Tools and Methods: instructor model responses

Worked snapshot queries, a provenance record and a raw-capacity estimate.

These responses use synthetic data. This is a public, formative answer key.

1. Query results

Snapshot A: s1 and s2, 2 rows. Exclude s3 for score, s4 for region and s5 for postsynaptic identity. Snapshot B: s1, s2, s3 and s6, 4 rows. Exclude s4 and s5. The change consists of a score revision for s3 and a newly listed row s6.

With a strict score > 0.80 rule, s2 is excluded and the counts become 1 and 3. An undocumented boundary convention can therefore change both results without any dataset revision. The fixture has no evidence of tissue change: new or revised annotations do not establish biological growth.

2. Model methods record

“For the synthetic teaching dataset in the archived query source, I selected snapshot-a, postsynaptic label 101, region label core, and score ≥ 0.80. The endpoint was the number of qualifying rows, not distinct presynaptic partners or physiological strength. The selected IDs were s1 and s2, giving n = 2 from five input rows; three rows were excluded for the reasons recorded above.”

Append the actual execution date, exact command, saved output, Python version and source SHA-256 emitted by the program. Archive the source itself; a hash alone cannot recover a missing file. On the paper route, record the worksheet revision, manual filtering procedure and peer re-run instead of claiming a code execution. Do not copy a fabricated hash or environment from an exemplar.

For a real query, additionally record the datastack/table and schema, materialization version and timestamp, segmentation state, version-appropriate object IDs, region definition and coordinate units/frame, client/dependency versions, and restrictions on access or redistribution. This fixture has region labels only and no coordinates or segmentation graph; those fields are not applicable, not silently inferred. No stochastic operation occurs here, so a seed is not applicable either.

If a prior version is missing, inspect archived outputs, notebook history, manifests and query logs. A count alone does not uniquely identify its source version. If the provenance cannot be established, label that result unreproduced and rerun a documented analysis. Do not declare either count correct merely because both are plausible.

The CAVEclient materialization guide describes versioned queries; the MICrONS versioning tutorial shows the association between versions and timestamps. The toy snapshot labels are not identifiers from either service.

3. Capacity and transfer

Convert the volume to 100,000 × 100,000 × 40,000 nm.
Divide by voxel size: 10,000 × 10,000 × 1,000 voxels.
Product: 100,000,000,000 voxels. At one byte per voxel: 100 GB raw.
Three total copies require 300 GB before overhead.
One ideal transfer: 100 × 10⁹ / (100 × 10⁶) = 1,000 seconds, about 16.7 minutes. This is not an acquisition-time estimate or a transfer guarantee.

Omissions include segmentation labels, image pyramids, chunk padding, metadata, intermediate processing arrays, backups beyond the three copies, and retry/protocol overhead. Compression can change storage and throughput; its ratio was not supplied. Neither hardware price nor staff time can be inferred from this byte estimate. Accept binary units only if explicitly labeled and converted correctly.

Feedback guide

Score four dimensions 0–2 each: exact filtering/counts, recoverable provenance, capacity arithmetic/units, and interpretation/limitations. Two means explicit and correct, one means a recoverable omission, and zero means absent or contradictory. A proficient response has at least 6/8 and no zero in provenance or interpretation. This is a local teaching rubric, not a validated assessment instrument.

Common feedback: “Which snapshot?” for a missing version; “Is 0.80 included?” for an ambiguous threshold; “Where are the archived inputs?” for a hash without files; and “What changed in the evidence?” for an unsupported biological explanation.

Teaching material: CC BY-SA 4.0, NeuroTrailblazers.