Trading and Data Engineering
StrikeZone Crypto
A private trading-systems R&D workspace for indicator, market-intelligence and signal-delivery research.
Outcome
A research framework for reproducible experiments across indicator development, market intelligence and governed delivery workflows.
Media
Select any frame to enlarge
The Market Structure Analyzer working on BTC 1h. Every BOS, CHoCH and order block you can see also emits a JSON alert payload — the drawing and the Discord message come from the same event. The same suite on SOL. Captured by the agent driving a scoped research browser over the DevTools protocol, not screenshotted by hand. The same detection dropped to 15m, which is where the entries actually get taken. Denser structure, same rules. Where a chart event becomes a message. Each structure event calls buildAlertJson() and fires once per bar close — the line that turns a drawing into a webhook. The delivery side. One channel per signal per timeframe, so a member follows only the feeds they trade. The market thesis desk, which is the system's flagship output. Every thesis carries the evidence timestamp it was built from — here 2026-07-24T11:55:43Z — then per-asset price, 24h range position, funding and open interest, then a posture with the exact levels that would upgrade or invalidate it. The chart composite above the text is assembled in the same run. The same channel refusing. When the scheduled refresh cannot validate a complete evidence set, it publishes a hold notice instead of a thesis — no forced signal and no invented levels. This is the more important screenshot of the two: a system that only ever shows you its good days is not telling you what it does. One asset, read in full. Consolidation and bearish paths are stated as conditions rather than predictions, and the level response plan says in advance what each outcome would mean — written before price resolves, so it cannot be rewritten afterwards. Levels and their invalidations kept in their own channel, so the price that would prove a read wrong is as visible as the read itself. Session opens. Crypto trades continuously but does not behave uniformly, so the session boundary is published as context rather than left implied. Bias flip alerts on their own feed. Separating them means a member can subscribe to a change of direction without subscribing to every structure event. Liquidity sweeps. Same routing rule as every other feed — one signal type per channel, so nothing arrives mixed. The daily close, which is what the next morning's thesis is measured against.
Case study
The market thesis desk
The flagship output is a full market read, not an alert. Each thesis opens with the evidence timestamp it was built from, then gives per-asset price, position within the 24h range, funding and open interest for BTC, ETH and SOL, then a posture stated with the exact levels that would upgrade or invalidate it — a reclaim above one price, a loss of another. The chart composite is assembled in the same run, so the picture and the text describe the same moment rather than being paired after the fact. When the scheduled refresh cannot validate a complete evidence set, it publishes a hold notice instead: no forced signal, no invented levels, and an explicit statement that the previous read still stands until validation passes.
Two engines, one server
The member-facing half is a suite of proprietary TradingView indicators — market structure, scored fair-value gaps, liquidity sweeps, EMA crossovers, session killzones and an alt risk-regime index. Each one builds a JSON alert payload and fires it at bar close into its own Discord webhook, so a member subscribes to exactly the signals they trade rather than one firehose. The second half is a research runtime that reads evidence, drafts market analysis, and grades itself.
A browser the system drives itself
Research capture runs through an isolated Chromium profile controlled over the DevTools protocol. It is scoped to StrikeZone research sources only: the accounts used for community posting are bound to separate profiles and explicitly excluded from capture, so a research automation cannot reach a public account even by mistake. The chart images in this gallery were captured through that lane rather than taken by hand.
Sources are declared, not scraped
Every source the system may read is a registry entry with an operational status and a selector contract naming the fields a capture must produce — task identity, completion timestamp, evidence cutoff, freshness, canonical URL, body quality. A capture missing a contract field is not a weak source; it is not a source. Adding one means registering an adapter and letting it run in shadow until it earns usable status; retiring one is a config change, and past captures keep their provenance so old analysis stays auditable.
It reviews itself weekly, and the review has teeth
Each week the runtime audits its own output — false reads, missed moves, gate failures, which sources actually changed an analysis versus which added noise — and writes a patch plan. Patches become upgrade proposals carrying before-and-after configuration hashes and a set of governance flags. If a proposal would mutate policy, methodology, credentials or member-facing behaviour, it stops and waits for a human. Failed verification after apply rolls the change back.
Failure is recorded, not hidden
Every scheduled job stores its last status, last error and last delivery error. A bounded-retry watchdog re-attempts missing output within a cap, and a repair pass fixes publication failures inside a defined policy and escalates whatever it may not touch. A refusal is deliberately distinguished from a crash: when a gate declines to publish because the evidence is stale or incomplete, that is the system working, and reporting it as a failure would train the operator to ignore the error channel.
Absence detection, because silence is the hardest failure
The hardest thing to notice is a job that stopped rather than one that broke — a stopped job raises nothing. A drift detector holds a committed baseline of expected jobs and alarms when one disappears, and a freshness checker alarms when an artifact stops growing even though its job still reports success. Both exist because a green status is structurally blind to work that quietly ceased.
What it deliberately cannot do
There is no wallet, no exchange key and no order placement anywhere in the system — not disabled, absent. Analysis is evaluated on paper against public candles, and publishing to member channels sits behind an allowlist with an explicit operator approval phrase. The boundary is structural rather than a setting someone could flip.
Verified architecture
System map
Technology
- Pine Script
- Python
- Discord infrastructure
- ChaseOS workflow pack
Inside this system
TradeSync
ResearchPaper-first market-intelligence and agent-systems research with explicit evidence, advisory and execution boundaries.
Repository ↗Pine Script indicators
ResearchStrategy and indicator development for the StrikeZone Crypto system.
Repository ↗StrikeZone Market Intelligence
ResearchA ChaseOS workflow pack running governed daily market intelligence: evidence acquisition, thesis drafting and publication with fetch-back proof. No trade execution.
Repository not public
Scope and boundaries
No live trade execution, no financial performance guarantees, and no production signal subscription service.