Skip to content

Crypto Research Labs v0.3.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 21 Oct 09:32
v0.3.0
5df8479
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Plain-language dashboard, every rule checked on every coin, weekly digest and alerts in 7 languages,
a free LLM option, and a redesign that works on phones.

Added

  • Coin pages on the dashboard (coins/<SYMBOL>.html, 7 languages): price, changes, trend, volatility,
    12-month chart and today's state of each tested rule; a Coins table on the home page.
  • Translated titles for H-012…H-019.
  • Coin pages: distance from the highest close, value today of $100 invested a year ago, 90-day correlation with Bitcoin.
  • Exchange flows (Coin Metrics community, BTC since 2011, ETH since 2015) stored in datasets/exchange_flows.csv and refreshed daily; BTC net-flow regime card; H-020/H-021 (heavy exchange deposits).
  • Coin pages show today's state of every data-driven rule tested on the coin (e.g. H-020 on BTC); an "H-020 rule today" regime card alerts when the rule switches.
  • H-022: forward replication of H-020 on as-published flow data (rows older than 14 days are never rewritten), first test 2027-10-15.
  • Gram (GRAM, prev. Toncoin) from Gate.io: coin page, market breadth (now 13 coins), H-023 (SMA50 trend rule, same as H-018 for EVER).
  • Market diagnosis in plain words on the home page (7 languages) and a diagnosis on every coin page.
  • Coverage scan (crl scan, weekly): every tested rule on every coin with FDR control; matrix on the home page, per-rule tables on hypothesis pages, per-coin tables on coin pages, reports/COVERAGE.md.
  • Sortable tables (coins, hypotheses, scan); forward-data progress bar for hypotheses waiting for new data (H-022).
  • Cross-section studies (kind: cross_section, daily ranking vs an equal-weight basket, rotation null). H-024…H-026 from a ChatGPT proposal: all fail.
  • Weekly digest page (digest.html, 7 languages) and the latest digest on the home page: the model writes each week's digest in every dashboard language; files in reports/digest/<week>/.
  • Free LLM option: any OpenAI-compatible provider via LLM_API_KEY (+ CRL_LLM_BASE_URL), Groq by default.
  • Setup page on the dashboard (setup.html, 7 languages): Telegram alerts, alert language, the optional AI model key (what it does, how to get it), running your own copy, all settings in one table.
  • Telegram alerts in 7 languages: repository variable TELEGRAM_LANG; alerts end with a link to the dashboard in that language.
  • Telegram test: Run workflow with test_alert, or crl notify-test.
  • New look: IBM Plex Sans, deep-navy palette with high-contrast text, brand header, diagnosis panel with a signal meter, chip matrix; on phones every table turns into labelled cards and the matrix into one chip grid per rule (no horizontal scrolling).
  • replicates field: a declared replication is not flagged as a duplicate.
  • test_after field: a hypothesis is not run (even with --force) before its pre-registered start date.
  • Market breadth monitor: share of the 13 coins above their 200-day average (regime card + alerts).
  • Course for beginners on the website (course.html), 10 lessons in 7 languages, live numbers, Course structured data.

Fixed

  • research-agents stopped parsing after duplicated env keys; a test now loads every workflow with a strict YAML loader.
  • Weekly Telegram message without a model is the plain-language diagnosis, not a dump of raw reports.
  • LLM model is discovered from the provider's /models list (Groq retired the hard-coded one); the digest is written
    once in English and translated, retries rate limits and never publishes a cut-off answer.
  • Diagnosis says "uptrend"/"downtrend" only with a clear majority of signals.

Live dashboard: https://f1devbin.github.io/Crypto-Research-Labs/

Assets 5
Loading