Public tooling scaffold. Release status: scaffolded. License posture requires human review.
epi-bench is the Franzabner public scaffold for Energy Per Intelligence (EPI) calculation and reporting discipline. It is meant to separate measurement claims from measurement readiness: schemas, examples, and review gates can be public before real benchmark results are public.
This repository is not a released benchmark suite, not a validated evaluation result, not a dataset release, not a model release, and not a Hugging Face Space.
| Item | Status |
|---|---|
| Public status | Tooling scaffold |
| Release status | Scaffolded |
| Package status | Prototype code present; no release claim |
| Benchmark status | No validated public benchmark results |
| Dataset status | No released dataset |
| Hugging Face status | No model, dataset, or Space created by this repo |
| License posture | Existing license file is unchanged; human review required before license changes or external reliance |
epi-bench is the public-safe tooling lane for:
- EPI calculation interfaces.
- Power trace and benchmark result schemas.
- Synthetic example workflows.
- Future report generation discipline.
- Review gates before any EPI result is called measured, validated, released, or benchmarked.
The code in this repo may contain prototypes and synthetic tests. Those do not establish benchmark validity.
| Repo | Role |
|---|---|
| franzabner-proof-stack | Public navigation and proof-routing spine |
| energy-per-intelligence | EPI research framing |
| epi-meter | Planned public-safe measurement instrument scaffold |
| hf-card-templates | Hugging Face release-readiness and boundary templates |
EPI = joules per output unit / reviewed quality score
Any public use of the denominator must state the evaluation protocol. A benchmark score, task score, or domain rubric is not valid just because it appears in an example file.
- Prototype EPI calculator structure.
- Synthetic examples and tests.
- Placeholder data directories.
- Public-safe schema examples.
- Planned command-line interface shape.
- Documentation for claim discipline and future review.
This repo does not claim:
- validated benchmark results;
- released EPI reports;
- released datasets;
- hosted models;
- hosted Hugging Face datasets, models, or Spaces;
- deployed infrastructure;
- client or customer use;
- revenue outcomes;
- production readiness.
Human review is required before:
- publishing real power traces;
- publishing benchmark outputs;
- describing a result as evaluated, benchmarked, released, or validated;
- creating or linking Hugging Face artifacts;
- changing license posture;
- using the repo as evidence of client, deployment, or production claims.
Keep this repository public-safe. Do not add private model weights, private training corpora, private endpoints, credentials, customer files, private infrastructure, or sealed YOSO-YAi implementation details.
Build the harness. Mark the status. Do not let a scaffold pretend to be a result.