Vigil
Vigil is a small framework for behavioural verification under controlled variation.
It executes a system across declared variations, evaluates the resulting behaviour with checks, and records everything in a single structured report.
Vigil is system-agnostic and focuses on behavioural properties rather than performance benchmarks.
Running Vigil
Run Vigil on a project directory:
vigil [project_dir] [--trace]
Vigil opens an interactive menu to:
- select a specification
- run all variations and checks
- inspect previous reports
Reports are written next to the specification file.
Project layout
A project directory contains:
- exactly one backend
- one or more specifications
- optional local checks and variations
project/
llm_backend.py
language.yml
prompts.yml
checks/
entity_types_agree.py
variations/
differ_language.py
Files ending in *.report.yml are treated as results, not specs.
Specification
A specification defines what behaviour is tested.
At minimum it contains:
- hypothesis
- inputs
- variations
- checks
Optional metadata:
- title
title: Behavioural verification of DatsLlm with respect to language
hypothesis: Annotation behaviour remains stable across language routing
inputs:
- text: "Hello world"
variations:
- type: set_input
language: de
checks:
- matches_baseline
Backend
A backend wraps the system under evaluation.
It combines:
- function configuration (what is executed)
- environment configuration (how and where it runs)
Backends implement:
update_environment(environment)
compute(input, function) -> output
The framework manages execution, isolation, and cleanup automatically.
Variations
Variations introduce controlled changes.
Each variation targets exactly one domain:
- input
- function
- environment
Variations:
- transform inputs or patch configuration
- are applied explicitly and sequentially
- never inspect outputs
Baseline execution is represented explicitly with none.
Vigil also supports a small convenience expansion:
variations: [{ type: repeat, times: N, do: [...] }]
This expands into a flat list by repeating the entries in do N times.
Syntactic sugar
The specification format provides light syntactic sugar for common patterns. This sugar expands into ordinary variations and does not add new semantics.
Supported forms include:
- repetition of a variation block
These constructs exist purely to reduce duplication in specs. After parsing, the engine operates only on plain variations.
Checks
Checks evaluate observed behaviour.
Three intents exist:
- unary
- reference
- group
Checks may be:
- assertive (produce PASS / WARN / ERROR)
- diagnostic (INFO only)
Checks operate only on recorded slices and never trigger execution.
Reports
Each run produces a single structured report containing:
- metadata (title, hypothesis, timestamps)
- backend configuration
- all inputs and variations
- all check results
Reports are YAML by default and meant for inspection, comparison, and archiving.
Vigil is intentionally small, explicit, and extensible.
New backends, variations, and checks can be added without modifying the core.
Release files for vigil-bv 0.1.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| vigil_bv-0.1.2.tar.gz | 25.0 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| vigil_bv-0.1.2-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 54.2 kB
Release files / vigil_bv-0.1.2.tar.gz
| Download URL | vigil_bv-0.1.2.tar.gz |
|---|---|
| Size | 25.0 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
2f669bb5c1a98368ef9f1d97ccda9f25966d0ef895fc12bbf88832f77d8816da
|
|
BLAKE2b-256 checksum How to use checksums |
2faadc0c9344100f535fdf81e765039b60b84cb7faa7addb3f208bcc5623763c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.7
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Feb 10, 2026.
Transparency logRelease files / vigil_bv-0.1.2-py3-none-any.whl
| Download URL | vigil_bv-0.1.2-py3-none-any.whl |
|---|---|
| Size | 29.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
98b6011d6580991a1561dc2d02a7cd0d6606546d6c124e747114ef5a93576b99
|
|
BLAKE2b-256 checksum How to use checksums |
9407001a5a7907fa24c78f7f6c43b90c2bd899dc926786fd2135879180f3fbb0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.7
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Feb 10, 2026.
Transparency log