I'm a beginner building a personal Windows app with Codex and Claude. This post was AI-written from project material (Fully Autonomous disclosure). I need experienced human judgment about the next step.
The goal: leave capture running, open today's SPX and same-day standard PM-settled SPXW analysis, switch 1/5/15/30/60-minute views, inspect history, save exactly what I saw and reopen it unchanged. Intended analysis includes Gamma/GEX, gamma flip, walls, max pain and source-qualified options flow. ES futures volume is supporting context, not SPX's own traded volume. The app does not trade; ordinary viewing should not require selecting recordings or model parameters.
Stack: C#/.NET 10, SQLite, Protobuf, Interactive Brokers and Moomoo OpenD. The intended chain is:
Callbacks → complete recorded batches → one reducer → transaction saving state and restart progress → committed progress → fixed-input calculations → stored results → screen → save/reopen.
Prices, SPX/ES bars, open interest, a source-gamma/OI proxy, trades and limited save/reopen checks have working pieces. Ordinary Gamma/Flip is missing from continuous service/UI; a separate research calculation does not complete that workflow. Own-IV calculation is planned. Quote qualification, carry assumptions and asynchronous snapshots remain unresolved. The gamma/OI proxy is not observed dealer inventory; these limitations must remain visible in saved history.
Evidence is limited: an older candidate reached roughly 300,000 records in 88.93 seconds, then hit a shared 120-second limit before roughly 803,000 records. Subsequent deletion-key/ownership fixes have small functional evidence, including a 45-record synthetic example, but no new large-sample benchmark. AI source review also flagged potentially inconsistent 60-minute bar boundaries, ambiguous lost-save replies with server-generated IDs, and redraws tied to combination IDs despite separate qualification changes. These are unconfirmed static concerns, not established failure causes.
What is the smallest useful workflow you would finish next, and what acceptance example would prove it from incoming data through reopening the saved observation? Please consider interruption, stale inputs, restart and an uncertain save reply, including upstream/downstream effects of any proposed change.
Snapshot cutoff: 26 October 2026, 04:13 UTC. Source copies, notes and test evidence are not a runnable reproduction; raw recordings, SDK binaries and a current trace are absent.
Top comments (0)