From 63cecf8526ee9a4a4f0587f435b762e7d1184532 Mon Sep 17 00:00:00 2001 From: augiemazza <139812003+augiemazza@users.noreply.github.com> Date: Sun, 8 Mar 2026 19:04:51 -0500 Subject: [PATCH 1/3] Add VARRD to Available Skills table --- README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/README.md b/README.md index 5e0e644b..b35d9f27 100644 --- a/README.md +++ b/README.md @@ -25,6 +25,7 @@ Bankr Skills equip builders with plug-and-play tools to build more powerful agen | yoink | [yoink](yoink/) | Social on-chain game. "Yoink" a token from the current holder. Uses Bankr for transaction execution. | | [Neynar](https://neynar.com) | [neynar](neynar/) | Full Farcaster API integration. Post casts, like, recast, follow users, search content, and manage Farcaster identities. | | [Hydrex](https://hydrex.fi) | [hydrex](hydrex/) | Liquidity pools on Base. Lock HYDX for voting power, vote on pool strategies, deposit single-sided liquidity into auto-managed vaults, and claim oHYDX rewards. | +| [VARRD](https://varrd.com) | [varrd](varrd/) | Statistical edge discovery engine. Turn plain-English trading ideas into validated setups with event studies, backtesting, and Bonferroni-corrected K-tracking. `pip install varrd` β€” CLI, Python SDK, or MCP server. | ## Adding a Skill From c2a11c5ec942532c4391e4023e18050bbda0cae5 Mon Sep 17 00:00:00 2001 From: augiemazza <139812003+augiemazza@users.noreply.github.com> Date: Sun, 8 Mar 2026 19:04:51 -0500 Subject: [PATCH 2/3] =?UTF-8?q?Add=20varrd/SKILL.md=20=E2=80=94=20statisti?= =?UTF-8?q?cal=20edge=20discovery=20skill?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- varrd/SKILL.md | 44 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+) create mode 100644 varrd/SKILL.md diff --git a/varrd/SKILL.md b/varrd/SKILL.md new file mode 100644 index 00000000..e6f24b2d --- /dev/null +++ b/varrd/SKILL.md @@ -0,0 +1,44 @@ +--- +name: varrd +description: Statistical edge discovery engine β€” test any trading idea with real market data. Use when the user wants to validate a trading hypothesis, find edges, scan live signals, or generate trade setups. +version: 1.0.0 +tools: ["Bash"] +homepage: https://varrd.com +metadata: {"openclaw": {"requires": {"bins": ["varrd"]}, "emoji": "πŸ“Š"}} +--- + +# VARRD β€” Statistical Edge Discovery + +Turn any trading idea into a statistically validated trade setup. + +## Install + + + +## What It Does + +VARRD is a quant-in-a-box. You bring a trading idea in plain English, VARRD loads real market data, charts the pattern, runs rigorous statistical tests, and tells you definitively whether there’s an edge β€” with exact entry, stop-loss, and take-profit prices. + +## Commands + + + +## Statistical Guardrails + +Every test enforces institutional-grade controls automatically: + +- **K-tracking** β€” significance bar rises with each test +- **Bonferroni correction** β€” multiple comparison penalty +- **OOS lock** β€” out-of-sample is one-shot, no re-running +- **Fingerprint dedup** β€” can’t retest same formula/market/horizon +- **Lookahead detection** β€” catches formulas using future data + +## Also Works As + +- **Python SDK:** +- **MCP server:** (Claude Desktop, Cursor, OpenBB) +- **Trading bot integrations:** and generate ready-to-run strategy files + +## Cost + + free on signup. ~/usr/bin/bash.25 per idea tested. Scan/search always free. From bcab2b1753d6ecdfe621f156c9cf10b21b4d9778 Mon Sep 17 00:00:00 2001 From: augie mazza Date: Mon, 9 Mar 2026 10:08:38 -0500 Subject: [PATCH 3/3] Update SKILL.md with complete guardrails, install, commands, and bot integrations --- varrd/SKILL.md | 117 +++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 95 insertions(+), 22 deletions(-) diff --git a/varrd/SKILL.md b/varrd/SKILL.md index e6f24b2d..ca0c6661 100644 --- a/varrd/SKILL.md +++ b/varrd/SKILL.md @@ -1,7 +1,7 @@ --- name: varrd -description: Statistical edge discovery engine β€” test any trading idea with real market data. Use when the user wants to validate a trading hypothesis, find edges, scan live signals, or generate trade setups. -version: 1.0.0 +description: AI-powered trading research β€” validate any idea with real statistical testing. Use when the user wants to test a trading hypothesis, find edges, scan live signals, or generate trade setups. +version: 0.3.1 tools: ["Bash"] homepage: https://varrd.com metadata: {"openclaw": {"requires": {"bins": ["varrd"]}, "emoji": "πŸ“Š"}} @@ -9,36 +9,109 @@ metadata: {"openclaw": {"requires": {"bins": ["varrd"]}, "emoji": "πŸ“Š"}} # VARRD β€” Statistical Edge Discovery -Turn any trading idea into a statistically validated trade setup. - -## Install - +Turn any trading idea into a statistically validated edge β€” or find out it doesn't work β€” in about 3 minutes. +## What it does -## What It Does +Ask a question in plain English: -VARRD is a quant-in-a-box. You bring a trading idea in plain English, VARRD loads real market data, charts the pattern, runs rigorous statistical tests, and tells you definitively whether there’s an edge β€” with exact entry, stop-loss, and take-profit prices. +- "Does buying BTC after a 5% weekly drop actually work?" +- "Is there a seasonal pattern in crude oil before OPEC meetings?" +- "When VIX spikes above 30, is there a bounce in ES futures?" -## Commands +VARRD loads real market data, runs event studies with Bonferroni correction, tests across multiple markets, validates out-of-sample, and returns a proven trade setup with entries, exits, stops, and targets. Or tells you there's no edge β€” which is just as valuable. +## Autonomous edge discovery +Point it at any market and say "find me what works." It scans, tests, discards noise, and surfaces only what survives real statistical rigor. Then monitors your edges live and alerts you when setups are firing with fresh levels. -## Statistical Guardrails +## Coverage -Every test enforces institutional-grade controls automatically: +- **Futures (CME):** ES, NQ, CL, GC, SI, ZW, ZC, ZS + 20 more +- **Stocks/ETFs:** Any US equity (15,000+ instruments) +- **Crypto (Binance):** BTC, ETH, SOL + more -- **K-tracking** β€” significance bar rises with each test -- **Bonferroni correction** β€” multiple comparison penalty -- **OOS lock** β€” out-of-sample is one-shot, no re-running -- **Fingerprint dedup** β€” can’t retest same formula/market/horizon -- **Lookahead detection** β€” catches formulas using future data +## Install -## Also Works As +```bash +pip install varrd +``` -- **Python SDK:** -- **MCP server:** (Claude Desktop, Cursor, OpenBB) -- **Trading bot integrations:** and generate ready-to-run strategy files +The agent connects to VARRD's MCP server at `https://app.varrd.com/mcp`. $2 free credits on signup β€” enough for 6-8 research sessions. -## Cost +## Commands - free on signup. ~/usr/bin/bash.25 per idea tested. Scan/search always free. +```bash +varrd research "When crude oil drops 5% in a week, what happens next?" +varrd discover "Find edges in gold futures" +varrd scan # Scan all your validated edges against live data +varrd search "momentum" # Search saved strategies +varrd hypothesis # Load a specific strategy +varrd balance # Check remaining credits +``` + +Or use the Python SDK: + +```python +from varrd import VARRD + +v = VARRD() +result = v.research("Does buying SPY after a 3-day losing streak work?") +print(result.context.edge_verdict) # "STRONG EDGE" or "NO EDGE" +``` + +## Bot integrations + +VARRD generates ready-to-deploy strategy files for trading bots: + +```python +from varrd.freqtrade import generate_strategy +from varrd.jesse import generate_strategy as jesse_strategy +``` + +Validate first, deploy second. Works with Freqtrade, Jesse, Hummingbot, OctoBot, and NautilusTrader. + +## Guardrails (all infrastructure-enforced) + +Every test runs through 18 categories of automated guardrails. You can't turn them off. + +### Statistical rigor +| Guardrail | What it prevents | +|-----------|-----------------| +| K-Tracking | Multiple comparison abuse | +| Bonferroni Correction | P-hacking | +| OOS Lock | Re-running out-of-sample after seeing results | +| Fingerprint Dedup | Retesting the same thing twice | +| Lookahead Detection | Using future data in formulas | +| Similarity Detection | Testing near-duplicate hypotheses | + +### Returns & risk +| Guardrail | What it prevents | +|-----------|-----------------| +| ATR-Normalized Returns | Cross-market comparisons on raw returns | +| Beats-Market Validation | Claiming edge when it doesn't beat buy-and-hold | +| Edge Decay Tracking | Trading edges that no longer hold on recent data | +| Max Drawdown Tracking | Ignoring worst-case equity decline | +| Sharpe & Sortino | Ignoring risk-adjusted performance | + +### Execution & portfolio +| Guardrail | What it prevents | +|-----------|-----------------| +| SL/TP Optimization | Unoptimized or overfit stop-loss and take-profit levels | +| Slippage Modeling | Fantasy fills that ignore real market impact | +| Volatility-Based Sizing | Position sizes that ignore current volatility | +| Gap Handling | Unrealistic fill assumptions on price gaps | + +### Workflow enforcement +| Guardrail | What it prevents | +|-----------|-----------------| +| State Machine | Skipping steps in the research process | +| Sandbox Guardrail | AI doing its own math instead of using tested tools | +| Tools Calculate, AI Interprets | Fabricated statistics or edge claims | + +## Links + +- **Web app:** [app.varrd.com](https://app.varrd.com) +- **GitHub:** [github.com/augiemazza/varrd](https://github.com/augiemazza/varrd) +- **PyPI:** [pypi.org/project/varrd](https://pypi.org/project/varrd/) +- **MCP endpoint:** `https://app.varrd.com/mcp`