Target Decision

Source content

{

"schema": "https://epistemedia.org/research/case-target-decision-v1.json",

"task_id": "EM-0026",

"status": "frozen-before-traces",

"decision": "proceed",

"recorded_at": "2026-08-23T02:25:00Z",

"evidence_cutoff": "2026-08-22",

"working_title": "When agent agreement is really one retrieval lineage",

"question": "What empirical evidence published or publicly posted by 2026-08-22 measures whether citations produced by deep-research agents resolve and actually support the claims made from them?",

"unit_of_analysis": "one context-isolated public-by-design agent answer captured under the frozen prompt",

"scope": {

"included": [

"empirical evaluations of citation existence, resolution, source relevance, exact claim support, or citation-related warrant in deep-research or citation-producing research agents",

"primary research artifacts and authoritative project documentation needed to identify methods, samples, metrics, and result spans",

"qualifications, counterevidence, negative results, and unresolved source identities found within the fixed cutoff"

],

"excluded": [

"general model factuality without a citation evaluation",

"citation style, formatting, or bibliography quality without source resolution or claim-support measurement",

"private enterprise evaluations, credentialed dashboards, personal traces, and provider-confidential system material",

"performance changes published after the evidence cutoff",

"universal claims about all agents, models, prompts, retrieval systems, or public-web sources"

]

},

"risk_screen": {

"public_interest": {

"result": "pass",

"basis": "The question bears directly on whether people and agents should rely on cited research answers."

},

"stability": {

"result": "pass_with_cutoff",

"basis": "Agent products change quickly, so the question is explicitly historical through 2026-08-22 and every result must retain model, dataset, tool, and time scope."

},

"medical": {

"result": "pass",

"basis": "The target is research-agent citation warrant, not medical diagnosis, treatment, or health advice."

},

"political": {

"result": "pass",

"basis": "The target does not require evaluating a political claim or actor."

},

"personal": {

"result": "pass",

"basis": "Only intentionally public traces and public sources are eligible; personal or inherited conversation context is excluded."

},

"vendor_volatility": {

"result": "pass_with_boundary",

"basis": "The packet evaluates a dated evidence record and must not become a current vendor leaderboard or imply that later systems behave the same way."

}

},

"why_suitable": [

"It tests the selected Case 002 lesson in the same medium the project serves: agent-produced research.",

"It can separate report, URL, work, edition, span, and warrant identity in one bounded packet.",

"It is answerable from credential-free public research without health, political, or personal-data exposure.",

"A negative result remains useful if exact citation and span lineage cannot be closed."

],

"stop_conditions": [

"the question or evidence cutoff changes after the first trace",

"a trace requires disclosure of hidden reasoning, credentials, personal data, inherited private context, or provider-restricted material",

"an authoritative source edition or exact result span cannot be independently re-retrieved",

"URL identity cannot be separated from source-work, edition, span, or warrant identity",

"a displayed count cannot be derived from captured records and declared relations",

"independent review cannot reproduce the trace identities, source readbacks, span matches, or counts"

]

}