Carving finance_lab out of Sqetch
Short project-hygiene note. In December 2025 the QCBM + TimeGAN + regime-detection work that had been living inside the Sqetch fintech platform got moved into its own repo, finance_lab. The QCBM + TimeGAN post covers the technical content; this one is just about why the split.
Why separate
- Dependency weight. The QCBM work pulls in PennyLane, PyTorch, scikit-learn, hmmlearn, Optuna. The rest of Sqetch doesn’t. Keeping them in the same repo meant every Sqetch build paid for an import of the whole quantum stack.
- Release cadence. The research code iterates faster than the platform code. Separate repos let each release on its own cadence without the research commits cluttering the platform’s diff.
- Reviewability. A PR touching “the research notebook plus the production config plus the TEE attestation flow” is unreviewable. Three PRs across three repos are each reviewable.
- Open-source optionality.
finance_labcan plausibly go public once the claims are honest (see the QCBM post’s “what I can say without overclaiming” section). The platform code can’t. Keeping them together would have held the research hostage to the platform’s IP posture.
What didn’t change
- The APIs each side exposes to the other. The carve-out produced a cleaner import boundary, not a new protocol.
- The data. Both sides still read from the same DuckDB store;
finance_labconsumes, doesn’t ingest. - The overall architectural story. Nothing was rewritten; files moved.
What did change
- Tests that depended on cross-repo fixtures had to be either inlined into each side or lifted into a small shared
sqetch-test-datapackage. The latter is cleaner; the former is what actually happened on day one. - CI per repo is faster individually. The combined pipeline used to be the bottleneck; two shorter pipelines are a better day-to-day experience.
The wider point
Splitting a repo is a boring, unglamorous operation. It’s easy to defer and easy to regret. In this case the split landed late enough that a few tests broke during the move; it should have happened a quarter earlier.
Private (finance_lab), part of the broader Anzaetek codebase.
Written on December 18, 2025