0001 — Provenance-first project store
- Status: accepted
- Date: 2026-06-24
- Deciders: bioedca
- PRD anchor: §3 (vision), §5 (data model); NFR-REPRO
- Milestone: M0
Context and problem statement
smFRET pipelines routinely sever the link between a processed trace and the pixels, coordinates, corrections, and software version that produced it. Once that link is gone, results cannot be re-derived, audited, or trusted. How should Tether structure its data so that provenance is never lost?
Decision drivers
- Reproducibility is a non-functional requirement (NFR-REPRO): a result must
resolve back to a specific verified commit + frozen
conda-lock. - The trace ⇄ movie round-trip (ADR-0002) requires coordinates and patches to travel with every molecule.
- Single self-describing artifact beats a scatter of sidecar files for a desktop, lab-shared workflow.
Considered options
- A. Provenance-first single HDF5 store — coordinates, correction factors,
parameters, labels, and app-version stamped into one
.tetherfile. - B. Processed traces only (CSV/
.dat), provenance in ad-hoc notes. - C. Relational/DB-backed store.
Decision outcome
Chosen option: A. Every datum carries its origin: /molecules hold stable
identity + coordinates, corrections store their factors and method, every
analysis stamps parameters + app-version (git describe). Provenance is a
property of the store, not a convention.
Consequences
- Good: results are auditable and re-derivable; round-trip and staleness tracking become possible.
- Trade-off: a richer schema that must be designed up front — hence the M0 schema freeze (ADR-0005).
- Follow-up: enforced by the
schema-guardgate and the §0.4 DoD "provenance stamped" item.
More information
PRD §5.1 group skeleton; ADR-0002, ADR-0005.