Grant applications
Trace outcome claims back to source artifacts before submission.
Test a claimEvidence-grounded verification
Compile a claim against cataloged DataHub artifacts, classify each segment as observed, derived, or unsupported, and record the verdict as a verification assertion.
Mock and live modes · Browser-local history · Citation-level traceability

CLAIM SEGMENTS
$6.30/hr average wage increase
[A-102]Improved employment outcomes
[A-102, A-208]40% improvement
no citation01 / 04 · Pipeline //
A compact evidence pipeline that makes the source, inference, and unsupported gaps visible at every step.
Step 01
Parse a claim into precise, independently verifiable segments.
INPUTStep 02
Match each segment to cataloged DataHub artifacts and metadata.
MAPStep 03
Classify support as observed, derived, or unsupported—with reasoning.
ASSESSStep 04
Record verification assertions so the verdict stays catalog-visible.
WRITE-BACK02 / 04 · Workflows //
Designed for moments when a claim must survive scrutiny—not merely sound convincing.
Trace outcome claims back to source artifacts before submission.
Test a claimSeparate directly observed facts from derived conclusions and gaps.
Test a claimGive reviewers a compact packet with citations, reasoning, and assertion status.
Test a claim03 / 04 · Browser history //
04 / 04 · System notes //
Observed means an artifact states the claim directly. Derived means the conclusion follows from cataloged evidence, with the inference and its limits explained.
No. When no cataloged artifact backs a segment, the result is explicitly marked unsupported instead of being presented as evidence.
The app uses its mock catalog so the complete classification and packet workflow can still be tested safely.
Packets are stored locally in the current browser. They are not shared between browsers or devices.