Veriscope is the CLI for auditable training runs
CLI-first training monitor for auditable capsule bundles: record runs, validate artifacts, and gate cross-run diffs by explicit comparability predicates.
# Run monitored training (runner wrapper)
$ veriscope run gpt --outdir run_01 -- --dataset tiny --seed 123
# Validate the capsule bundle
$ veriscope validate run_01
# Diff two runs only when comparable under contract
$ veriscope diff run_01 run_02An artifact contract, not a dashboard
Veriscope's core output is a capsule bundle you can validate, report, and diff. Comparisons are only admitted when runs remain comparable under contract.
Auditable capsule bundles
Every run emits standard files such as window_signature.json, results.json, and
results_summary.json, plus optional governance and provenance artifacts.
Validate / Report / Inspect
Use the CLI to verify bundle integrity, inspect key fields, and generate text or Markdown reports without trusting embedded hashes.
Diffs gated by comparability
Cross-run diffs are blocked unless the runs are non-partial and still match the same window signature, preset identity, and compare-time governance checks.
Governance without mutation
Record approvals and notes as append-only overlays without altering raw artifacts. What happened and what your team decided stay separate and auditable.
Window-scoped calibration
Calibration emits FAR, delay, and overhead results together with either a preset candidate or an explicit rejection, all scoped to one window-signature hash.
Runners + pilot kit
Start with runner wrappers for CIFAR, nanoGPT, and Hugging Face workflows, then use the pilot harness for control versus injected-pathology runs with shareable outputs.
From training run to decision record
Declare a window signature, run alongside training, emit capsule artifacts, validate them, then report and diff only when the contract admits comparison. Governance overlays remain append-only.
- Decisions live in artifacts with pass, warn, fail, and skip states
- Comparability gates prevent silent-drift comparisons
- Governance overlays preserve raw artifacts without mutation
- Calibration protocol remains scoped to the window that produced the preset
# 1) Run a monitored training job (runner wrappers)
$ veriscope run gpt --outdir OUTDIR -- --dataset tiny --seed 123
# 2) Validate the capsule bundle
$ veriscope validate OUTDIR
# 3) Produce a shareable report
$ veriscope report OUTDIR --format md > report.md
# 4) Diff two runs only when comparable under contract
$ veriscope diff OUTDIR_A OUTDIR_B
# 5) Record a manual judgement without mutating raw artifacts
$ veriscope override OUTDIR --status pass --reason "Known infra noise"Run Veriscope in a pilot
Produce auditable artifacts, validate capsule integrity, and gate cross-run comparison under an explicit artifact contract.