Version 0.1.0, beta. MIT licensed. $1.6T platform · $4.5B a day · 36-month program · 12+ weeks in production parallel.
Each asks one question and leaves you holding a document. Run them in order, or jump to the one you need.
Scored red, amber or green, never averaged into one number. Green on ten and red on two is one bad weekend from being a case study.
Day one of the flagship platform, before any work began. My judgment of a program I led, not output from the tool.
A date exists and it predates the scope. What this program has is the two things that cannot be bought later: one named person accountable for correctness, and a book that can be split by account and fund group. Either the date moves or the bar does.
Every red above was cleared: over 50% of the platform migrated, zero loss of trading accuracy. Archaeology with every rule ratified by a human, a parity harness reconciling every record, 12+ weeks in production parallel.
How it was done →Both came out of running the work rather than writing about it. Both replace a metric or a practice that is standard and quietly misleading.
Both give you the same thing. Copying lets you read the instructions first, which is reasonable for something about to review your architecture.
Corrections are worth more than additions: if a criterion here is wrong, say so on GitHub.