evaluate 0 0 0

phionyx/airep-embedded-evaluation-profile

published by @phionyx

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
208
Last 30 days
86
Likes
-
Size
50 KB
Created on HF
2026-09-13
Age
1 months ago

Description

AIREP Embedded Evaluation Profile v0.1

This is not a training dataset or benchmark. It is a Hugging Face distribution mirror of an

experimental evaluation-evidence profile, its schema basis, registry and fixtures. The canonical

specification history lives in the AIREP GitHub repository. Byte identity between this mirror and

its source commit is a distribution-integrity property, not independent scientific verification.

Experimental evaluation-evidence contract for AIREP v0.2.… See the full description on the dataset page: https://huggingface.co/datasets/phionyx/airep-embedded-evaluation-profile.

Tags

license:othersize_categories:n<1Kformat:jsonmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantarxiv:2608.21363region:usai-evaluationai-safetyai-governanceauditprovenancereproducibilityairep

README current version from Hugging Face


pretty_name: AIREP Embedded Evaluation Profile v0.1
license: other
license_name: airep-dual-license
license_link: https://github.com/halvrenofviryel/ai-runtime-evidence-protocol/blob/85f99b79fbe78377a53f03d2fafc5331eab42f71/LICENSE-CC-BY-4.0.txt
tags:

  • ai-evaluation
  • ai-safety
  • ai-governance
  • audit
  • provenance
  • reproducibility
  • airep

AIREP Embedded Evaluation Profile v0.1

This is not a training dataset or benchmark. It is a Hugging Face distribution mirror of an
experimental evaluation-evidence profile, its schema basis, registry and fixtures. The canonical
specification history lives in the AIREP GitHub repository. Byte identity between this mirror and
its source commit is a distribution-integrity property, not independent scientific verification.

Experimental evaluation-evidence contract for AIREP v0.2.

Profile identity

Profile identifier airep.embedded-evaluation
Profile version 0.1
Carrier AIREP wire version 0.2
Schema basis digest sha256:ce57bd493d1a3166051bc9029972d6be215793edcd3f1b314534959659b3da8b
Canonical GitHub source https://github.com/halvrenofviryel/ai-runtime-evidence-protocol/tree/85f99b79fbe78377a53f03d2fafc5331eab42f71/spec/airep/v0.2/profiles/embedded-evaluation
Source commit of this mirror 85f99b79fbe78377a53f03d2fafc5331eab42f71 — fixed source snapshot for the files in this publication (including its historical README)
First publication commit 2b62580aa05302cc978c2cb5a2a55d8a0a215b55 — schema, registry and both fixtures are byte-identical to that commit; only PROFILE_README.md changed since
Status Experimental companion profile — not a ratified standard, not a fifth AIREP artifact family

The five files here are byte-identical to the source commit's spec/airep/v0.2/profiles/embedded-evaluation/
directory (the profile README is published as PROFILE_README.md); MANIFEST.sha256 lists their digests.

Publication policy and current exporter scope

Frozen, source-pinned publication; not a continuously updated mirror of canonical main.
The schema, registry, fixtures and PROFILE_README.md retain the exact source pin above.
Card clarifications do not refresh those source bytes. A new source edition requires an explicit
new publication revision and pin.

PROFILE_README.md is historical source text, not a current parser-support statement.
The current exporter parses LightEval results_*.json; native Inspect .eval and arbitrary
OpenEvals formats are not currently supported. Explicit timezone-aware start/end timestamps
are required; a timezone-naive filename is never silently labelled UTC. Profile validation is
not an assurance result or safety certification. Declared context is recorded, not established.

What it binds

Evaluator and evaluated-party context with a declared independence basis; target model/system/
checkpoint identity; evaluator access tier, capabilities and limitations; harness/task/environment/
safeguard configuration; explicit measurement execution and observed states; content-addressed
evidence with visibility/withholding state; redactions, deviations, retries and incidents; and
optional verification references.

Three negative-state guards are enforced by the schema: NOT_RUN can only pair with
NOT_MEASURED/NOT_APPLICABLE; PASS/FAIL require RAN/PARTIAL; INVALIDATED never yields
PASS/FAIL. example.not-measured.json preserves NOT_RUN → NOT_MEASURED explicitly.

Claim limitations

A profile PASS means only that the profiles["airep.embedded-evaluation"] payload validates
against the exact digest-pinned schema basis. It does not establish:

  • evaluator independence or competence (independent-declared is a declaration, not proof);
  • that the named model, revision or configuration was the one actually executed;
  • truth, representativeness, completeness or reproducibility of observations;
  • model safety, alignment, security, regulatory conformity or deployment suitability;
  • any AIREP assurance class — Core/Authenticated/Witnessed semantics are unchanged;
  • validity of a platform verifyToken, which the profile carries only as platform-verification evidence.

Without an accepted basis for the exact identifier, the correct verifier result is NOT_EVALUATED,
never an implicit pass.

Files

  • embedded-evaluation.schema.json — self-contained Draft 2020-12 schema (basis)
  • registry.json — digest-pinned registry for the AIREP v0.2 --profile-bases mechanism
  • example.measured.json — synthetic measurement that ran and failed its declared criterion
  • example.not-measured.json — synthetic planned evaluation that did not run
  • PROFILE_README.md — the canonical profile README at the source commit
  • MANIFEST.sha256 — digests of the mirrored files

Related

Licensing

The canonical AIREP repository is dual-licensed by file type: code under Apache-2.0
(LICENSE), specification text and schemas under CC BY 4.0 (LICENSE-CC-BY-4.0.txt), with a
patent non-assertion statement. The mirrored files are specification/schema/example material and
follow the CC BY 4.0 terms; this card declares license: other so those file-specific terms are
not collapsed into a single repository-wide SPDX claim.

Paper

AIREP: A Protocol for Per-Decision Evidence in AI Runtime Governance
https://arxiv.org/abs/2608.21363

The Embedded Evaluation Profile is an experimental AIREP v0.2 companion profile.
Its presence does not create an AIREP assurance class or constitute independent validation.

README history 6 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-09-18docs: link AIREP arXiv v2 paperdfebf606.1 KB
    Loading...
  2. 2026-09-14docs: clarify source pins and evaluation exporter boundaries (#1)e2417505.8 KB
    Loading...
  3. 2026-09-13Card: not a benchmark/training dataset; source commit 85f99b7 (profile README...9cd45a65 KB
    Loading...
  4. 2026-09-13Card: state mirror scope (not a benchmark), pin source commit 85f99b7 (profil...322b3704.4 KB
    Loading...
  5. 2026-09-13Publish AIREP Embedded Evaluation Profile v0.1 (mirror of 2b62580aa05302cc978...da0c33d4.4 KB
    Loading...
  6. 2026-09-13initial commit4b43ad30 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 2 threads · 1 open · 3 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-09-15[bot] Conversion to Parquetopen1 💬#2
    Loading...
  2. 2026-09-14PRdocs: clarify source pins and evaluation exporter boundariesmerged2 💬#1
    Loading...

Related datasets same stage · evaluate

Read further

Metadata

License
other

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)
from datasets import load_dataset
ds = load_dataset("phionyx/airep-embedded-evaluation-profile")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="phionyx/airep-embedded-evaluation-profile", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/phionyx/airep-embedded-evaluation-profile
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration