human decision. Capital allocation is decided after that review.

not_live_ready: still missing observation, score, retention health or live path. It does not appear as a Review card.
ready_but_capital_blocked: reviewable; live trading remains blocked until capital is assigned.
ready_but_risk_or_venue_blocked: reviewable when approval_eligible=true; live trading remains blocked until the listed live-allocation blocker is resolved.
live_ready: reviewable; capital is available, but the backend still must not auto-activate without Plak approval and the separate live activation gate.

Shadow evidence Plak review Retirement rules

Live Review Gate to Live Trading

Moves forward when Plak explicitly approves the candidate, capital allocation and activation parameters are set, backend live gate accepts the candidate, and kill switches plus monitoring are active.

Stays blocked when approval is absent, capital is unset, live queue data is stale, activation side effects are present, or any risk blocker remains.

Review queue Live gate Decision labels

Shadow Timeframes

Shadow-Abschluss gilt je Timeframe fuer die Live-Review-Readiness: Shadow observation days and closed Shadow trades must both meet the minimum. It is not live approval and not an order signal.

Timeframe inventory complete: A timeframe is filled when the active Shadow inventory has at least 50 candidates for that timeframe. This controls discovery/refill pressure only.
Strategy evaluable: A single strategy becomes evaluable after it has fresh Forward Shadow data and enough per-strategy closed Shadow trade evidence from the live-readiness gate.
Final negative closeout: Hard retirement needs the retention guard: enough new gate-failure observations, repeated Profit Factor or solid-gate misses and concrete negative Forward Shadow evidence. The default guard is 12 gate-failure observations plus 2 repeated misses.
Live-ready: Shadow complete is not live approval. Live still requires backend review, venue/risk gates, capital allocation and explicit Plak approval.

Timeframe	Min. Shadow observation	Min. closed Shadow trades	Source	Current meaning
5m	30 Tage	15	DEFAULT_MINIMUM_SHADOW_OBSERVATION	Short-timeframe Shadow is review-complete only after both minima are met.
15m	30 Tage	15	DEFAULT_MINIMUM_SHADOW_OBSERVATION	Current Micro-Live strategy family uses this lane; completion still does not force live trading.
30m	30 Tage	15	DEFAULT_MINIMUM_SHADOW_OBSERVATION	Same default readiness minimum as 5m and 15m until explicitly overridden.
1h	30 Tage	15	MINIMUM_SHADOW_OBSERVATION	Explicit 1h rule in live_queue.
4h	90 Tage	12	MINIMUM_SHADOW_OBSERVATION	Longer observation window; fewer closed trades required because signals are rarer.

Inventory completeness is separate: each timeframe targets 50 active Shadow candidates for discovery balance. Candle balance is not a retirement cap and not a live-trading approval.

Runner / Dashboard Boundary

Trading Runner runtime root: /opt/bitcoin-trader-publisher/runtime/live/crypto-trader is the live runner source of truth for Micro-Live inputs, runtime state, preflight snapshots and execution artifacts.
Dashboard ist Read-Model/Monitoring: The dashboard consumes published artifacts one-way. Dashboard files must not be the execution input path for the Trading Runner.
Publisher boundary: The publisher builds runner-owned artifacts first and mirrors read-models to /opt/dashboard / k-dash after generation.

Live Microtest

Scope: Current bounded Microtest uses 25 EUR je Market for BTC/USD:USD and ETH/USD:USD.
Per-market guard: Each market has its own signal work and max one entry position. BTC does not block ETH, and ETH does not block BTC.
Exit behavior: Existing owned positions may close only through a reduce-only Exit. Scale-in, duplicate same-symbol legs, unknown symbols and open-order ambiguity trigger safe-off.

Monitoring and Alerting

Shadow/Live Drift ist monitor-only: Shadow signals are compared with production signals for drift detection, but Shadow is not a live execution prerequisite.
Auto safe-off: Drift, Kraken/Telemetry mismatch, unexpected open orders, duplicate positions or executor blocker switch live trading back to safe-off and raise a message.
Protected Shadow statuses: Strategies in live or waiting for the live gate continue Shadow observation as live_monitoring or live_gate_monitoring and are retirement-protected.

Table Columns

Annual median: Median annualized net return across markets when the artifact provides annualized values, otherwise the reported median return.
Annual worst: The weakest market result after annualization. This guards against one market carrying the whole strategy.
DD: Maximum drawdown in the backtest or shadow run.
PF: Profit factor. Values above 1 mean gross wins exceeded gross losses in the tested sample.
MtM: Mark-to-market return since shadow activation, including conservative exit-cost handling when present in the artifact.

Strategy Families

ema_regime_dual: Trendfolge ueber schnelle und langsame EMAs. Die Strategie handelt nur, wenn das Marktregime klar genug wirkt und versucht Seitwaertsphasen zu meiden.
supertrend_dual: ATR-basierte Trendfolge. Sie reagiert auf Trendwechsel ueber Supertrend-Linien und versucht frueh in neue Bewegungen hineinzugehen.
volatility_compression_breakout_dual: Breakout nach Ruhephase. Erst wird enge Volatilitaet gesucht, dann wird der Ausbruch aus dieser Kompression gehandelt.
trend_pullback_dual: Trend mit Ruecksetzer-Entry. Statt nackte Breakouts zu jagen, wartet die Strategie auf Pullbacks innerhalb eines bestaetigten Trends.
adx_breakout_dual: Breakout mit Trendstaerke-Filter. ADX soll helfen, schwache oder verrauschte Ausbrueche auszusortieren.
weekday_session_trend_dual: Verbindet Trendlogik mit wiederkehrenden Mustern aus Wochentag und Handelssession. Ziel ist, zeitliche Marktstruktur auszunutzen.
attention_breakout_dual: Breakout plus Aufmerksamkeitsschub. Preisbewegung wird nur bevorzugt, wenn zusaetzlich Narrativ- oder Attention-Signale den Move stuetzen.
onchain_impulse_dual: Preislogik plus On-Chain-Aktivitaet. Die Strategie sucht handelbare Impulse, wenn Blockchain-Daten auf echte Aktivitaet oder Spannungen hindeuten.
crowding_reversion_dual: Antizyklische Rueckkehr zum Mittelwert. Sie sucht ueberfuellte oder einseitig positionierte Maerkte und setzt auf Gegenbewegungen.
oi_confirmation_breakout_dual: Breakout mit Open-Interest-Bestaetigung. Ein Ausbruch gilt nur dann als glaubwuerdig, wenn die Positionierung ihn mittraegt.
fear_greed_reversal_dual: Sentiment-Reversal. Extreme Angst- oder Gierphasen werden als moegliche Wendepunkte interpretiert.
fear_moon_disagreement_dual: Handelt Widersprueche zwischen euphorischen Narrativen und breiterem Sentiment. Solche Dissonanzen koennen Reversal- oder Filter-Signale liefern.
session_open_attention_dual: Fokus auf Session-Open plus Aufmerksamkeit. Sie prueft, ob Marktstart und externer Impuls zusammen genug Qualitaet fuer einen Trade ergeben.
funding_squeeze_failure_dual: Sucht ueberdehnte Positionierung ueber Funding-Druck und handelt die Gegenbewegung, wenn der erwartete Squeeze scheitert.
oi_flush_reclaim_dual: Beobachtet Open-Interest-Spuelungen und versucht den Reclaim nach Kapitulation oder Positionsbereinigung zu handeln.
premium_compression_breakout_dual: Nutzt enge Premium-Spannen als Ruhephase und setzt auf den Ausbruch, wenn Preis und Derivate-Premium wieder auseinanderlaufen.
premium_disagreement_reversal_dual: Sucht Reversals, wenn Spot-Preis und Derivate-Premium in unterschiedliche Richtungen deuten.
delayed_crowding_trend_dual: Kombiniert ueberlaufene Positionierung mit verzoegerter Trendfortsetzung. Nicht sofort gegen den Move, sondern nach Bereinigung wieder mit ihm.
funding_flip_breakout_dual: Handelt Breakouts rund um deutliche Funding-Regimewechsel, wenn sich die Positionierung schnell auf eine neue Seite dreht.
attention_exhaustion_reversal_dual: Sucht das Ende ueberhitzter Aufmerksamkeit. Wenn Narrativ und Aufmerksamkeit auslaufen, setzt die Strategie auf Reversal statt Fortsetzung.
fear_rotation_disagreement_dual: Nutzen Widersprueche zwischen Angst-Signalen und Rotationen im Markt, um moegliche Wendepunkte oder Filter zu finden.
month_turn_attention_dual: Koppelt Monatswechsel-Effekte mit Aufmerksamkeitssignalen. Ziel ist, wiederkehrende Turn-of-Month-Impulse nur dann zu handeln, wenn auch externes Interesse mitzieht.
onchain_rotation_breakout_dual: Sucht Breakouts, bei denen On-Chain-Rotation oder Kapitalbewegung den Richtungswechsel bestaetigt.
weekend_pressure_reversal_dual: Reagiert auf typische Wochenend-Verwerfungen und versucht die Gegenbewegung nach ueberschossenem Druck mitzunehmen.
unknown: Die Familie ist aus dem Artefakt nicht sauber ableitbar. Dann hilft nur die konkrete Kennzahlentabelle und der Strategiename selbst.

Retirement Reasons

solid_gate_failed_on_latest_run: Die Strategie war frueher im Shadow-Set, hat aber im neuesten Hauptlauf das Solid-Gate nicht mehr bestanden und wurde deshalb aus dem aktiven Monitoring genommen.
shadow_solid_gate_failed_streak: Das Shadow-Solid-Gate wurde ueber mehrere neue Beobachtungen in Folge verfehlt. Das ist ein Drift-/Qualitaets-Exit, kein fehlender Artefaktgrund.
shadow_profit_factor_drift: Der Profit Factor lag ueber mehrere neue Beobachtungen in Folge unter der Shadow-Schwelle.
shadow_runtime_spec_missing: Die Runtime-Spezifikation zur Strategie fehlte und konnte nicht rekonstruiert werden; der Kandidat wurde fail-closed aus dem aktiven Shadow entfernt.
shadow_data_integrity_failure: Kritischer Daten- oder State-Fehler im Gate. Der Kandidat wird sofort retired, damit kein kaputter State weiterlaeuft.
shadow_max_drawdown_breach: Der Shadow-Drawdown hat die konfigurierte Exit-Schwelle ueberschritten.
shadow_rolling_return_breach: Der rolling Return blieb ueber die konfigurierte Anzahl Auswertungen unter der Mindestschwelle.
shadow_inactivity_weak_performance: Der Kandidat war zu lange ohne neuen Trade und zugleich Mark-to-Market schwach.
unknown: Der Shadow-Status nennt keinen klaren Retirement-Grund. In dem Fall sollte das Artifact kuenftig einen maschinenlesbaren Grund plus Kurztext liefern.

Decisions

Valid: A strategy passed the artifact's baseline or solid criteria and is displayed as a candidate for deeper review.
Exploratory: A leaderboard row that remains useful for context but should not be treated as a deployment candidate.
Shadow tracking: Signals, fills, positions and drawdown are simulated on new data. It is a monitoring gate before real execution.

Cost Model

Cost model names identify the fee and slippage assumptions used by the trader artifact. Strict taker models are more conservative than blended or maker-lean models. A strategy that only works in optimistic models is fragile.

Position

flat: No open simulated exposure.
long: The strategy is simulated as holding the asset after a long entry signal.
short: The strategy is simulated as holding short exposure where the venue and instrument allow it.

Venue

Venue describes where the strategy can realistically be executed, such as Binance Spot or Binance USD-M futures. Spot venues and futures venues have different instruments, fee models, execution constraints and live-readiness.

Statuses

active: The strategy is currently tracked by the shadow status artifact.
live_monitoring: The strategy is already live-active or live-owned and continues in Shadow as monitoring-only evidence. It remains retirement-protected while that live ownership exists.
live_gate_monitoring: The strategy is waiting in the live gate and continues in Shadow with retirement-protected monitoring status until the gate changes.
retired: The strategy was removed from active shadow tracking but remains visible for review.
retirement-protected: The strategy is exempt from automatic Shadow retirement because it is connected to a live or live-gate state. Drift still alerts; protection does not hide execution problems.
robust: The strategy passed the artifact's robustness checks across the relevant markets.
fragile / exploratory: The strategy appears in broader ranking context but fails one or more practical gates.

Info / Help