Scientific decision records
This directory records consequential scientific and API decisions, including decisions to defer or reject a method. It is an audit trail, not a claim that every decision is permanent.
Status vocabulary
- Draft: proposed and open to change; not normative.
- Accepted: current project policy.
- Rejected: considered but not adopted.
- Superseded: replaced by a later numbered record, which must link back.
Decision records state their date, evidence available at that date, alternatives, consequences and an explicit revisit trigger. Later evidence is appended or linked; the original reasoning is not silently rewritten.
Relationship to other documents
docs/drafts/contains working designs and questions. Drafts may change freely and must not be cited as package guarantees.benchmarks/protocol-*.mdfreezes hypotheses, scenarios and thresholds before aggregate execution.- benchmark JSON files retain complete machine-readable outcomes.
- reports describe results after execution and link to the governing decisions.
Index
| Record | Status | Decision |
|---|---|---|
| SDR-0001 | Accepted | Record consequential scientific decisions |
| SDR-0002 | Accepted | Keep initial control-free methods experimental |
| SDR-0003 | Accepted | Represent subtraction and division as different transformations |
| SDR-0004 | Accepted | Require explicit alignment for externally derived data |
| SDR-0005 | Accepted | Do not infer photometry channel identity from row position |
| SDR-0006 | Accepted | Require explicit NWB session metadata and avoid invented hardware |
| SDR-0007 | Accepted | Require explicit TDT stream, channel, and epoc semantics |
| SDR-0008 | Accepted | Preserve open metadata while versioning readiness profiles |
| SDR-0009 | Accepted | Preflight structural incompatibility without accessing outcomes |
| SDR-0010 | Accepted | Report scalar mixed models as secondary sensitivity summaries |
| SDR-0011 | Accepted | Regularize irregular clocks only through prospective, auditable policy |
| SDR-0012 | Accepted | Protect missing event windows instead of silently reconstructing them |
| SDR-0013 | Accepted | Report event selection before and after preprocessing |
| SDR-0014 | Accepted | Resample animals for peri-event uncertainty |
| SDR-0015 | Accepted | Use a frozen public NWB golden path |
| SDR-0016 | Accepted | Require named, justified multiverse alternatives |
| SDR-0017 | Accepted | Couple preprocessing outputs and separate incompatible units |
| SDR-0018 | Accepted | Summarize robustness within complete unit lanes |
| SDR-0019 | Accepted | Reject irrelevant parameters and declare incompatibility prospectively |
| SDR-0020 | Accepted | Store one reference signal and the complete multiverse ledger |
| SDR-0021 | Accepted | Verify directory bundles and mark standalone NWB trust |
| SDR-0022 | Accepted | Separate byte identity from scientific reproduction |
| SDR-0023 | Accepted | Sign manifest attestations with domain-separated OpenSSH |
| SDR-0024 | Accepted | Generate deterministic deposits from neutral metadata |
| SDR-0025 | Accepted | Stop Zenodo automation at validated drafts |
| SDR-0026 | Accepted | Organize documentation by scientific question |
| SDR-0027 | Accepted | Hold out complete groups for event-kernel models |
| SDR-0028 | Accepted | Retain weak event-kernel validation and keep the API experimental |
| SDR-0029 | Accepted | Treat grouped event-kernel intervals as conditional sensitivity |
| SDR-0030 | Accepted | Delegate behavioral learning trajectories to Unspool |
| SDR-0031 | Accepted | Treat spontaneous transients as a method family |
| SDR-0032 | Accepted | Preserve external behavior semantics at typed boundaries |
| SDR-0033 | Accepted | Retain validity masks without compressing time |
| SDR-0034 | Accepted | Fit only explicit matched-pulse clock transforms |
| SDR-0035 | Accepted | Compare event-kernel models only on common evidence |
| SDR-0036 | Accepted | Reconstruct kernels from explicit typed bases |
| SDR-0037 | Accepted | Model event history as explicit within-session modulation |
| SDR-0038 | Accepted | Model variable-duration behavior with physical intervals and progress |
| SDR-0039 | Accepted | Treat predictor-family drops as paired predictive sensitivity |
| SDR-0040 | Accepted | Keep simultaneous kernel bands opt-in after failed calibration |
| SDR-0041 | Accepted | Use format-specific readers behind one acquisition boundary |
| SDR-0042 | Accepted | Separate transient detection from quantification |
| SDR-0043 | Accepted | Infer transient contrasts at the animal level |
| SDR-0044 | Accepted | Require comparability evidence before longitudinal handoff |
| SDR-0045 | Accepted | Never bridge gaps or state boundaries in spectral analysis |
| SDR-0046 | Accepted | Require explicit pairs and shared evidence for multisignal analysis |
| SDR-0047 | Accepted | Separate prospective optical masks from observed validity |
| SDR-0048 | Accepted | Freeze transient thresholds and retain waveform QC |
| SDR-0049 | Accepted | Make interval-policy order explicit and auditable |
| SDR-0050 | Accepted | Preserve ndx-pose values and declare link omissions |
| SDR-0051 | Accepted | Name observable multiscale estimands and preserve denominators |
| SDR-0052 | Accepted | Require independent identification for optical unmixing |
| SDR-0053 | Accepted | Treat spatial edges as within-mouse evidence |
| SDR-0054 | Accepted | Make forward models primary and deconvolution conditional |
| SDR-0055 | Accepted | Materialize animal estimates before population inference |
| SDR-0056 | Accepted | Form interactions from animal-condition contrasts |
| SDR-0057 | Accepted | Preserve domain denominators at one population boundary |
| SDR-0058 | Accepted | Require explicit axes and pointwise support for population curves |
Use four-digit monotonically increasing identifiers. Copy
template.md, fill every heading, and add the record to this index.