CRYPTYX — The Intelligence Layer for Digital Assets

One-call agent grounding bundle. Returns latest anchor `asof_day` plus per-asset payload across four slim queries: `profile` (comp_score, comp_rank_eq, comp_rank_pct, ret_1d/7d/30d, vol_rv_7d/30d, sharpe_7d/30d), `factors[class][horizon]` cube of `t_score_pct` + `t_band` across 8 classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), `regime.primary` and `regime.secondary` with confidence + `resolution_mode`, and raw `liquidity` OB depth row. ~200-asset universe, single aligned `asof_day`. Drop into a prompt for quant-aware agents.

Per-asset bundle of latest atomic + composite signal snapshot plus N-day event lookback, pulled from four api.v_*_ux views and grouped by asset. Response: {ok, asof_day, assets[]} where each asset carries {asset, asof_day, regime (expansion/consolidation/deleveraging), regime_confidence, atomic:{snapshot,events}, composite:{snapshot,events}}. 143 atomic signals across 8 classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL) and 9 IC-weighted composites, ~200 assets. Filter via assets (CSV), events_days (1-90, default 7), min_conf_pct (0-100). One call carries the full signal picture + regime gating for an agent prompt.

Single-asset factor t-scores across CRYPTYX's 8 canonical classes (CORR, EFF, FLOW, FUT, OB, OPT, TR, VOL) and 8 horizons (1d/7d/14d/30d/60d/90d/180d/365d), sourced from the policy-aware metric stack (440+ winsorized, z-capped metrics rolled up by class x horizon). Returns a flat top-level array of rows with asof_day, asset, horizon, class, t_score_pct, t_band. mode=snapshot returns the latest anchor day; mode=series returns a daily window (up to 365d) optionally filtered to specific horizons. Use to inspect one asset's factor surface for grounding before composite/signal calls.

Per-asset spot order-book depth timeseries from api.asset_liquidity_timeseries_v1: ob_bid_50bp_usd, ob_ask_50bp_usd, ob_bid_100bp_usd, ob_ask_100bp_usd, ob_bid_200bp_usd, ob_ask_200bp_usd, plus spot_vol_1d_usd, spot_vol_7d_avg_usd, spot_vol_30d_avg_usd. Daily aggregate anchored to UTC−1, full history or trailing days window. Handler returns a bare JSON array of rows ordered by asof_day ASC; include=futures switches the envelope to {spot:[...], futures:{...}} where futures carries latest-day fut_ob_bid/ask_*bp_usd from compat.fut_ob_agg_1d_*. Use for pre-trade depth checks and venue routing across ~200 CRYPTYX-tracked assets.

Bare JSON array of per-asset rows from api.asset_universe_timeseries_v1 (analytics.v_composite_ux + TR_RET + VOL_RV). Snapshot mode returns latest canonical day cross-section; series mode returns last N days (default 30, max 365). Filter by assets= and/or universes= CSV. Each snapshot row: asof_day, asset, universe_tag, comp_score, comp_rank_eq, comp_rank_pct, price_anchor, ret_1d/7d/30d, vol_rv_7d/30d, sharpe_7d/30d, plus macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence`. ~200-asset master ranking feed for portfolio construction and rotation, with regime gating in the same call.

Daily aggregated-exchanges close funding rate (fr_close) for one asset, last N days, sourced from compat.fut_funding_1d (CoinGlass/Deribit, UTC-1 anchor day, ALL_EXCHANGES scope). Response is a raw JSON array of {asof_day, asset, fr_close} rows ordered DESC — no wrapper object. No annualization, no z-score, no FUT_FUNDING_STRESS signal context — use /api/ai/signals or composite endpoints for enrichment. Cheap historical funding feed for carry-trade backtests and crowding diagnostics.

Daily OHLCV bars for a single asset over a configurable lookback (1-1825 days, default 90), anchored to the latest pipeline day. Response is a bare JSON array of rows with keys asof_day, asset, open_px, high_px, low_px, close_px, volume_usd, sourced from api.v_daily_ohlcv. UTC-1 canonical anchor, ~200-asset universe, up to 5 years of backfilled history. Use as the grounded spot price tape feeding all CRYPTYX factor scores and signals.

Universe-wide factor breadth from metrics.factor_scores, bucketed by (asof_day, horizon, factor class). Each row returns positive_count (t>=1.0), negative_count (t<=-1.0), neutral_count, total, avg_t_score, expand_share_pct, factor_state (EXPANSION/COMPRESSION/NEUTRAL) and dominant_regime_id across ~200 active assets. 8 factor classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL). Series mode returns the lookback window; snapshot collapses to latest_day. Filter by horizons/classes. Use to time risk-on/risk-off rotations and breadth divergence.

Universe-aggregate cross-factor-class divergence detector. Averages t_scores across ~200 active assets from metrics.factor_scores at the latest asof_day per horizon (excludes OPT), then runs three hardcoded patterns over TR/VOL/FLOW only: Ignition (TR+VOL extreme), Capitulation (FLOW+TR collapsed), Distribution (outflows while trend holds). Live path uses fixed thresholds, not the adaptive percentile engine. Returns ok, asof_day, total_alerts, by_type counts, by_horizon grouping, ranked alerts array (severity-weighted by horizon), durationMs. Market-wide regime read, not per-asset; use for top-down inflection scanning.

Macro-regime synthesis from analytics.macro_regime_day — NOT per-asset. Returns ok, asof_day, current (object keyed by horizon: 1d/7d/14d/30d/60d/90d/180d/365d, each with regime_label, regime_confidence, cross_class_state, breadth JSONB, horizon_alignment JSONB, n_assets, meta JSONB), history (raw rows over lookback window), count, durationMs. Built on ~200 assets, 8 factor classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL). Filter to one horizon via horizon param; slim=true drops JSONB blobs for timeline use. For per-asset regimes hit /api/asset-regimes. Read multi-horizon risk-on/off posture in one call.

Latest-day firing signals across ~200 assets, returned as `{ atomic, composite }`. Atomic rows: signal_id, signal_name, class (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), horizon, tags, z, confidence, confidence_pct, confidence_band, event-delta (event_type, prev/new_triggered, prev/new_confidence, delta_confidence), plus macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence` per row. Composite rows: composite_id, title, composite_type, same confidence + event-delta + regime block. Filtered to live registry; excludes anti-predictive signals via analytics.signal_health.

Enabled signal registry. Per row: signal_id, name, description, class (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), horizon, tags, geometry_type, active_version, active_params, params_created_at, recent_total_assets, recent_triggered_assets, recent_days_evaluated, unique_assets_observed, unique_assets_ever_triggered, trigger_rate_per_asset_day, health_grade, predictive_grade, operational_grade, plus a `walk_forward` block (latest 180d/70-30 IS/OOS split from analytics.signal_eval_wf_day: is_ic, is_hit_rate, oos_ic, oos_hit_rate, overfit_ratio, ic_delta, reliability_grade green/yellow/red). Top-level: ok, count, signals, meta.ic_context='cross_sectional', durationMs. logic_sql withheld (protected IP). Discovery surface before backtest or fork.

Per (asset, day) factor t-score breakdown plus composite score, wrapped under an `explanation` JSONB blob with the queried `signal_id`, `day`, `asset` echoed back at the top level. Surfaces class x horizon t-scores (CORR, EFF, FLOW, FUT, OB, OPT, TR, VOL) and the weighted composite from metrics.factor_scores and metrics.composite. Note: the underlying signals.signal_explain() function does not currently read signal_log or signal_registry, so the explanation is asset-day scoped, not signal-trigger scoped. Use for inspecting which factor classes were hot on a given asset-day, not for a per-signal trigger justification.

Signal trigger-activity leaderboard over a configurable lookback (window=7-365 days, default 30). Reads signals.signal_log and joins signals.signal_registry to return per-signal rows: signal_id, total_days, triggered_days, trigger_events, trigger_rate_pct, avg_confidence, confidence_std, assets_triggered, last_triggered, class, horizon, plus a per-horizon `walk_forward` block (latest 180d/70-30 IS/OOS split: is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade) so an agent reading the leaderboard sees firing frequency AND robustness in one pass. Ranks 143 atomic + 9 composite signals across ~200 assets by avg_confidence DESC, then trigger_rate_pct.

Recently triggered atomic signal events from CRYPTYX's 143-signal registry, served from api.v_signals_atomic_ux. Returns up to 50 events as { events: [{ asset, signal_id, confidence, triggered_at }] } where triggered_at is the day-level asof_day, filtered to is_triggered=true. Single query param: since (ISO date, default 24h back, clamped to max 7 days). Ordered by asof_day DESC then asset. Use as a change-detection digest of which atomic signals fired across ~200 assets — pair with /api/signals/catalog or /api/signals/explain to resolve class, geometry, grade, and regime context.

Curated top-10 atomic signals from CRYPTYX's 143-signal registry, gated to enabled + non-anti-predictive + A/B health-graded names ranked by 7d cross-sectional IC. Each row returns signal_id, name, description, class (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), horizon, tags, geometry_type, active_version, active_params, ic_7d, hit_rate_7d, n_obs_7d, n_triggers_7d, health_grade, unique_assets_observed, unique_assets_ever_triggered, trigger_rate_per_asset_day, plus a 7d `walk_forward` block (latest IS/OOS split: is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade) so the in-sample IC and the OOS check sit side by side. Response also carries count, selection_criteria block, durationMs. Default discovery surface MCP agents call before /api/signals/catalog to short-list which signals to wire into a live strategy.

Last N days (default 30, max 365) of taker aggressor flow per asset, split into spot and futures arrays of recent rows with asof_day, asset, taker_buy_usd, taker_sell_usd, bs_ratio - pass-through of compat.spot_taker_bs_1d and compat.fut_takerbs_1d with no z-score, FLOW factor t-score, regime label, or signal context. Feeds custom flow models on CRYPTYX cleaned spot and perp taker tape

One-shot LLM grounding snapshot. Six parallel reads in ~200-400ms. Returns asofDay, factorBreadth (assetsWithScores + activeClasses array of CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), compositeScores (topAssets top-10 + bottomAssets bottom-5, each carrying macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence`), recentSignals (day, totalTriggered, bySignal aggregation), platformCounts (metrics, signals, classes), pipelineStatus (lastRunDay, status, totalDurationMs). Optional ?asset= adds perAssetMetrics: top-10 metrics ranked by absolute time-series IC with TS + CS health grades. Call before any market reasoning.

Single-call LLM grounding bundle — top 10 and bottom 5 composite-ranked assets, all signals triggered today, 30-day factor breadth across 8 classes, plus a ready-to-drop plaintext brief for an agent prompt.

Service health probe — build version, last pipeline run, current signal/composite/asset counts. Sub-second, no DB writes. The cheapest end-to-end pay-per-call check before issuing real queries.

Compact metric feed shaped for agent prompts — z-scores, percentiles, t-scores by factor class for chosen assets. Same coverage as /api/asset-factors but optimized for LLM context efficiency. Use as raw context for quant-aware agents.

Compact signal feed shaped for agent prompts — normalized field names, comma-separated asset filter, JSON-only. Same 143-signal coverage as /api/signals/active but optimized for LLM context efficiency. Use when injecting signal data directly into an agent's prompt.

Class × horizon factor t-scores per asset across CRYPTYX's 8 canonical factor classes (CORR, EFF, FLOW, FUT, OB, OPT, TR, VOL). Computed from 440+ underlying metrics with policy-aware winsorization and z-capping. Use for cross-sectional ranking, sector rotation, and factor exposure analysis.

Order-book depth per asset at 50, 100, and 200 basis points — bid USD notional, ask USD notional, imbalance ratio. Aggregated across Binance spot books, refreshed every 15 minutes. Use for slippage estimation, exchange routing, and pre-trade liquidity checks.

Macro regime classification per asset — expansion / consolidation / contraction — with confidence (0-1). Regimes are factor-derived (not price-derived), so they shift before price action confirms. Use to gate strategies by market state — mean-reversion logic fails in trending regimes.

Composite scores and ranks across ~200 digital assets — comp_score, comp_rank, 1d / 7d returns, factor breadth, conviction count. Snapshot or time-series mode. Use as the master ranking feed for portfolio construction, watchlist generation, and rotation strategies.

Perpetual futures funding rate per asset — current rate, annualized, z-score vs trailing 30-day distribution. Sourced from Deribit and CoinGlass aggregated feeds. Use for carry trades, crowding detection, and timing perp entries when funding stretches to extremes.

Daily OHLCV for an asset (up to 365 days lookback) — open, high, low, close, volume. UTC−1 canonical anchor day. Sourced from CoinMarketCap. Use as the spot price feed underlying all CRYPTYX metrics — same data our factor scores compute on.

Market-wide signal pulse — total triggers across 143 atomic signals, positive/negative counts per factor class, net breadth, asof day. Single-call market state snapshot for the universe of ~200 assets. Use to time risk-on / risk-off shifts and detect early breadth divergence.

Cross-factor-class divergence flags per asset (e.g. TR bullish while VOL bearish) with spread magnitude. Configurable horizons. Use to find non-consensus setups before they become crowded — divergences mark inflection points in factor agreement.

Regime distribution across the universe — count of assets in each macro regime (expansion/consolidation/contraction), per-asset regime + confidence, summary totals. Use to read the breadth of risk-on/off positioning across the universe at a glance.

All currently-firing atomic and composite signals across ~200 digital assets — signal_id, asset, triggered, confidence (0-1), geometry_type (5 types: single_metric / state_transition / cross_timeframe / multi_factor / regime_conditional). 143 atomic signals + 9 IC-weighted composites, walk-forward validated against 5+ years of price data. Use as a live "what is lighting up right now" feed for agent reasoning.

Full signal registry — 143 atomic signals and 9 IC-weighted composites (6 thematic, 3 regime) with active parameters, 30-day trigger statistics, geometry type, factor class (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), and category. Logic SQL deliberately excluded (protected IP). Use as the canonical capability-discovery call before backtests or live signal subscriptions.

Per-day explainability for a single (signal × asset × day) — full trigger reason, factor scores by class, composite score, parameters in effect, regime context. Audit-grade output. Use when an agent (or human) needs to justify a position or audit a model's reasoning.

Ranked signal leaderboard with IC (Spearman rank correlation), hit rate, IR (information ratio), grade, and asset coverage at a chosen forward horizon. Evaluated against ~200 assets × 5+ years of returns. Use to surface the strongest predictors across the universe and benchmark new signals.

Recent signal firings stream (last N events) — signal_id, asset, day, triggered, confidence — sorted newest-first. Use for change-detection feeds, agent digests, and "what changed since I last checked" summaries.

Top-grade signals ranked by IC-weighted performance at a forward horizon (7d / 14d / 30d) — signal_id, grade (A/B/C/U), hit rate, mean forward return, sample size. Updated daily from rolling Spearman-rank evaluations across ~200 assets × 5+ years. Use to select which signals to wire into a live trading strategy.

Aggregated taker buy/sell volume per asset with buy-sell ratio. Reads aggressor pressure independently of price — flow can lead price by hours. Use for high-frequency entry timing and tape-reading at the asset level.

Full state snapshot for LLM grounding — factor breadth (bullish/bearish), top/bottom composite rankings, signal trigger summary, pipeline run status, asof day. Single parallel-read call returning everything an agent needs before reasoning. Use BEFORE any market analysis call.

Daily AI-readable market brief — top-ranked assets by composite score, total signal triggers across the universe, anchor day, structured narrative summary. Built for LLM context windows. Use as a single-call grounding before any market reasoning task.

Per-asset triggered signal feed on the latest anchor day from api.v_asset_signals_enriched. Returns ok, symbol, asof_day, macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence`, total, high_grade (count of A/B), signals[], meta. Each signal row: signal_id, name, description, class (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), horizon, value, z, confidence, ic, hit_rate, ir, regime_accuracy, predictive_grade, operational_grade, health_grade, is_anti_predictive, plus `walk_forward` block per signal at its registered horizon (is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade) from analytics.signal_eval_wf_day. Anti-predictive signals filtered out, ordered by health_grade then |z|. IC/grades are cross-sectional (universe-wide) — flagged in meta. Per-asset live firing panel with regime gating + walk-forward robustness in one call.

Cross-class positioning panels per asset — futures funding stress, options 1M risk-reversal skew, order-book imbalance, and 7-day taker buy/sell ratio. Each panel: current z, percentile, 30-day sparkline, one-line narrative.

Nearest-k peers in CRYPTYX's 8-class factor t-score space (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL) for a target asset at the requested horizon (7d/14d/30d/90d). Point-in-time Euclidean distance over per-class t-scores from metrics.factor_scores on the latest available day — not a cluster label, not rolling correlation. Returns target {asset, composite, ret_7d, classes} plus peers ranked by distance with composite, divergence vs target, ret_7d, and shared_classes. Use for hedge selection, basket construction, and divergence trades grounded in the protected factor engine.

Per-asset macro regime context: returns current regime id, confidence, days_in_regime streak from signals.asset_regime_day, plus conditional forward-return stats (n, mean_return, median_return, hit_rate) at 1d/7d/14d/30d horizons from analytics.forward_returns, and top-12 historical regime transition pairs. Computed live across ~200 assets — grounds sizing and strategy selection in what historically tends to happen next when this asset sits in this regime.

Per-asset triggered signal ledger over last N days (7-365, default 90), filtered to grade A/B non-anti-predictive signals only. Returns ok, symbol, days, events[] with asof_day, signal_id, name, class (one of 8: CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), direction (long/short/neutral derived from registry tags + z sign), z, confidence, horizon, grade, close_px for chart overlay, fwd_return_7d realized outcome, plus meta.ic_context=cross_sectional. Use for "when did X last fire on SOL and what happened next" backtests across ~200 assets and 143 atomic signals.

Claude Haiku-narrated per-asset conviction call. Returns `thesis` (3-5 sentence prose), macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence` at the top level, plus an `evidence` pack: `composite` (score, rank, rank_pct, score_yesterday, score_delta) vs ~200 assets, `returns` (ret_1d/7d/30d, vol_rv_30d), `factors` (t_score_pct + t_band per CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL class at chosen horizon), `active_signals` (top 8 health A/B/C, anti-predictive filtered, with z/confidence/ic/hit_rate/grade), `active_signals_total`/`_high_grade` counts, and `top_ts_metrics` (5 strongest per-asset predictors w/ ts vs cs grade_divergence). Plus `ok`, `symbol`, `asof_day`, `horizon`, `source`, `model`, `generated_at`, `cache_hit`. 4h cache.

Per-day composite conviction trajectory for one asset across a 7-90 day window. Reads metrics.composite for the score timeline and LEFT JOINs analytics.asset_thesis_daily for optional one-liner overlay (direction + one_liner + generated_at). Returns ok, symbol, days, points[] with {d, composite, direction, one_liner, generated_at}, thesis_count, and a note flagging when the thesis populate job has not yet written rows. Score timeline is always complete; narrative overlay is honest about gaps. Use to track conviction evolution before a major move.

Per-asset predictor explorer for one symbol at 7d/14d/30d horizon. Returns up to 5 top atomic signals AND 5 top metrics ranked by |IC| times hit_rate, gated to A/B health grades, anti-predictive filtered, family-deduped, max 2 per factor class. Signals carry ic, hit_rate, ir, grade, n_triggered_180d, last_triggered, plus `walk_forward` block at the requested horizon (is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade). Metrics carry ts_grade vs cs_grade divergence, polarity, per-asset time-series ic, current_z, current_pctl (walk-forward currently signal-only — metrics blocks omit it). Use to pick the strongest, walk-forward-validated inputs for one asset before building a strategy.

Per-asset historical signal edge. For one symbol at 7d/14d/30d, returns up to 6 health-grade A/B atomic signals that actually worked on THIS asset, grouped into reversal/continuation/breakout lanes (max 2 per lane). Each setup: signal_id, name, class, grade, tags, n_triggers, hit_rate, mean_fwd_return, median_fwd_return, last_triggered, ic_global, hit_rate_global, score, plus `walk_forward` block per setup at the requested horizon (is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade) from analytics.signal_eval_wf_day — the per-asset trigger evidence and universe-wide walk-forward robustness side by side. Anti-predictive signals filtered (Phase N11 gate); MIN_TRIGGERS=3. Scope: signals_only.

Latest-day Deribit IV term structure for one asset plus current put/call open-interest ratio. Returns term_structure (per-tenor rows: tenor, atm_iv, rr_25d 25-delta risk reversal, bf_25d 25-delta butterfly, iv_25c, iv_25p), pcr block (pcr, oi_put, oi_call), and asof_day. Coverage limited to Deribit venues (BTC, ETH) — out-of-scope assets return empty arrays. Raw upstream telemetry feeding the OPT factor class; use for skew reads, vol-term structuring, and options positioning context.

Top-N and bottom-N assets by CRYPTYX IC-weighted composite score for the latest asof_day. Each row carries asset, comp_score, comp_rank, ret_1d, ret_7d, funding_z (FUT_FUNDING_1D), oi_roc_z (FUT_OI_ROC_30D), n_atomic and n_composite signal counts, optional sharpe_7d, optional thematic_scores keyed by composite_id, plus macro `regime` (expansion/consolidation/deleveraging) + `regime_confidence`. Stablecoins filtered. One call returns top, bottom, asof_day, durationMs — the daily "where to look" leaderboard across ~200 assets with regime gating inline.

Factor term-structure grid from metrics.factor_scores: 7 classes (CORR/EFF/FLOW/FUT/OB/TR/VOL, OPT excluded) x 8 horizons (1D/7D/14D/30D/60D/90D/180D/365D) of market-average t_score with n_assets, positive_count, negative_count breadth and 1-day momentum_delta. mode=asset returns the same grid for one symbol plus market_avg and delta_vs_market per cell; optional sparkline history for a single class x horizon cell. Use to read factor breadth and 1d rotation across CRYPTYX's ~200-asset universe.

Single-metric z-score backtest over 5+ yr fact history. POST {metric_id, asset, operator (gt/lt/gte/lte/abs_gt), threshold, cooldown_days?}. Returns ok, echoed params, full timeseries [{asof_day, close_px, z, pctl, triggered}], triggers [{asof_day, z, pctl, fwd_1d/7d/14d/30d/60d/90d/180d/365d, max_dd_*}] (1d-30d from analytics.forward_returns, 60d-365d on-the-fly), and summary {n_observations, n_triggers, n_triggers_raw, n_suppressed, trigger_rate, horizons.{h}.{n, mean_return, median_return, max_return, min_return, std_return, hit_rate, sharpe, max_dd_mean, max_dd_p95, max_dd_worst}} over 8 horizons. Cooldown suppresses repeat triggers. Pre-composite single-factor edge testing across 440+ metric catalog, ~200 assets.

Multi-factor z-score intersection backtest for a single asset. POST 2-4 conditions (metric_id + operator gt/lt/gte/lte/abs_gt + threshold z) plus optional cooldown_days; route intersects triggered days across all conditions from metrics.fact_stats (365d window) and returns metricSeries (per-metric daily z+pctl+triggered), priceSeries, compositeTriggeredDays, triggers[] with per-horizon fwd_return + max_dd at 1d/7d/14d/30d/60d/90d/180d/365d, and summary.horizons stats (n, mean, median, hit_rate, sharpe, drawdown p95/worst) plus perMetric trigger counts. Pre-computed forward returns for 1d-30d, log-return on the fly for 60d-365d. Use to validate confluence stacks against ~5y of CRYPTYX 440-metric history before live deployment.

Cross-sectional z-score scanner: pick any of 440+ metrics across 8 classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), apply gt/lt/gte/lte/abs_gt vs threshold on the 365d-window z, get ranked triggering assets from the full ~200-asset universe on the latest asof_day. Each row carries z, pctl, cross_section_pctl, price, volume_usd, ret_1d/7d/30d, comp_score, comp_pct, ts_grade, ts_ic. Response also returns metric_name/class/polarity, total_assets, triggering_count and a health block (ic_7d, hit_rate_7d, n_obs_7d, sample_sufficient, is_anti_predictive, data_state, grade). Live anomaly hunting for agents — "who is at 3-sigma on TR_MOMO_CONT_14D right now?"

Backtest any of 143 atomic or 9 composite signals over up to 365 days. Returns totalDays, triggeredDays, triggerRate, totalTriggerEvents, avgConfidence, per-day rows[], plus a performance block: 1d/7d/14d/30d horizons each with sample_size, mean_return, median_return, hit_rate, bootstrap_ci_95, statistical_significance, is_reliable, capped by reliability_grade (green/yellow/red) and explicit caveats. The user-bounded backtest above is in-sample only; alongside it the top-level `walk_forward` block returns the latest nightly out-of-sample read at 7d/14d/30d (is_ic, oos_ic, overfit_ratio, ic_delta, reliability_grade) from analytics.signal_eval_wf_day so the IS conclusions can be cross-checked against the OOS evidence in the same response. dryrun skips signal_log writes.

Class-level attribution waterfall for one asset and horizon — master score, per-class contributions ordered by impact, per-signal triggers with walk-forward IS/OOS, and per-signal health grades.

Walk-forward backtest of an agent-supplied 8-class composite weight vector (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL, 0-2 each) over [from, to] at 7d/14d/30d horizon. Loads class composite scores from signals.class_composite_log and forward returns from analytics.forward_returns, then runs three blends in parallel — custom, IC-weighted default, equal-weight — applying ±threshold long/short rules. Returns performance.{custom,default,equal_weight} with total_days, avg_daily_return, hit_rate, sharpe, max_drawdown, cumulative_return, avg_assets_triggered, plus a daily[] array with n_bullish, n_bearish, n_total, avg_score, top_asset, top_score. Strategy-design primitive for autonomous agents benchmarking custom factor blends against CRYPTYX's production IC-weighted moat. Institutional+ tier.

Universe-wide master composite breadth at a chosen horizon (7d/14d/30d) — bullish/bearish/neutral counts and percentages across ~200 assets, sliced by primary asset regime, with intraday micro_regime overlay and a full per-asset stance list (asset, score, rank, stance, regime, regime_confidence). Threshold-tunable (0-1). Reads signals.master_composite_log + asset_regime_day. Use to read participation: is this rally broad or concentrated, and which regimes are driving it.

Bundles four arrays for the latest N days (max 30): master composite scores per (asset, horizon) with rank, n_classes, meta; latest-day class composites per (asset, class) with n_signals, n_triggered, meta; latest-day macro regime context (primary/secondary regime_id, confidence, resolution_mode); current intraday micro_regime state per asset and regime_type. Drives multi-horizon factor heatmaps. Scores are raw, 4 decimals — not 0-1 normalized.

Pipeline primitive that re-runs the database-side signal evaluator for a single YYYY-MM-DD day. Pass signal_id+dryrun=true → returns wouldTrigger row count from signals.signal_dryrun_count (no writes). Pass signal_id alone → executes signal_eval_day, returns insertedRows written to signal_log. Omit signal_id → executes signal_eval_all_for_day across enabled signals, returns results[] of {signal_id, inserted_rows} plus durationMs. Row counts only — no per-asset, IC, regime, or grade. Use as smoke-test or to recompute a missed eval day.

Forks a registered signal with a new params JSONB into an INACTIVE version row in signals.signal_params — production firing is untouched. Verifies signal_id exists in signals.signal_registry, then inserts COALESCE(MAX(version),0)+1 with is_active=false. Returns ok, signal_id, version, params, is_active, created_at. Experimentation primitive for the agent optimization loop: fork, then call /api/signals/backtest dryrun, then activate. Covers all 143 atomic signals across CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL classes.

Pre-flight threshold simulator for the signal registry. POST {signal_id, threshold_key, threshold_value, asof_day?} — looks up the signal's underlying metric (z-score from metrics.fact_stats window=365, or t-score from metrics.factor_scores for TR signals), counts how many of the ~200 assets would trigger at the proposed cutoff for one asof_day, returns {ok, signal_id, threshold_key, proposed_threshold, current_threshold, asof_day, n_triggered, n_total, estimated_trigger_rate, estimated_trigger_pct}. Single-day scalar only — no IC delta, no per-asset breakdown, no sweep. Use before forking params via /api/signals/fork.

All currently-triggered signals for a single asset with confidence, grade, and trigger reason. Filtered to one name. Use as the per-asset live signal feed.

Return asymmetry stats per asset — skew, tail ratio, expected shortfall (5%). Identifies fat-tailed names. Use for tail-aware position sizing and risk-parity overlays.

Peer cluster for a single asset — cluster label and similarity-ranked peers with correlation. Computed from rolling 30-day factor correlations. Use for pairs trading, basket construction, and hedge selection.

Current macro regime label and confidence for a single asset. Single-call regime check. Use before sizing — strategies that work in expansion regimes often fail in consolidation.

Time series of signal firings on a single asset over the last N days — day, signal_id, triggered, confidence. Use for signal-level backtesting on a specific name and "when did X last fire" lookups.

Complete per-asset thesis — composite score, rank, dominant factor class with t-score, top firing signals with grades, regime, key metrics. The canonical "should I trade this asset" call. Replaces 10+ separate API calls.

Daily history of an asset's thesis — composite score, regime, signal count, dominant factor class by day. Use to track conviction evolution over time and study how an asset behaved before a major move.

Top metrics for a single asset ranked by IC and hit rate. Asset-specific — the strongest predictors for SOL aren't the same as for BTC. Use to find the most reliable inputs for an asset before designing a strategy.

Top-grade signals firing on a single asset right now with hit rate and mean forward return per signal. Use as the entry point for asset-level deep dives — "why should I look at SOL today?" answered in one call.

Deribit implied-volatility surface for an asset — expiry × strike grid with IV, delta, and gamma. Use for options structuring, skew reads, and risk-reversal pricing.

Highest- and lowest-conviction assets by composite score, with the count of atomic signals contributing per name. Pre-ranked, top and bottom of the universe. Use for filtered watchlists, short-list generation, and daily "where to look" briefings.

Asset × factor-class t-score matrix with rank per class — full universe of ~200 assets. Use for factor-tilt construction, sector rotation, and "show me the strongest momentum assets right now" queries.

Single-metric z-score backtest — hit rate, mean forward return, per-asset breakdown across a configurable z-threshold. Backed by 5+ years of fact history (440+ metrics available). Use to test a single metric's standalone edge before incorporating it into a composite.

Multi-factor z-score intersection backtest — 2-4 metric conditions chained with AND logic. Hit rate, mean forward return, sample count. Use to test confluence strategies ("when TR is bullish AND VOL is compressing AND OB depth is rising") before live deployment.

Scan a single metric across all ~200 assets for current z-score extremes — ranked, with percentile context. Use to find live anomalies and outliers — "who is at 3+ sigma on TR_TREND_STRENGTH_14D right now?"

Backtest a single atomic or composite signal over a custom date range — per-day events plus aggregate trigger rate, total events, average confidence. Backed by 5+ years of indexed factor history (~200 assets). Optional dryrun mode (no writes). Use to validate any signal before deploying it in a live strategy.

Per-factor-class contribution and weight to an asset's composite score at a given horizon — decomposes the composite into its drivers. Use to explain WHY an asset is ranked where it is — essential for client-facing narratives and model audits.

Backtest a custom-weighted composite across the 8 factor classes — per-day composite return series with Sharpe, Sortino, max drawdown. Backed by 5+ years of factor history. Use to design and validate multi-factor strategies before promoting them to live composites.

Bullish / bearish breadth per factor class at a chosen horizon — total counts, per-class splits, net positioning. Use to read market participation: is this rally broad, or concentrated in a few names?

Composite intensity heatmap — asset × factor-class intensity grid (0-1 normalized) over the last N days. Use to spot factor concentration visually — when one class is hot across many assets, regime is forming.

Evaluate one or all signals for a given day — triggered count, total assets considered, per-asset breakdown. Optional dryrun (no signal_log writes). Use as a smoke test for the eval pipeline or to recompute a missed eval day.

Fork a signal with new parameters into an inactive parameter version — does not disrupt production firing. Versioned through signals.signal_params with full audit trail. Use as the experimentation primitive for agent-driven signal optimization.

Estimate a proposed threshold change's effect on a signal — projected trigger rate, sample triggers, per-asset distribution. Runs against 5+ years of fact history without writing. Use to size parameter changes before committing them to the registry.

Top-N composite-ranked assets from the ~200-asset universe, bundled as a single agent starter pack. Each idea returns asset, rank, composite_score, returns{ret_1d,ret_7d}, sharpe_7d, derivatives{funding_z,oi_momentum_z}, regime{regime,confidence}, dominant_factor{class,t_score} across the 8 factor classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL), signal_count{atomic,composite}, top_setups (top-3 atomic signals gated by health_grade A/B, not anti-predictive, n_triggers>=5, with hit_rate + mean_fwd_return + last_triggered), and top_predictors (top-3 TS-graded metrics per asset). Replaces 5 separate calls.

Top-ranked trade ideas across the universe — each idea bundles composite score, regime context, top backtested signal setups with hit rate, and derivatives overlay (IV, funding) in a single call. Pre-ranked by conviction. The canonical "where should my agent look right now?" call for autonomous trading systems.

Re-rank the ~200-asset CRYPTYX universe under your own class weights across the 8 factor classes (CORR/EFF/FLOW/FUT/OB/OPT/TR/VOL) at 7d/14d/30d horizon. Reads signals.class_composite_log + the active versioned default weight set, returns asof_day, horizon, total_assets, custom_weights, default_weights, summary {bullish, bearish, neutral}, ranked assets [{asset, score, default_score, delta, class_contributions, rank}], and top_movers (10 largest |delta| vs default). Encode a house view, A/B test a composite recipe, or stress-test the production weights against your own.

Natural-language intelligence query over the full CRYPTYX state — returns answer plus structured sources cited (factor scores, signal triggers, regime context). Use when the question doesn't fit a fixed schema, or when piping an LLM agent into CRYPTYX without per-endpoint plumbing.

Re-rank the entire ~200-asset universe by a custom class-weight composite — pass your own weights across the 8 factor classes. Use to encode an institutional house view as a ranking system or A/B test alternative composite recipes.