Capabilities

A running index of what the AI Edge collaborator can actually do — price action, backtesting, live data, ML, chart engineering, finance, and repo ops. Use it as a central place to keep track and to ask “what should we add?”

Price action & Al Brooks

Working knowledge of the Brooks corpus and how it maps onto live bar-by-bar action on AI Edge charts.

  • Brooks vocabulary

    H1/H2/L1/L2, signal vs entry bars, breakout / breakout-pullback / failed-breakout, ii/iii, wedges, double tops/bottoms, final flags, micro double bottoms, magnets, measured moves, climaxes, channels, trading ranges, BTC vs MTR, FTC.

  • Trend / range diagnosis

    Always-in long vs short, trend-from-the-open vs trading-range-day, tight vs broad channel, gap context, and how those frames change which setups are takeable.

  • Probability framing

    Talking in 40/60 terms, premium vs discount entries, when to take a setup vs when to wait one more bar, when the trade is "always-in but late."

  • Brooks-tour DTW matching

    Hybrid DTW (5-channel skeleton + 10-channel Brooks features) against the Brooks book corpus with vertical-flip search and the most-relevant passages pulled per match.

  • Wisdom reference

    Hallmarks, guidelines, and principles quoted verbatim from the Brooks book corpus — each snippet tagged by kind and cited to its source book and figure.

Backtesting & verification

Pulling the actual historical record before trusting an intuition. Trade-by-trade, R-multiple-based, no curve fits.

  • Setup-by-setup ledgers

    5-min TFO, gap-up + FT, BGU, opening-setup labelers, and Brooks-style patterns each get their own ledger with fill time, exit time, stop, +2R target, realized R, MFE, and EOD R.

  • Bar-accurate fills

    Stops and targets resolved against 1-min bars from the live-bars store, so the R number is what would actually have hit — not a daily approximation.

  • Cohort + regime slicing

    Splitting trade lists by ticker, direction, opening setup label, entry family, gap %, ATR regime, day-of-week, and synced_at vintage to find where edge concentrates.

  • History + Analogs

    Pull the closest historical analogs for a current setup, score them, and use them as a sanity check before the bar closes.

Live data & infra

The Fly aggregator → /api/live-bars → /api/bars → chart pipeline. The skill is keeping it boring and dedup-clean.

  • Aggregator pipeline

    Fly-hosted live-bars aggregator (Dockerfile.live-bars, fly.live-bars.toml) with deploy via GitHub Actions on push to main.

  • Subscribe / reconnect discipline

    Dedup keys, no reconnect-loops, single-source bar truth — the invariants that the aiedge-live-data skill enforces every time the feed is touched.

  • Operator diagnostics

    The operator endpoint and Vercel production logs (/logs) used to confirm a feed is healthy before declaring a deploy live.

  • Supabase + Vercel wiring

    Server-only service role, .env.local.example as the single source of truth for env vars, no client-side leakage.

ML & scoring

Lightweight, explainable models that rate setups the moment the bar closes — not black boxes.

  • Setup probability ratings

    Every scanner fire is scored at bar close so the surfaced edge is calibrated, not just "it printed."

  • Feature engineering for bars

    Brooks-aware features (channel slope, pullback depth, prior-leg measured move, vol ratio, length ratio) feeding both the scoring models and the DTW similarity search.

  • Repro environment

    scripts/requirements-ml.txt pinned, models reproducible from raw bars + the labeler outputs; nothing depends on a one-off notebook.

Chart engineering

Everything under /chart — indicators, overlays, the ƒx menu, watchlist, live-status badge, settings.

  • Indicator + overlay system

    EMA / HTF overlays, SR strip, Always-in toggle, ƒx menu — wired through chart conventions captured in the aiedge-chart skill.

  • Mobile watchlist

    Behavior tuned for the iPhone width so the 9+ nav routes and the watchlist coexist without cramming.

  • Smoke testing

    npm run test:chart for the browser smoke test, plus curl -I on the local /chart route before publishing.

Finance & market structure

Enough domain background to talk through trades without translating jargon every sentence.

  • Intraday US equities

    RTH session structure, opening drive vs opening reversal, lunch chop, last-hour reversion, MOC dynamics, gap classification (gap-up FT, gap-fade, micro-gap).

  • Risk units in R

    Per-trade R, MFE, eod_r, +2R targets, stop discipline — the language the journal and review tooling already speak.

  • Catalyst & calendar context

    Earnings, FOMC, CPI/NFP, holiday-shortened sessions and how they should change which setups are takeable.

Repo & deploy ops

The boring part that keeps shipping safe: Go Live workflow, boundary hook, rollback.

  • Go Live workflow

    Quality check → concise proof → push origin HEAD:main → poll Vercel Production until Ready → verify the live route. No "live" claim before the URL responds.

  • Product boundary hook

    .claude/hooks/check-boundary.sh scans every push for the categories of content that don’t belong on aiedge.trade and blocks the deploy when it finds them. Surfaces, never bypasses.

  • Smoke / Logs / Rollback

    /smoke for pre-flight, /logs for Vercel production tail + summarize, /rollback for one-shot revert of the latest production deploy.

  • Ideas triage

    Brain-dump → triage → autonomous execution via /ideas when you drop multiple things at once.

Living document. Ask “what should we add to capabilities?” any time and the list gets a proposed diff.