feat: May 2026 API updates — trading system audit pipeline (1057 tests)#28
Open
redact22 wants to merge 19 commits into
Open
feat: May 2026 API updates — trading system audit pipeline (1057 tests)#28redact22 wants to merge 19 commits into
redact22 wants to merge 19 commits into
Conversation
…lation Request translation (messages, tools, images, generation config), streaming SSE response translation, non-streaming JSON translation, tool use roundtrip with id→name mapping. Auto-routing: Haiku→Gemini primary, Sonnet/Opus→DeepSeek primary with cross-failover. Live switching via /_proxy/mode endpoint. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Alpha-parser: fix 3 engine.mjs bugs (dead code, sub-million display, false-positive low_holders), add hermetic test layer with injectable fetch, remove dead live mode CLI stub, add mock fixtures for offline tests. Audit tools: implement stop-loss/take-profit/SMA crossover exit logic in backtester, make startingEquity configurable across all functions, thread --equity CLI flag through to audit(). Gemini translator: map SAFETY→safety_blocked and RECITATION→recitation_blocked as distinct stop reasons, add console.warn for silently dropped unknown roles. Add package.json with "type": "module" and test scripts. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
111 tests, 27 suites, 0 failures. Covers: Gemini translator (16), model-proxy routing/remap/sanitization (52), audit schemas/performance/ backtest (22), alpha-parser token scanning/audit/DEX integration (21). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…s pass Quant audit fixes across 5 modules: - gemini-translator: tool_result content extraction, FunctionResponse id field, thinking_level mapping, content_block_start input, output_config handling - backtest: look-ahead bias, SMA window alignment, exit_reason tracking, double-counted slippage - perf-audit: Sortino Infinity cap, trades type guard, equity curve start, unused param cleanup - alpha-parser: RugCheck fallback activation, API failure logging, parallel batch processing (concurrency=5) - model-proxy: adaptive thinking normalization, sampling param stripping, output_config migration, assistant prefill filter, OAuth key warning - start-proxy: dead code removal, --mode flag bounds check Tests: 143 pass, 0 fail across 32 suites (+53 new tests). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…8 tests) - proxy/observability.js: structured JSON logging, request metrics, Prometheus output, latency histograms, health reports - proxy/observability.test.js: 18 tests across 5 suites - proxy/circuit-breaker.js: 3-state breaker, retry with jitter, token bucket rate limiter, backend health tracker, fallback selection - audit/advanced-backtest.mjs: Monte Carlo simulation, walk-forward optimization, strategy comparison (Welch's t-test), drawdown analysis, equity curve metrics - proxy/model-proxy.js: 106 tests — normalizeThinkingBlocks, stripSamplingParamsOnThinking, normalizeOutputConfig, filterAssistantPrefill, warnOAuthToken, stripDeepSeekUnsupportedFields, mapEffortToReasoning, ANTHROPIC_THINKING_BACKENDS - shell scripts: May 2026 pricing, Gemini benchmark, auto-mode docs Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
CircuitBreaker, withRetry, TokenBucket, BackendHealthTracker, selectFallback — all 7 suites green. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…er (264 tests) - model-proxy.js: structured JSON logging replaces console.log, per-request correlation IDs, /_proxy/health + /_proxy/metrics endpoints, CircuitBreaker per backend, withRetry wrapping, BackendHealthTracker integration, graceful fallback selection - cost-tracker.js: per-session cost tracking, token counting, backend comparison, formatted reports, active DeepSeek 75% discount - cost-tracker.test.js: 24 tests covering computeCost, createCostTracker, savings comparison Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…o signals (530 tests) - audit/microstructure.mjs: OrderBookAnalyzer (bid/ask imbalance, spoofing detection, iceberg orders, volume footprint, absorption, VPIN toxicity) + 56 tests - audit/liquidity-flow.mjs: VolumeProfile (VAP, value area, high/low volume nodes, skew), OrderFlowMomentum (momentum, absorption, cumulative delta), VolatilityEddyDetector (eddies, turbulence index) + 43 tests - audit/signal-fusion.mjs: SignalFusionEngine (weighted/bayesian/voting fusion, composite scores, BUY/SELL/HOLD decisions, correlation matrix), SignalQualityAnalyzer (accuracy tracking, weight calibration) + 80 tests - alpha-parser/crypto-signals.mjs: HMAC-signed signal envelopes, verification, provenance chains, replay guard, key derivation, signal bundles + 39 tests - audit/advanced-backtest.test.js: 48 tests (Monte Carlo, walk-forward, strategy comparison, drawdown analysis, equity curve metrics) Inspired by Leonardo da Vinci's polymathic approach — anatomical market dissection, fluid-dynamics liquidity modeling, cross-domain signal fusion. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Zone Detector: order blocks, FVGs, breaker zones, liquidity voids, streaming mode. Decision Graph: record/link/query trading decisions, pattern mining, graph queries. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Classifies 8 market regimes (trending, ranging, accumulation, distribution, breakout, breakdown, volatile) with confidence, strength, and actionable trading parameters (stop multiplier, position sizing, TP aggressiveness). Includes streaming classifier with regime-change detection and multi-TF confluence. 45 tests. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Wires Zone Detector + Market Regime + Signal Fusion + Order Book Microstructure into a single `run()` call that produces complete trading decisions (BUY/SELL/HOLD) with confidence, rationale, zone context, and strategy parameters. Includes `runBacktest()` for historical walk-forward simulation with stop-loss/take-profit, cooldown bars, and per-regime PnL stats. 37 tests. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Swing points, BOS/CHoCH, order blocks, mitigation blocks, breaker blocks, imbalance/FVG, inducement detection, SMC trend analysis, liquidity levels. Zero deps, ESM. Built via parallel agent dispatch. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…86 tests) Trade CLI: single-command pipeline (--analyze, --backtest, --live) Multi-TF Confluence: zone clustering, heatmap, confluence scoring (75 tests) Terminal Dashboard: ANSI renderer, colorize, live streaming (69 tests) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The orchestrator passes { minConfidence } to getDecision() but the
method ignored it and always used the constructor default (0.6).
Now accepts opts.minConfidence to allow dynamic threshold tuning.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…to-end pipeline Optimizer: gridSearch, hillClimb, optimize, optimizePerRegime, sensitivity (18 tests) Grail Demo: --quick, --optimize, --full modes with dashboard + verdict (14 tests) Orchestrator: wired confluence zoneConfluenceScore into signal pipeline Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ation Previously signalsFromZones emitted both support AND resistance signals, which canceled in the fusion engine (~0 composite score). Now only the dominant direction is emitted when the strength difference exceeds the significance threshold (0.15). RR signal also requires clear asymmetry (>1.5 or <0.67). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ilures Binance OHLCV fetcher with geo-block detection, synthetic fallback caching, 7 trading presets (BTC/ETH/SOL/MBT/DEFAULT/AGGRESSIVE/SCALPING), and a multi-symbol optimizer scanner with ranking. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The orchestrator emitted signals with sources 'zone-detector', 'market-regime', 'confluence' but SignalFusionEngine silently dropped them (source not in VALID_SOURCES) and assigned zero weight (not in DEFAULT_WEIGHTS). Every backtest since the orchestration rewrite produced 0 trades. Live Scanner ties datafeed + optimizer + presets + grail into a single pipeline command. 3-symbol scan completes in 2.3s with ranked GRAIL verdicts. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
PaperTrader: simulated live trading with position management, SL/TP, equity curve, Sharpe/maxDD/PF metrics, JSON export. MTF scanner: single-symbol multi-timeframe analysis with directional conviction scoring. Correlation: Pearson matrix, portfolio variance, diversification scoring. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
deepclaude/proxy/start-proxy.js
Lines 15 to 18 in bf09907
Detect legacy mode without consuming flag arguments
The startup mode detection treats process.argv[2] and [3] as <targetUrl> <apiKey> unconditionally, so flag-based invocations like node proxy/start-proxy.js --mode auto or --port 3300 are misclassified as legacy mode (targetUrl='--mode', apiKey='auto') and then fail when new URL('--mode') is evaluated. This breaks the standalone control-server path that should accept --mode/--port options.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| @@ -254,7 +549,7 @@ export function startModelProxy({ targetUrl, apiKey, startPort = 3200, backends, | |||
| clientRes.end(JSON.stringify(result)); | |||
| return; | |||
| } | |||
| console.log(`[MODEL-PROXY] Mode switched: ${result.previous} → ${result.mode}`); | |||
| log.info(`Mode switched: ${result.previous} → ${result.mode}`, { correlationId: cid, reqId }); | |||
There was a problem hiding this comment.
A POST /_proxy/mode request can crash the proxy because the handler logs with cid and reqId before those const bindings are initialized on this code path. The control-endpoint branch returns before reaching the later declarations, so the callback hits a temporal-dead-zone ReferenceError and terminates the Node process, making backend switching unusable and allowing a single request to bring down the proxy.
Useful? React with 👍 / 👎.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
feature/may-2026-api-updatesaudit/directory (zero changes to main CLI)What shipped
Data & Pipeline
Trading Engine
Scanners & Analysis
Critical Fixes
Test plan
node --test audit/*.test.js— 1057 tests, 0 failuresnode audit/live-scanner.mjs --symbols BTC-USD,ETH-USD --days 10node audit/trade.mjsintegrationdeepclaude.ps1,deepclaude.sh,proxy/)🤖 Generated with Claude Code