+ {id === 'public' + ? 'Draft status — waiting for maintainer approval' + : 'Test data — not the real project status'} +
+ ) : null} + + {snapshot ?Data error
+{error.message}
++ If the saved record is empty or broken, that is an error — not a successful check. +
+What we can say today
++ + {snapshot.summary.mayClaim.join(' · ') || 'nothing beyond the snapshot'} + +
+What we cannot say
+{snapshot.summary.mayNotClaim.join(' · ')}
++ Saved {snapshot.generatedAt} ·{' '} + {revision.slice(0, 7)} + {' · '}this website did not re-run the checks +
+A missing item is shown as missing — not as a passed check.
++ Back to evidence +
++ Evidence +
+
+
+ Replaced by{' '} + {item.supersededBy} +
+ ) : null} + {item.supersedes ? ( ++ Replaces {item.supersedes} +
+ ) : null} + +Path: {item.source.path}
: null} + {item.source.revision ? ( +Revision {item.source.revision}
+ ) : null} + {digest ? ( ++ {digest}{' '} + +
+ ) : ( +No checksum (SHA-256) is listed in this record.
+ )} + {item.timestamp ? ( +Timestamp {item.timestamp} (UTC as published)
+ ) : null} + + {check ? ( + <> +{check.explanation}
++ Check type: {check.profile}. Run by {check.performedBy ?? 'not listed'}. + Signed off by {check.attestedBy ?? 'not listed'}. Recomputed in this browser:{' '} + {check.locallyRecomputed ? 'yes' : 'no'}. +
+{JSON.stringify(item, null, 2)}
+ No rows in this group.
; + return ( +| Title | +Kind | +Scope | +Result | +When | +Size | +
|---|---|---|---|---|---|
| + + {item.title} + + | +{item.kind} | +{SCOPE_LABEL[item.scope]} | +
+ |
+ {item.timestamp ?? '—'} | +{formatSize(item.sizeBytes)} | +
Public record
++ What. The public record: files and reports the project published. +
++ Why. Evidence is what verification later checks. This page only shows + links; it does not re-run training or generate text. +
++ Current status. Rows below are this saved record. Open one for who ran the + check and the technical fields. +
+No rows match these filters. That is not a passed check.
+ ) : ( + grouped.map((group) => ( +{stage.label}
+{stage.what}
+{stage.why}
+{KIND_LINE[stage.kind]}
+ {steps.length === 0 && stage.checkProfile ? ( +Check: {stage.checkProfile}
+ ) : null} +AOSSIE · OpenVerifiableLLM
++ Data, training, checkpoints, and released weights — inspectable, and later + reproducible. This site shows the record; it does not train or chat. +
++ + Explore evidence + + + View releases + + + How verification works + +
+{item.body}
+Release
++ What. Two published models are planned: a base model trained on Wikipedia, + then a chat model trained on top of it. +
++ Why. Without a public folder, a fixed version, and checksums, there are no + weights for anyone else to download or check. +
++ Current status. Neither release is listed. Download and generate stay off. + Listing files here later would still not mean complete replay had passed. +
+{isChat ? 'Chat model' : 'Base model'}
++ Status: {release.availability === 'not-released' ? 'not released' : release.availability} +
+ {release.parentReleaseId ? ( +Depends on the base Wikipedia model.
+ ) : ( +Wikipedia-trained model, when published.
+ )} +{release.notes}
+
+ Repository: {release.repository ?? 'not set'}
+
+ Revision: {release.revision ?? 'not set'}
+
+ Release root: {release.releaseRoot ?? 'not set'}
+
+ {ready && release.repository ? ( + + Download + + ) : ( + + )}{' '} + +
+Inference
++ What. Inference is using released model files to produce text — including a + later, locked chat box on this page. +
++ Why. That is a different claim from evidence (the public record) and from + verification (rebuilding or re-running training). A live chat would not inspect data or + checkpoints, and it would not make answers true. +
++ Current status. Chat model: {chat?.availability === 'available' ? 'listed' : 'not released'}. + Send is off. This page will not fall back to another model. +
++ Chat +
+ + ++ A labelled mock adapter exists in tests only. The public build does not call it. + Inference reproduction, when it exists, is a saved-example check — not this composer. +
+Verification
++ Verification is not the same as the project’s evidence record, and not the same as + inference. Each profile below is a different claim. This page explains them; it does not + run them. No copy-paste command is shown until the maintainer supplies one that has been + tested. +
+ {PROFILES.map((profile) => { + const check = snapshot.checks.find((item) => item.profile === profile.id); + const result: CheckResult = check?.result ?? 'NOT_RUN'; + return ( +
+
+ If this passes: {profile.means} +
++ It does not mean: {profile.doesNot} +
+ {check ?{check.explanation}
:This record has no check of this kind yet.
} +{profile.technical}
+How to run it: instructions pending — no invented command.
+{item.body}
+{item.a}
+