エージェントを強化
公開元自身のコマンドです。PlugMCP のエンドポイントではありません。
キーとログ付きでコピーをホスト。
npx -y [email protected]Claude Code にはローカルコマンド用のインストールリンクがありません。下のコマンドをコピーして、ターミナルで実行してください。
claude mcp add canli-validation-mcp -- npx -y canli-validation-mcp@0.16.0ターミナルで実行します。
{ "mcpServers": { "canli-validation-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
リモート URL か、手元のマシンのコマンドか 2 つのインストール方法の違い
PlugMCP で実行
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
起動公開元より
Open quant research where every published number comes with the file that produced it, including the strategies that failed. Live at canlicapital.com.
If a check here stopped you from trusting a backtest that wasn't real, a ⭐ helps other quants find it.
Check whether a backtest is real: deflated Sharpe, CSCV probability of backtest overfitting, minimum track record and backtest length, haircut Sharpe and luck-equivalent trials, as an MCP server. Local mode, hosted endpoint, signed receipts.
# Ask your AI assistant whether a backtest is real (Claude Code shown; Cursor, VS Code and
# Claude Desktop setups are in mcp/README.md)
claude mcp add canli -- npx -y canli-validation-mcp
# Or call the free API: get a key, then deflate a Sharpe ratio for the number of trials you ran
curl -X POST https://canlicapital.com/api/v1/keys -H "Content-Type: application/json" -d '{"label":"readme"}'
# Or check the published record yourself (Python 3)
pip install cryptography
curl -sO https://canlicapital.com/glassbox/reproduce.py
for f in capacity_commitment founder_commitment; do curl -sO https://canlicapital.com/glassbox/$f.json; done
python3 reproduce.py --dir .
Backtest validators (deflated Sharpe, CSCV overfitting probability, minimum track record, haircut Sharpe, luck-equivalent trials) as an MCP server and a free API, each result with a signed receipt anyone can recompute.
SEC fundamentals point in time as an MCP server
(npx -y canli-fundamentals-mcp): what a company first reported, what was known on any date, and
every later restatement, each value with its filing, computed locally from a hash-checked SEC
snapshot.
The research record as an MCP server (npx -y canli-research-mcp):
papers, killed candidates, trial counts and the live paper record.
Company reference pages built from SEC XBRL filings: financial histories where every value links to the filing it came from, with the year-on-year change and growth rate worked out.
Research, including what failed: the kill log, the Null Zoo (overfitting corrections scored where the truth is known) and luck-equivalent trials.
Open data: FilingFacts, financial questions with machine-checked answers from SEC filings (CC BY 4.0).
ALPHAC, the three-strategy paper-trading book whose record is published daily; its engine is github.com/arhancanli/alphac.
Help check it: open review tasks take 15 to 60 minutes and are credited on /review. To build with it, see CONTRIBUTING.md. If the project is useful to you, a star helps other people find it.
Created and maintained by Arhan Canli for Canli Capital.
Machine-readable software citation metadata is provided in CITATION.cff.
Current work and verified limitations: persistent goal status.
The site's whole claim is "a quant fund proving itself in public before it asks you to trust
it." — the tagline in config/brand.js. A site that makes that claim and hides its own source
is asking for a trust it hasn't earned. So: this is the source, including the parts that enforce
honesty on us.
The load-bearing one is docs/retracted_claims.txt in the engine repo. When a number is
withdrawn it goes on a blocklist, and check_retracted_claims.py scans dist/ and public/ for
it before each deploy. It cannot be satisfied by deleting the number — a retracted figure must
still be quotable inside its own retraction, so a match counts only when the explanation is
absent from the surrounding window.
That check exists because the retraction here had already failed twice: a withdrawn DSR of 0.83 stayed on the homepage and in the social-unfurl card for six days after the signed chain formally withdrew it. The pipeline was publishing the correction and the error in the same run.
The gate is now fail-closed for publication. Since 2026-08-19, both live_tick.sh and
live_publish.sh run check_retracted_claims.py after regeneration and skip the deploy when it
fails. Trading remains outside that blast radius: a publication defect can stop the website from
shipping, but cannot place, cancel, or delay an order.
public/paper-state.json and public/glassbox/* are written by the engine's publish job, not by
hand. They are the machine-readable form of every claim the pages make. Current corpus counts are
derived during the build from public/research-index.json,
public/glassbox/trial_packet_manifest.json, and the generated measurement directory; the sitemap
is generated from the same files rather than maintained separately. The present build contains
372 canonical URLs in the sitemap (all indexable), plus a public noindex evidence page for every incomplete registered
trial and one archival HTML paper per registered sleeve. It publishes
identity-level packets for all 228 recorded hypotheses, while
honestly marking 226 of those packets incomplete.
The local paper-journal example connects supplied sizing, order checks and signed local journal evidence. The execution package remains private and Unreleased, with journal writes explicitly enabled by the caller. It has no broker connection and places no orders.
npm install
npm run build # Vite multi-page build -> dist/
npm run preview # serve the built dist
npm run dev # dev server with hot reload
Three.js, GSAP + ScrollTrigger and Lenis are self-hosted — they install from npm and Vite
fingerprints them into dist. Nothing is fetched from a CDN at runtime. The build must stay
green, and source plus dist must contain zero em dashes (U+2014); both are audited before deploy.
config/brand.js is the one place names and numbers live. js/shell.js renders the nav and footer
from it so every page ships byte-identical chrome, and js/main.js binds data-brand,
data-flagship, data-tagline and data-fact nodes from it.
STATS and FACTS are the only numeric claims permitted on the hand-authored marketing
surfaces. Generated papers and measurement pages obtain their figures from engine exports and
carry their own source paths and claim boundaries. audit-published-numbers.mjs reconciles the
shared site-level figures; do not add a number to either layer without binding it to an
authoritative artifact.
Nothing on this site or in this repo is investment advice, an offer, or a solicitation. The record
published here is paper trading; the published ALPHAC strategy record includes no funded performance. Simulated and past
performance do not indicate future results. See LICENSE: provided "as is", without warranty.
The platform direction and quality contract
connect the research engine, developer API/MCP and the company-reference collection. The local
expansion candidate adds 49 SEC-backed reference pages; it is not a million-page deployment.
The search goal is at least 800,000 indexed pages, targeting 1,000,000, recorded in
config/search-growth-goal.json. npm run seo:inventory reports built counts separately from
actual indexing evidence; the indexed count remains unverified until Search Console evidence
is available. --require-indexed-minimum fails while that evidence is missing.
npm run seo:capacity tests sitemap transport with synthetic URLs in a temporary directory.
Refresh selected company records explicitly with npm run companies:import -- CIK [CIK ...].
Builds use captured public JSON and original compressed source snapshots, with no network ingestion
during publication. The selected latest-filed histories may contain restatements and are not
point-in-time backtest data. Public JSON downloads do not extend the validation API's capabilities.
Check whether a backtest is real: deflated Sharpe, CSCV probability of backtest overfitting, minimum track record and backtest length, haircut Sharpe and luck-equivalent trials, as an MCP server. Loca
{ "mcpServers": { "canli-validation-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}概要
Check whether a backtest is real: deflated Sharpe, CSCV probability of backtest overfitting, minimum track record and backtest length, haircut Sharpe and luck-equivalent trials, as an MCP server. Local mode, hosted endpoint, signed receipts.
ツール
ツール一覧はまだありません。リモートサーバーは、最初のヘルスチェック後に取得されます。
よくある質問
このページで Cursor、Claude Desktop、Claude Code、VS Code 用の stdio 設定をコピーできます。PlugMCP のプロキシではなく、公開元のインストール方法を使います。
このレコードに必須の環境変数は記載されていません。
Canli Validation は stdio 経由でローカル実行されます。
最終チェックは 2026-11-15 です。 Check whether a backtest is real: deflated Sharpe, CSCV probability of backtest overfitting, minimum track record and backtest length, haircut Sharpe and luck-equivalent trials, as an MCP server. Local mode, hosted endpoint, signed receipts.