Full scans

Four recursive profiles with explainable fan-out.

A profile sets cumulative compatibility, depth, family, source, time and safety bounds. Evidence-qualified results may create further compatible work until a bound or the frontier is reached.

Profile comparison

Increase breadth deliberately.

Each level adds compatible coverage, not a fixed result count. Actual work depends on the clue, retained evidence, available sources, permissions and request budget.

Deterministic fixture measurements
Local, network-free profile fixture
ProfileFixture jobsPlanned credits
Quick11
Balanced33
Deep Research44
Advanced Investigation55
These values come from one deterministic, PostgreSQL-backed reserved example fixture with standalone plug-in subprocesses and zero external requests. They verify profile behavior; they are not expected customer result counts or live-source coverage.

Recursive fan-out

Evidence quality shapes the next branch.

The orchestrator uses retained evidence and plug-in metadata to decide whether a result is eligible for more work. It does not turn every discovered value into an unbounded target.

Evidence classes stay distinct

Verified, observed, candidate and context objects retain their own tier. Candidate or shared context does not become a direct asset by presentation alone.

High-value branches go first

Deterministic ranking considers evidence class, validation, target affinity, source quality, capability family and cost.

Every branch remains bounded

Profile, stage, family, depth, source, runtime, safety and credit controls can suppress or stop scheduling with a recorded reason.