Calibration, determinism, and regression protection
`invarlock calibrate` arrives, determinism utilities mature, and regression harness + golden tracking help prevent silent policy drift.
Release: InvarLock 0.3.2 - First-class calibration and golden-output tracking
Highlights
- New calibration CLI (
invarlock calibrate) and tuning modules. - Determinism flows for repeatable presets and runs.
- Regression harness + tracked golden to catch silent gate/output shifts.
This is the release where calibration stops being “something you do around InvarLock” and becomes something the framework supports directly. Having invarlock calibrate as a first-class command makes tuning policies and guard behavior feel more systematic—and easier to explain to other people.
Determinism utilities also get a serious boost here, which matters if you care about repeatability beyond a single machine or day. And the regression harness + golden tracking is the quiet hero: it’s basically insurance against subtle drift. If something shifts in gates or outputs, you want to catch it early—before it becomes a messy “which commit changed this?” hunt.
Sources
Website documentation explains the maintained workflow. The changes described here belong to InvarLock v0.3.2; the tagged release record preserves that historical scope.
More in Release
Explore nearby related posts.
Release
Token-weighted paired statistics and stricter release gates
Token-weighted paired bootstrap lands across the pipeline, strictness toggles expand, and CI/release pairing expectations become explicit and enforceable.
Release
Large-model reload stability and B200 controls
Fixes a GPU memory leak during reload fallback, hardens B200 scripts, and adds practical controls for acceptance ranges and overhead measurement.
Release
Fail-closed baseline pairing with lower-memory retries
CI/release baseline pairing is fail-closed (pairing evidence is required), and adapters reduce peak memory during retries via chunked snapshot/restore.