Vale construir sua versão de Grow or Die?

Build me a single-tenant AI contribution-profit dashboard for one product, replacing the narrow personal core of Grow or Die.
Use Node.js 22, TypeScript, Fastify, better-sqlite3, and server-rendered HTML with no frontend framework.
Run as one process behind Caddy; store everything in one SQLite file with a documented backup command.
Ship a first-party browser tracker that records pageviews and product events with a random anonymous visitor ID.
Add POST /identify so my app can bind that visitor ID to my own stable account_id; never use email as the join key.
Verify Stripe webhooks for checkout.session.completed, invoice.paid, charge.refunded, and subscription deletion.
Read account_id from Stripe metadata and store each payment or refund idempotently in an append-only revenue ledger.
Provide a thin wrapper around the official OpenAI Node SDK that returns the provider response unchanged.
The wrapper records account_id, model, token usage, cached input tokens, latency, status, and occurred_at.
Never collect prompts, generated output, or the OpenAI API key; the key stays in .env and goes only to OpenAI.
Keep model prices in a versioned JSON catalog with effective dates and calculate cost on the server.
Join product events, revenue, and AI cost only by account_id and only across the same finalized 7, 30, or 90 day window.
Never turn a missing payment, unknown model, partial import, or unmatched identity into zero. Show it as unknown.
Dashboard: visitors, paying accounts, revenue, AI cost, contribution profit, margin, and a customer profit table.
Every total links to the underlying ledger rows and shows source freshness, unmatched counts, and unpriced calls.
Recommend one next action only when complete observed data supports it; otherwise recommend the missing connection or identity fix.
Protect the dashboard with one admin bearer token from .env; add no accounts, billing, telemetry, cookies, or third-party analytics.
Include clearly labelled demo data that can be deleted in one command and never appears after real data arrives.
Write unit tests for webhook idempotency, identity joins, price effective dates, refunds, and unknown-state propagation.
Add one end-to-end test that tracks a visitor, identifies an account, records a payment and model call, and shows profit.
README: setup, tracker and identify snippets, Stripe CLI testing, SDK wrapper usage, deployment, backup, and limitations.
Explicitly exclude GA4, PostHog, Lemon Squeezy, Anthropic, multi-currency, cross-device identity, teams, and automated model-price discovery.
Run tests and a production build before finishing, and fix every failure.