Skip to content
LAB · BUILD LOG

The Lab

The most complex systems I run. Data-intense research and analysis platforms on institutional-grade pipelines: scheduled ingestion from multiple providers and a self-hosted node, deduplication and reconciliation across sources, and AI models embedded in the apps doing the analysis. Private, built for my own use.

CHAPTERThe Node: what it is

THE NODE · ON-CHAIN INTELLIGENCE

LOG node-001, WHAT IT IS

A private Bitcoin cycle-intelligence platform. Price, on-chain and sentiment data flow into one deterministic compute layer; a language model writes the narrative on top of the numbers, never the numbers themselves. House rule: the model narrates, the code computes.

LOG node-002, THE PIPELINE

A self-hosted Bitcoin Core full node, exposed only through a Cloudflare Tunnel. Scheduled collectors pull on-chain valuation metrics (MVRV Z-Score, NUPL, SOPR, Puell Multiple), market and sentiment feeds, and an exchange-premium series sampled every 30 minutes and stored permanently. Supabase holds state and scheduling; serverless functions run the jobs; the AI layer calls Anthropic's Claude API directly, with no gateways or routers in between.

LOG node-003, THE LOGIC

Every intelligence brief starts from computed inputs: trend windows, cluster scores, scenario evidence, and trigger gates with hard numeric thresholds. The narrative layer can only describe what the computation already established. That is the liability firewall, and the reason the output can be trusted at 06:00 without re-checking its math.

node · on-chain intelligence
The Node · Real data
Node console showing block height, peer count, sync status, network hashrate and mempool size
Metric cards showing price, MVRV Z-Score, MVRV ratio and net unrealised profit and loss with valuation gauges
Metric cards showing spent output profit ratio, a sentiment index, the Puell multiple and node sync status
"The model narrates, the code computes."

THE DESK · EQUITY SCENARIO ENGINE

LOG desk-001, WHAT IT IS

A daily decision-support system for stock trading. Six dependency-ordered stages run server-side every morning: market-regime classification, a deterministic scan, a filter stage enriched with TipRanks analyst and catalyst intelligence, thesis building, brief composition, and a midday re-check.

LOG desk-002, THE PIPELINE

Daily price bars from a market-data provider, analyst ratings and catalyst data from TipRanks, and execution against an Interactive Brokers paper account as full bracket orders. A single server-side orchestrator drives all stages in dependency order with semantic completion checks; a sequencing race caught during validation is exactly why those checks exist.

LOG desk-003, THE DISCIPLINE

Nothing trades real money. A validation gate requires a minimum number of closed trades and weeks of live paper history before real capital is even discussed, and a benchmark panel tracks performance against the index, net of costs. Paper trading exists to find where the system breaks, not to confirm the thesis.

desk · scenario engine
Daily brief showing the current market regime, position ceilings and paper validation day count
A closed paper trade showing the thesis, the invalidation condition and the time stop

Want this pattern in your company?