Where should we look?
Discover sources, check new metadata and propose leads. A newly discovered site is not automatically an endorsed source or an active subscription.
From AI noise
to a decision you can defend.
Follow the evidence. Inspect the judgement.
See what the agents do—and what remains a human decision.
01 / DISCOVERY
A useful signal is a starting point.
Choose a stage to see what happens and what is retained.
THE ARTIFACTIllustrative workflow, not a live research run. These examples do not evaluate or recommend a real product.
01 / DIFFERENT JOBS, DIFFERENT RESPONSIBILITIES
Automation does the legwork.
Judgement stays accountable.
The cloud collector and local editorial agents have different roles. A model can propose a finding; it cannot certify an experiment or approve its own advice.
What supports the claim?
Gather source-linked evidence, investigate the use case and look for limitations or contradictory findings. Long-form roles run through the local operator workflow.
What can we responsibly say?
Separate unknowns from measured results and prepare a clear, contextual dossier. Deterministic policy controls eligibility; a person approves the exact revision.
02 / MAKE THE SCORE INSPECTABLEA score needs
A score needs
a visible working.
This source-quality example uses the same five dimensions and equal weights as the research rubric. Move a slider or remove enough evidence to see why a score can become unavailable.
ILLUSTRATIVE CALCULATOR · NOT A SOURCE ASSESSMENT
03 / EVIDENCE SETS THE LIMITFour rings.
Four rings.
No shortcuts to endorsement.
The broader radar framework distinguishes what to investigate from what has been tested. The current executable policy supports documentary Assess recommendations only.
Assess
Worth a closer look for a defined problem. Documentary evidence is available; practical results may still be missing.
CURRENT POLICYTrial
A bounded practical evaluation supports trying it in a suitable setting.
REQUIRES TRUSTED TEST EVIDENCEAdopt
Credible operational experience supports use in a stated context.
REQUIRES OPERATIONAL EVIDENCECaution
Material concerns call for a clear explanation of the risk and who it affects.
FRAMEWORK; NOT AUTOMATED HERETrial, Adopt and Caution placements in the design preview are fictional interface examples. They are not outputs supported by the current policy.
04 / WHAT IS KEPTEnough to inspect.
Enough to inspect.
Relevant to the decision.
Backing records retain the source URL, dates, bounded text and the exact model questions behind new decisions. Private records also preserve confidence, model version, sample hashes and the score calculation.
Published editions expose an approved advisory projection. The public site is not a window into private research tables.
EXAMPLE EVIDENCE RECEIPT / STRUCTURE ONLY
- Source
- Original article URL + publisher
- Captured
- Publication and retrieval dates
- Relevant material
- Bounded excerpt + text fingerprint
- Decision
- Questions, answers and confidence
- Score
- Rubric version + arithmetic
- Publication
- Exact reviewed revision
Does popularity put a tool on the radar?
No. Hype can create a research lead. It cannot replace a relevant use case, inspectable evidence, an appropriate policy decision and human review.
How does the system avoid reading the same thing again?
Conditional feed requests, canonical URLs, content fingerprints, persistent caches and request allowances reduce repeat work. A failed or interrupted job can still repeat a provider call; this is not an exactly-once system.
Is Jev running the whole research system?
No. Jev is a stateless decision service used for bounded metadata and source-quality questions. Fedworks owns the registry, evidence store, check queue, editorial workflow and public website.
Is everything running automatically in AWS?
No. The AWS worker supports bounded collection and checks. Long-form editorial work and approval remain operator workflows. Scheduling and paid calls require explicit configuration, and no agent can publish itself.
Can it guarantee finding a launch within a day?
No. Early discovery is a target that needs measured recall, source coverage and failure monitoring. A useful system should report what it missed as well as what it found.
LET’S MAKE IT PRACTICAL
A method is useful.
A decision makes it matter.
Explore the radar, or talk through the question facing your team.