ta4j Files

0.22.4 (2026-03-15)

Added

• Window-aware criterion evaluation API: AnalysisCriterion can now analyze exactly the slice you care about, including specific bar ranges, date/time ranges, lookback bars, and lookback durations, via AnalysisWindow/AnalysisContext and AnalysisCriterion#calculate(series, tradingRecord, window[, context]), with strict/clamp history policies and configurable open-position handling.
• Price-structure aggregators: Added RangeBarAggregator, VolumeBarAggregator, and RenkoBarAggregator for range-, volume-, and Renko-brick derived bar series with externally configurable thresholds and comprehensive fixture-driven regression coverage.
• One-shot multi-timeframe Elliott Wave analysis: You can now run ElliottWaveAnalysisRunner once and get an ElliottWaveAnalysisResult with base + neighboring degree outputs, ranked scenarios, and confidence context in one place.
• Reusable walk-forward framework with backtest integration: Added WalkForwardEngine, WalkForwardTuner, WalkForwardObjective, and StrategyWalkForwardExecutor, plus BacktestExecutor entrypoints so you can run backtest-only, walk-forward-only, or both in one consistent flow.
• Weighted strategy ranking across execution results: TradingStatementExecutionResult and BacktestExecutionResult#getTopStrategiesWeighted(...) now support normalized weighted ranking (for example net profit + drawdown + trade count) with pluggable normalization and deterministic ordering. You can jump in with WeightedCriterion.of(...), RankingProfile.weighted(...), or the direct weighted overloads, and the README plus SimpleMovingAverageRangeBacktest now show a concrete “net profit + RoMaD” shortlist flow you can copy directly.
• Shared scoring/weighting primitives for library extensions: Added NamedScoreFunction<I, S> and WeightedValue<T> so indicator, confidence, and walk-forward components can reuse the same scoring and weighted-aggregation contracts.
• Live Elliott preset demo support: ElliottWavePresetDemo now accepts live tickers (for example BTC-USD, ETH-USD, SPY) so you can run the same EW workflow on non-ossified daily data.

Changed (Trading Record and Execution Flow)

• Elliott APIs and demos now follow runner-centric naming and defaults: The project has moved from legacy analyzer naming to ElliottWaveAnalysisRunner, examples are organized under analysis.elliottwave.{demo,backtest,support}, and demo defaults now emphasize auto-degree selection with multi-degree context.
• HighRewardElliottWaveStrategy momentum confirmation now uses MACD-V: The strategy now uses VolatilityNormalizedMACDIndicator and drops redundant exit-rule guarding to keep rule flow cleaner.
• Release automation now favors safer incremental bumps: release-scheduler.yml and semver-rules-override.txt now drive explicit go/no-go decisions with patch|minor outputs only, normalize noisy AI bump values (for example MAJOR or minor), and keep major bumps disabled so automated releases stay predictable for library consumers and maintainers (#1477).
• @since policy is now explicit for contributors: .github/CONTRIBUTING.md now clearly requires introducing release versions without -SNAPSHOT (for example @since 0.22.4), plus a documented 5-minor volatility window so teams can adopt new APIs with clearer risk expectations (#1477).

Fixed

• Net momentum can now jump straight to later bars without falling over: NetMomentumIndicator now handles large first-lookups and constrained maximumBarCount series without blowing the stack, so replay/backtest flows can request a late bar first, warm up on pruned rolling windows, and keep the same momentum values they would get from sequential evaluation.
• Release workflow drift guardrails: Release maintainers now get near-immediate drift detection because release-health.yml runs on every master push and after Publish Release to Maven Central completes, and publish-release.yml now hard-fails if the pushed release tag does not resolve to the expected releaseCommit or is not reachable from origin/<default_branch>.
• Rolling variance now matches sample-statistics expectations by default: VarianceIndicator now computes sample variance out of the box (n-1 divisor), so spreadsheet-style checks like issue #1152 line up directly. You can now choose behavior explicitly with SampleType/factory helpers across VarianceIndicator, StandardDeviationIndicator, StandardErrorIndicator, SigmaIndicator, and CorrelationCoefficientIndicator (for example: VarianceIndicator.ofSample(...), VarianceIndicator.ofPopulation(...), StandardDeviationIndicator.ofSample(...), or CorrelationCoefficientIndicator.ofPopulation(...)).
• Enter-and-hold wrappers now keep return format metadata intact: EnterAndHoldCriterion now forwards getReturnRepresentation() from its wrapped criterion, so downstream consumers can reliably detect whether outputs are decimal, percentage, multiplicative, or log without special casing wrapper criteria.
• Elliott analysis hardening: Enforced bounded ElliottDegree.RecommendedHistory ranges, hardened ossified resource loading for classpath edge cases, and simplified redundant trend-bias null guarding in EW analysis reporting.
• Walk-forward fold metadata stability: Fixed fold-value reporting so criterion maps and fold views remain stable and deterministic when consumers rely on fold order for downstream comparisons.

Breaking

• Trade and record handling now has one obvious happy path: Build fill-aware trades with Trade.fromFill(...) or Trade.fromFills(...), then pass them to TradingRecord#operate(...). The older live-only wrappers (ExecutionFill, LiveTrade, LiveTradingRecord, and PositionLedger) are still available in 0.22.x as deprecated migration shims, but they are no longer the API you should reach for first.
• Open-position APIs now use Position end to end: TradingRecord#getOpenPositions() returns open Position snapshots, getCurrentPosition() remains the canonical net-open view, and getNetOpenPosition() is now just a compatibility alias. The separate OpenPosition type is gone.
• Lot bookkeeping is finally internal again: PositionBook and PositionLot now stay inside BaseTradingRecord, while FIFO, LIFO, average-cost, and specific-id matching keep the same external behavior.

Changed (Backtest Execution Models and Custom Records)

• Partial-fill recording is smoother in live-style flows: You can now stream one fill at a time with TradingRecord.operate(fill) when fills arrive incrementally, or keep using Trade.fromFills(...) plus operate(...) when you already have the whole batch for one logical order. TradeFillRecordingExample in ta4j-examples now walks through both styles and also shows how FIFO, LIFO, AVG_COST, and SPECIFIC_ID change partial-exit matching.
• Bring your own trading record in backtests: BarSeriesManager can now run directly against a caller-provided TradingRecord (run(strategy, tradingRecord[, amount, start, end])) and can also be configured with a default TradingRecordFactory, so you can keep standard BaseTradingRecord runs or wire custom record implementations without changing existing run(...) calls.
• BacktestExecutor now picks up the same backtest wiring without extra boilerplate: You can construct it directly with a TradeExecutionModel for the common slippage/stop-limit case, or hand it a preconfigured BarSeriesManager so custom TradingRecordFactory behavior flows through normal backtest, top-K, and walk-forward execution.
• Execution-model examples are easier to copy straight into your own backtests: TradingRecordParityBacktest in ta4j-examples now walks through next-open, current-close, and slippage fills side by side, then verifies the same behavior with provided and factory-configured BaseTradingRecord runs.
• Stop-limit/live parity hardening: StopLimitExecutionModel now expires stale pending orders before accepting new signals (so old orders cannot block fresh ones), commits partial expiry fills on unified BaseTradingRecord exit flows for better real-world fill progression, and keeps rejection metadata for the unfilled remainder.