An institutional-grade signal engine built outside the bank. Cold, precise, and falsifiable — operated as if every trade has to defend itself to a partner desk.
Pinnacle Dominion began as a pre-earnings signal engine inside a larger LLM-tooling ecosystem and was spun out into its own product when the framework hardened. The goal was straightforward: build a desk-grade analytical pipeline that anyone running a real book could submit to with confidence — beat-rate calibration, options sentiment, dark-pool flow, and macro regime context, fused into a single conviction score with falsifiable formulas and explicit caps.
Every trade competes for scarce capital. Conviction below 6 is watchlist, not capital deployment. No portfolio slot exists by default.
The system runs a seven-stage framework on every event: executive verdict → earnings edge → market positioning → technical structure → trade construction → risk review → final desk call. Every binary print is graded after the fact against both tracks — hold-through and pop-sell — regardless of which one we ran. That's the only way to honestly calibrate the Fade-Risk referee.
The journal is append-only. Every price fetch, regime change, IV shift, dark-pool print, and exit is logged with timestamp and citation. When the journal and the daily report conflict, the journal wins. When the trigger prompt and the journal conflict, the journal still wins. There is one source of truth, and it carries the audit trail.
Pinnacle Dominion runs on a vendor stack centered on FMP Premium and Unusual Whales Basic — $219/month total — with a Yahoo point-in-time collector replacing Zacks at zero cost. The composite engine, written in Python, executes against a SQLite registry of ~99 large-cap tickers, with a Flask dashboard fronted by a Cloudflare tunnel at signals.pinnacledominion.com.
Compute is self-hosted: Qwen3-32B-FP8 on RunPod for prompt routing and composite scoring. Webhooks fire on fills and regime changes via Discord and Telegram. The complete pipeline — from EPS revision detection to dual-track conviction publish — runs unattended on a 6-hour cadence, with a manual dual-track review at every print T-1 / T-2.
Builds and operates the PDS pipeline end-to-end. Authored the composite engine, the Fade-Risk referee, and the post-mortem calibration loop.
The daily desk discipline — pre-print pop-sell triggers, stop-trail decisions, override logging. Every override is journaled for calibration honesty.
We don't claim to predict earnings. We claim to calibrate the distribution of outcomes around earnings — including the failure modes (premium-multiple BEAT_AND_GUIDE_FADE, gamma-positive vol-suppressive tape, restrictive-regime long-duration drag) — and to convert that calibration into an explicit, falsifiable conviction score. The wins are real. The sample is still small. We're transparent about both.
Reach out about the methodology, the calibration record, or any of the published reports. Press inquiries, partnership conversations, and technical questions all welcome.