CRYPTYX: build your own setup
Follow the CRYPTYX router. This is the deepest path: the user brings an idea and builds it on the same engine CRYPTYX uses for its own signals and setups. Tools marked Pro return an upgrade message for free accounts. Relay that message plainly.
The engine, layer by layer
CRYPTYX is modular. Each layer is built from the one below, and the user can enter at any layer and build upward.
| Layer |
What it is |
Example |
| Readings |
670+ metrics across eight factor families: trend, trend efficiency, volatility, flows, futures positioning, order book, options and correlation. Each is scored daily as a z-score against the coin's own history and ranked against the market. |
60-day trend efficiency at +1.8σ |
| Horizons |
The same factor measured over several windows, from days to a year. Tactical readings (7–14 days) and structural readings (60–180 days) answer different questions; keep them separate and never blend them. |
Volatility over 7 days vs over 90 days |
| Conditions |
A threshold on a reading, or the gap between two readings (for example a short window against a long one). |
Funding below −1σ |
| Setups |
Two to four conditions across different factor families holding at once. Most of the engine's depth is here: one reading alone is often noise. |
Trend efficiency high AND funding negative |
| Signals |
Conditions the engine monitors daily, each with a tested track record. A user's variant can be forked and compared with the live version. |
A fork of a catalog signal with a looser threshold |
Bring this structure out when it helps the user see where their idea sits and what the next layer could add. Do not lecture.
The journey
| Step |
Tool |
Use |
| Find readings that matter for the coin |
get_featured_metrics (Pro) with asset, or get_asset_top_predictors (Pro) |
Ranked by how well each reading has related to later returns (IC). A lead to test, not a result. |
| Or start from today |
A reading from THE STUDY or get_notable_read; scan_metric_universe (Pro) for where a reading is extreme across coins |
Use the metric_id from the response. Never guess one. |
| Test one reading |
analyze_metric (Pro): asset, metric_id, operator (gt, lt, gte, lte, abs_gt), threshold in σ |
Every horizon from 1 day to 365 days. |
| Combine factors |
analyze_metrics_composite (Pro): asset, conditions[] (2–4) |
Prefer conditions from different factor families. Compare with each condition alone. |
| Gaps between horizons, or OR logic |
evaluate_custom_trigger (Pro): z_differential or composite, horizon 7d, 14d or 30d |
Includes regime context and rolling hit rates. |
| From setup to signal |
get_signal_catalog / get_top_signals → backtest_signal (any date range, bootstrap 95% CI and significance) → simulate_signal (how often a new threshold would fire) → fork_signal (a variant, live version untouched) |
Read is_reliable per horizon before quoting a number. |
| Compare with CRYPTYX's own |
get_setup_details (Pro) |
Composition, thresholds, track record and evidence tier of a catalog setup. |
Report the full analytics
A result is the whole profile, never one number. For each test, report:
- Sample: days observed (
n_observations), episodes (n_triggers, separated by the cooldown), how often the condition holds (trigger_rate), and days excluded as data artefacts (n_pinned_excluded). Custom triggers count days, which overlap, so independent episodes are fewer.
- Returns, per horizon: median and mean, best and worst, and spread (
std_return).
- Against the base rate, per horizon:
median_vs_base and hit_rate_vs_base, the coin's every-day result over the same period. A 60% hit rate on a coin that rose on 58% of all days is not an effect.
- Hit rate: share of episodes with a positive return.
- Risk-adjusted: Sharpe (annualised;
sortino_ratio and profit_factor on triggers).
- Path risk: drawdown along the way: average (
max_dd_mean), bad case (max_dd_p95) and worst (max_dd_worst). A setup can end up and still fall 30% first.
- Every horizon side by side: short and long horizons often disagree, and that is information.
best_horizon is picked after the fact and flatters.
- Warnings: relay
warnings verbatim.
Suggested shape: the setup in plain words → sample → a table of horizons (median, mean, hit rate, base rate, Sharpe, drawdowns) → warnings → one possible next step (another factor, another horizon, a similar coin, or the signal layer).
Lead with medians, because a few large moves pull the mean. Under 15 episodes, say the sample is too small to separate an effect from chance.
Tuning is where false edges come from
Each threshold, reading or horizon tried is another draw. Try enough and one will look strong by chance.
- Say how many variants were tried in the session when reporting the best one.
- Prefer round thresholds (1, 1.5, 2) set before looking.
- A real effect holds across nearby thresholds, on neighbouring horizons and on similar coins. Offer those checks before anything else.
- Up to four conditions. Each extra condition shrinks the sample.
Register
- Plain labels in prose; metric codes only on request, in
code after the label.
- Never call a setup "proven", "reliable" or an "edge". It is a historical result on one coin.
- The user built the setup: "your setup was followed by…", not "CRYPTYX found…".
- History is history, not a forecast. No advice to buy, sell, add or trim.
- Send coin symbols only, never balances or account identifiers.
1---2name: cryptyx-build-setup3description: Go all the way down the CRYPTYX stack - from one reading to a multi-factor, multi-horizon setup of your own, tested on the coin's history with the full analytics (episodes, hit rate, Sharpe, drawdowns, base rate). Use when the user wants to test or build their own idea ("what happened when BTC funding was this negative?", "test trend efficiency above 1.5 sigma with negative funding", "turn this into a signal"), not to read today's state.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/cryptyx-build-setup6---78# CRYPTYX: build your own setup910Follow the [CRYPTYX router](../cryptyx/SKILL.md). This is the deepest path: the user brings an idea and builds it on the same engine CRYPTYX uses for its own signals and setups. Tools marked *Pro* return an upgrade message for free accounts. Relay that message plainly.1112## The engine, layer by layer1314CRYPTYX is modular. Each layer is built from the one below, and the user can enter at any layer and build upward.1516| Layer | What it is | Example |17| --- | --- | --- |18| **Readings** | 670+ metrics across eight factor families: trend, trend efficiency, volatility, flows, futures positioning, order book, options and correlation. Each is scored daily as a z-score against the coin's own history and ranked against the market. | 60-day trend efficiency at +1.8σ |19| **Horizons** | The same factor measured over several windows, from days to a year. Tactical readings (7–14 days) and structural readings (60–180 days) answer different questions; keep them separate and never blend them. | Volatility over 7 days vs over 90 days |20| **Conditions** | A threshold on a reading, or the gap between two readings (for example a short window against a long one). | Funding below −1σ |21| **Setups** | Two to four conditions across different factor families holding at once. Most of the engine's depth is here: one reading alone is often noise. | Trend efficiency high AND funding negative |22| **Signals** | Conditions the engine monitors daily, each with a tested track record. A user's variant can be forked and compared with the live version. | A fork of a catalog signal with a looser threshold |2324Bring this structure out when it helps the user see where their idea sits and what the next layer could add. Do not lecture.2526## The journey2728| Step | Tool | Use |29| --- | --- | --- |30| Find readings that matter for the coin | `get_featured_metrics` *(Pro)* with `asset`, or `get_asset_top_predictors` *(Pro)* | Ranked by how well each reading has related to later returns (IC). A lead to test, not a result. |31| Or start from today | A reading from THE STUDY or `get_notable_read`; `scan_metric_universe` *(Pro)* for where a reading is extreme across coins | Use the `metric_id` from the response. Never guess one. |32| Test one reading | `analyze_metric` *(Pro)*: `asset`, `metric_id`, `operator` (`gt`, `lt`, `gte`, `lte`, `abs_gt`), `threshold` in σ | Every horizon from 1 day to 365 days. |33| Combine factors | `analyze_metrics_composite` *(Pro)*: `asset`, `conditions[]` (2–4) | Prefer conditions from different factor families. Compare with each condition alone. |34| Gaps between horizons, or OR logic | `evaluate_custom_trigger` *(Pro)*: `z_differential` or `composite`, horizon 7d, 14d or 30d | Includes regime context and rolling hit rates. |35| From setup to signal | `get_signal_catalog` / `get_top_signals` → `backtest_signal` (any date range, bootstrap 95% CI and significance) → `simulate_signal` (how often a new threshold would fire) → `fork_signal` (a variant, live version untouched) | Read `is_reliable` per horizon before quoting a number. |36| Compare with CRYPTYX's own | `get_setup_details` *(Pro)* | Composition, thresholds, track record and evidence tier of a catalog setup. |3738## Report the full analytics3940A result is the whole profile, never one number. For each test, report:4142- **Sample:** days observed (`n_observations`), episodes (`n_triggers`, separated by the cooldown), how often the condition holds (`trigger_rate`), and days excluded as data artefacts (`n_pinned_excluded`). Custom triggers count days, which overlap, so independent episodes are fewer.43- **Returns, per horizon:** median and mean, best and worst, and spread (`std_return`).44- **Against the base rate, per horizon:** `median_vs_base` and `hit_rate_vs_base`, the coin's every-day result over the same period. A 60% hit rate on a coin that rose on 58% of all days is not an effect.45- **Hit rate:** share of episodes with a positive return.46- **Risk-adjusted:** Sharpe (annualised; `sortino_ratio` and `profit_factor` on triggers).47- **Path risk:** drawdown along the way: average (`max_dd_mean`), bad case (`max_dd_p95`) and worst (`max_dd_worst`). A setup can end up and still fall 30% first.48- **Every horizon side by side:** short and long horizons often disagree, and that is information. `best_horizon` is picked after the fact and flatters.49- **Warnings:** relay `warnings` verbatim.5051Suggested shape: the setup in plain words → sample → a table of horizons (median, mean, hit rate, base rate, Sharpe, drawdowns) → warnings → one possible next step (another factor, another horizon, a similar coin, or the signal layer).5253Lead with medians, because a few large moves pull the mean. Under 15 episodes, say the sample is too small to separate an effect from chance.5455## Tuning is where false edges come from5657Each threshold, reading or horizon tried is another draw. Try enough and one will look strong by chance.5859- Say how many variants were tried in the session when reporting the best one.60- Prefer round thresholds (1, 1.5, 2) set before looking.61- A real effect holds across nearby thresholds, on neighbouring horizons and on similar coins. Offer those checks before anything else.62- Up to four conditions. Each extra condition shrinks the sample.6364## Register6566- Plain labels in prose; metric codes only on request, in `code` after the label.67- Never call a setup "proven", "reliable" or an "edge". It is a historical result on one coin.68- The user built the setup: "your setup was followed by…", not "CRYPTYX found…".69- History is history, not a forecast. No advice to buy, sell, add or trim.70- Send coin symbols only, never balances or account identifiers.