Search beyond
the brief.
Oracle opens adjacent, independent, historical, and unconventional search territories, then connects the evidence supporting each direction.
A team asks, “What should we pursue?” Ordinary AI researches the obvious directions. Quarry searches a wider opportunity field and connects evidence the original framing missed. Oracle produces a ranked decision brief with the evidence, uncertainty, smallest test, and walk-away condition. Quarry records the decision, observes the outcome, and learns. Across five exact historical clinical-development replays, Oracle put 21.1% more of the future within reach at the same 0.9% review limit than GPT-5.6 Sol.
Models make intelligence abundant.
Quarry decides where to direct it.
The expensive failure is not a weak answer. It is the strongest direction never entering the comparison before a team commits its next experiment, hire, or dollar.
Search beyond
Put more of the future
within reach.
At the same 0.9% review limit, Oracle put 21.1% more later-positive opportunities in view than GPT-5.6 Sol across five exact historical replays.
SAME 0.9% REVIEW LIMIT+21.1%more of the future within reach
Make every decision
useful twice.
The Opportunity Outcome Graph preserves what was knowable, what Oracle surfaced, what the team chose, and what happened later. Each outcome sharpens the next search.
lock Local by default · evidence retained
From QuarryProof,
Proof,
not promises.
Quarry is measured by what reaches action before the outcome is known, not by how impressive a research report sounds.
More of the future at the same attention budget
Oracle’s shortlist captured nearly half of the opportunities that appeared later, beating GPT-5.6 Sol by 21.1% at the same review limit.
A separate temporal exam reproduced the advantage
Oracle V5.1 again placed more later-positive opportunities within reach than GPT-5.6 Sol and replayed with zero rank mismatches.
Two papers submitted on reliable agents and valid discovery tests
The work covers deterministic control for tool use and fail-closed calibration for discovery benchmarks. Submitted; no acceptance claimed.
Quarry OracleBring Quarry the decision
Bring Quarry the decision
that matters next.
Start with one consequential question.