Skip to content

crosswalks: map the evidence axes to the agent-evidence-vocabulary registry - #275

Open
astrogilda wants to merge 2 commits into
aveproject:mainfrom
astrogilda:crosswalk-aee-vocabulary
Open

astrogilda wants to merge 2 commits into
aveproject:mainfrom
astrogilda:crosswalk-aee-vocabulary

Conversation

@astrogilda

@astrogilda astrogilda commented Sep 16, 2026 •

Copy link
Copy Markdown
Contributor

Both sides of this mapping share an author: I designed the evidence axes in #214 and maintain agent-evidence-vocabulary. It is a structural check and cannot count as independent corroboration.

This is the first crosswalk here for the evidence axes, in both of the directory's formats, and it validates against schema/crosswalk-1.0.0.schema.json. The two axes take the value sets and the weakest-input composition of observation_vantage and observation_directness in agent-evidence-vocabulary, and scripts/write_verification_basis.py already composes both into verification_basis.

None of the 80 record files fetched at 1e29789 carries evidence_vantage, evidence_method or verification_basis, which are optional in v1.1.0, so every row is evidence: inferred.

Edited 2026-09-22: added the first paragraph at chaksaray's request, shortened the rest, and pointed the registry link at its current path after the 2026-09-16 rename.

The crosswalks directory carries nineteen files and none of them maps
evidence_vantage, evidence_method or verification_basis, the three
evidence axes the record schema declares.

This adds the mapping in both formats the directory already uses. It
validates against schema/crosswalk-1.0.0.schema.json.

The mapping is unusually close because it is not a taxonomy pairing.
evidence_vantage takes substrate or artifact and evidence_method takes
intercepted or reconstructed; the aee-vocabulary registry defines
observation_vantage and observation_directness over byte-identical
value sets with the same weakest-input composition. verification_basis
is the composition of both, which scripts/write_verification_basis.py
already computes, so it maps structurally onto the pair.

Coverage is zero of eighty. Each of the 80 record files was fetched at
1e29789 and none carries any of the three axes; all three are optional
in v1.1.0. Every row is therefore marked inferred rather than emitted.

Signed-off-by: Sankalp Gilda <sankalp.gilda@gmail.com>
The registry repository was renamed from aee-vocabulary to
agent-evidence-vocabulary on 2026-09-16. The old URL still redirects, so
nothing was broken, but the crosswalk should name the registry as it is
named now.

Both crosswalk files are renamed to match, and the markdown cross-link
to the JSON follows the new filename. The mapping, the coverage counts
and the pinned commits are unchanged.

Signed-off-by: Sankalp Gilda <sankalp.gilda@gmail.com>
@astrogilda astrogilda changed the title crosswalks: map the evidence axes to the aee-vocabulary registry crosswalks: map the evidence axes to the agent-evidence-vocabulary registry Sep 16, 2026
@chaksaray

Copy link
Copy Markdown
Contributor

The coverage methodology and the mid-PR rename handling are both exactly right, worth naming specifically: pinning the commit you read coverage at, marking every row inferred rather than claiming false coverage, and catching the registry's own rename mid-PR with a clean, separate commit rather than folding it silently into the first one.

One thing worth saying directly rather than fixing quietly. agent-evidence-vocabulary is your own repository. That's a materially different situation from every other crosswalk here, which map AVE against something built by someone with no hand in AVE's own design. This maps the evidence axes you designed against a vocabulary you also authored. The structural claim itself, byte-identical value sets, same weakest-input composition, is real and checkable, and I'm not questioning it. But the crosswalk can't carry the same weight as independent corroboration, and it shouldn't be read that way by anyone finding it later.

Two things worth doing before this merges. First, the note field should say plainly that both sides of this mapping share an author, the same disclosure standard you set yourself on #94, unprompted, before anyone asked. Second, worth naming out loud that this is the second time in a row an external dependency turned out to be your own project, reason_code was the first, and neither was volunteered upfront. Not reading bad faith into it, the work itself has been excellent throughout, just worth naming as a pattern now rather than catching it silently a third time.

With the disclosure added, glad to merge this.

@astrogilda

Copy link
Copy Markdown
Contributor Author

Thanks, chaksaray, for saying it directly. I maintain probityai/agent-evidence-vocabulary, so both sides of this mapping share an author, and I edited the description to state it. You are right about the pattern too, since reason_code and this crosswalk both cite it and I volunteered neither.

The note field gets that sentence in the next commit, and the reason_code proposal I owe on #219 will open with it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants