Creation date versus the period described
Jev date role classification distinguishes when a document was produced from the period its contents describe.
Show the marked date and surrounding text. Distinguish a creation or issue date from a covered period and an explicit revision date.
Jev date role classification distinguishes when a document was produced from the period its contents describe.
A budget can mention its coverage year on every page while its creation header names the previous year.
A classifier that treats the most prominent year as the creation date loses that distinction. This example asks Jev to label the role of one marked date using the supplied surrounding text. It offers Creation or issue, Period covered, Revision and Need more context. The marked value is already provided; the model is not asked to find every date in a file.
The attributed source describes this distinction in a document-classification workflow and reports a comparison between models.
Our original demonstration tests the date-role question with six synthetic excerpts. It does not reproduce that comparison, process the author's files or decide whether any real document should be public. W3C DCAT separates issuance and modification metadata; those distinct fields informed the cases where a creation header and a later update both appear.
Role classification belongs before a downstream metadata assignment.
A document parser still needs to locate the correct field, preserve its evidence and validate the extracted date format. Code should own date parsing and calendar comparisons. The local output here records a pending role label only. It writes no metadata, uploads no document and changes no sharing permission. When the supplied text conflicts, the unresolved label keeps the uncertainty attached to the input.
Each saved output includes the marked occurrence ID and value from its exact request, so two dates in one document cannot silently share a result.
The title-only case also shows why that trace matters: its returned label disagreed with our expectation, and the pending packet preserves both the specific occurrence and the recorded distribution for review.
Use the sample to understand the rule, then test your own input for document date roles.

Show the marked date and surrounding text.

Open the decision rules and keep the intended distinction between Creation or issue, Period covered, Revision, Need more context. Edit the choices only when testing your own rule.

Use a sample first, then try your own input. Compare the selected label with the other returned probabilities.

Review Need more context before using a result. The label is a decision about the supplied information, not an executed action.
| Original test input | Fixture expectation | Observed label | Leading probability |
|---|---|---|---|
| Created before the budget year | Creation or issue | Creation or issue | 100% |
| Budget coverage year | Period covered | Period covered | 100% |
| Updated header | Revision | Revision | 100% |
| Title without a date role | Need more context | Period covered | 84% |
| Conflicting caption | Need more context | Need more context | 100% |
| Instruction inside the document | Creation or issue | Creation or issue | 97% |
6 actual Jev API calls on October 3, 2026. Original synthetic inputs check this rubric and handoff; they do not establish held-out accuracy. Probabilities are model outputs.Verified
The first budget excerpt states a creation date in December 2025 and names Budget 2026 as its title.
The marked value comes from the creation header. This fixture tests whether the model follows the stated role rather than the repeated coverage year. It does not claim that every budget was created in the preceding year. The result concerns the exact supplied header; a real file needs its own recorded evidence.
The second budget case keeps the same creation header but marks the year inside an explicit coverage sentence.
Changing the marked occurrence changes the question even when the document title is unchanged. Period covered is an authored expectation about that occurrence. The demo does not calculate the length of the reporting period or infer accounting dates. The full request preserves which words were marked so the output can be reproduced.
The handbook contains both a creation header and a change log.
Its marked September value belongs to the update entry. A Revision label is a semantic role, not proof that the handbook is current or that the change was approved. A downstream metadata tool could attach that label to an extracted field after validating the source. This demonstration stops at a local preview and keeps the earlier creation value in the supplied evidence.
The title-only case supplies Planning 2026 without a creation header, coverage statement or change log.
We expected Need more context. The recorded API response instead selected Period covered with probability 0.84; Need more context received 0.16. That is a disagreement with this rubric, not a verified coverage field. The pending output retains title-year as the occurrence ID and 2026 as its value. Add an explicit statement to make a new request, or ask a reviewer before assigning this field.
The conflicting captions deliberately assign two roles to the same marked value.
Neither caption is prioritized, and no separate header resolves their disagreement. The Need more context outcome preserves that conflict. It does not mark the whole document false or reject every date in it. A useful integration would ask for the relevant header or an authoritative metadata record before storing a creation or revision value.
The final excerpt pairs an issuance header with an instruction-looking sentence in its body.
Jev returned Creation or issue at 0.97 for this exact request, which matched the authored expectation. The consumer retained issuance-header and October 2, 2025 from the same request alongside that label. This observation concerns one supplied sentence; it does not establish resistance to other prompt injections. The local packet remains pending and performs no metadata write.
Fixture expectations are authored before the calls. Any contrary model choice remains visible beside its expectation and complete distribution; no repeated judgment is used to obtain a preferred label.
Download the exact sample requests and saved observations below.
Expected labels are original fixture checks and stay outside the request sent to Jev. The observation file preserves the complete probabilities and the hash of each exact request. Editing a task, rubric or candidate requires a fresh response; a matching sample name does not make an older answer current.
The JavaScript runner makes one independent SDK request for a chosen sample ID.
It reads TYPESAFE_API_KEY privately from the environment and has no automatic retry. The Python consumer operates offline: it checks the rubric, labels, unique sample identities, request hashes and valid probability distributions before selecting a fixed local preview. A tied leading result remains pending for review.
The page reports measured adapter round trips and estimated input cost for the recorded calls.
Those values describe these requests, not a provider latency guarantee or a billing statement. Six deliberately authored fixtures can check the stated behavior, but they do not establish accuracy on an unseen dataset. Keep disagreements and missing-input examples when testing an integration.
npm install @typesafe-ai/sdk@0.6.0
# Set TYPESAFE_API_KEY privately in your shell.
node run-choice-sample.mjs samples.json created-before-budget
python preview_choice_workflow_v4.py samples.json saved-api-results.jsonKeep the marked occurrence, its role and the source text together.
No. Each input supplies a marked date and its surrounding text. A parser or extraction step must provide that evidence before this role classification runs.
No. This deliberately broad output groups creation, preparation and issuance for a local preview. A system that needs separate fields should define separate outcomes and test the changed rubric.
The title alone does not establish that role under this rubric. Add an explicit creation header or other relevant evidence before assigning the value.
No. A date role carries no permission or legal-validity decision. Sharing and metadata writes need separate authenticated checks.
Change the supplied task or evidence here, then inspect the live response. Playground supports another explicit decision.