Goal — A decision instrument for day trading, not a piece of analysis writing: a surface where the picture across crypto assets, equity futures and the macroeconomic setting comes together, and which points out movements while they happen. The benefit arrives twice — in the moment, because the picture is in one place rather than 10 windows, and afterwards, because you can find out whether your own decisions were any good: the record was written before the outcome was known.

Challenge — The hard part is not the gathering. Sources age at different speeds and fail quietly: one that goes silent while still showing its last value is more dangerous than one that visibly crashes. On top of that comes hindsight, which makes every decision look more reasonable afterwards than it was. The only defence is a record written in the moment and untouchable after it.

Current state — Runs in the studio’s own daily operation, with several independent sources per instrument: the picture across crypto assets, equity futures and the macroeconomic setting comes together on one surface. The path outward is built — the notices go to the operator as a message with a rendered chart. He is the only recipient; there is nothing to subscribe to and no access from outside. Setups are written as rows before the outcome is known and evaluated later against fixed horizons — which makes it decidable afterwards whether a decision was any good, rather than remembered. Rows found to be faulty are refused rather than corrected after the fact. Every source carries an age and a staleness gate; on its first production run that gate immediately found real cases — silent sources that would otherwise have kept showing their last value. What is open is carried on a list rather than in a footnote. live

What part AI plays — In the judgement none: the evaluation is rules and statistics, the staleness gates are thresholds. A model at this point would cost exactly the traceability the whole thing is built for. In the building, throughout: collectors, indicators, evaluation and checks are developed by AI agents, in separated workspaces, with human approval at every step — including open defects, which were found that way.