Thinking in public, carefully.

Most retail traders watch price. Dealers watch their books. When market makers carry large positive gamma they sell rallies and buy dips to stay hedged. That dampens volatility. When gamma flips negative their hedging does the opposite and moves get amplified.

Gamma exposure, or GEX, is an attempt to estimate that aggregate dealer position from open interest. It is an approximation with real assumptions baked in. You have to guess who is long and who is short each strike. Still, tracking the sign and the concentration of exposure around key strikes tells you a lot about how a session is likely to behave.

On my platform I compute GEX across roughly one hundred tickers throughout the trading day. The most useful signal is rarely the level itself. It is the change. When exposure migrates from one strike cluster to another the character of intraday price action often shifts with it.

Delta and gamma get all the attention. But two second-order Greeks quietly drive a lot of hedging flow. Vanna measures how delta changes when implied volatility moves. Charm measures how delta decays as time passes.

Why care? Because dealer hedges respond to both. A drop in implied volatility changes dealer deltas through vanna and forces rebalancing even when spot has not moved. Charm does the same thing on a clock. It is one reason flows cluster around expiration dates.

Computing these across a full universe of tickers taught me a practical lesson too. Analytic Black-Scholes derivatives are cheap to evaluate but the data pipeline around them is the hard part. Getting clean chains, consistent implied vols and a sane treatment of dividends matters more than the calculus.

Selling strangles looks simple on paper. You collect premium from both sides and manage the position back to delta neutral as the market drifts. The hard part is deciding which strangles are worth selling and when to walk away.

I ended up building efficiency ratios from the Greeks themselves. Theta over gamma tells you how much decay you earn per unit of hedging risk. Vega over theta tells you how exposed that income is to a volatility spike. Ranking candidates by these ratios in real time turned a discretionary process into a systematic one.

The biggest lesson had nothing to do with math. An automated strategy is only as good as its failure modes. Timeouts, stale data and half-filled orders will find you. Design for them from day one.

Real-time options analytics sounds like it needs expensive infrastructure. It does not, if you separate what truly needs to be fast from what does not. My platform runs two loops. A slow loop refreshes full option chains on a schedule. A fast loop re-computes gamma and dealer exposure intraday from cached chains and fresh spot prices.

The stack is deliberately boring. Flask serves the dashboard. APScheduler drives the loops. SQLite in WAL mode handles concurrent reads while writes stream in. A thread pool fans out per-ticker computation and Cython accelerates the pricing engine where profiling said it mattered.

The result runs all trading day on a single small EC2 instance. Boring technology, chosen carefully, is a feature.

Markets and merchandise have more in common than people think. Both punish overconfidence. A demand forecast is a position and inventory is the cost of being wrong. Buying deep into a seasonal SKU is not far from selling a naked option. The upside is capped and the downside shows up in the clearance bin.

Trading taught me to think in distributions rather than point estimates. In planning that translates to buying to a confidence band, protecting the downside on volatile styles and letting proven sellers run. It also taught me to respect the base rate. Last season's sell-through is a better prior than this season's enthusiasm.

The tools differ. The discipline is the same. Size your bets, measure your errors and never fall in love with your forecast.

Most dashboards die quietly. They get built, demoed once and never opened again. The ones that survive share a trait. They answer a question someone already asks every week, faster than the old way of answering it.

My rule when building executive reporting is to start from the meeting, not the data model. What number gets asked for first? What comparison settles the argument? Build that view first and make it load fast. Depth can come later as drill-downs for the few who want it.

The second rule is to automate the refresh completely. A dashboard that needs manual care becomes stale the first busy week. Stale once, distrusted forever.