ZEUS is an evidence-first system for understanding what can, and cannot, be claimed about the order of events. B0LT verifies signed records and asserted links. QUIN investigates what evidence to seek next. Neither a signature nor a model answer alone proves physical occurrence or absolute UTC time.
Neither result proves physical occurrence or absolute UTC.
THE PRINCIPLE 01 / 05
An event can be signed without being true.
A sequence can be claimed without being time itself.
ZEUS keeps those boundaries visible, even when the answer is “not enough evidence.”
THE SYSTEM 02 / 05
Exact verification belongs to code. Choosing a useful next question belongs to the investigation layer. Neither gets to quietly become the other.
The part that checks.
Bo verifies source signatures, payload integrity, and asserted sequence or causal links. It preserves gaps, forks, conflict, and unknowns instead of smoothing them into a story.
The part that asks.
QUIN looks at Bo’s bounded state, chooses which available evidence to inspect next, then observes the verified result. Its learned source policies remain experimental and cannot override Bo.
A hard line between them: QUIN can recommend an investigation. Only Bo can admit signed evidence and report what the evidence supports.
EVIDENCE LAB 03 / 05
A simplified illustration of the boundary Bo enforces. Select an evidence state and watch the claimed-order result change. This is an explainer, not a live verifier.
The authenticated link supports this asserted order. It does not prove either event physically occurred or establish UTC time.
LOCAL BENCHMARK SNAPSHOT 04 / 05
These are p99 measurements of warm, resident Python prototype calls on one local host. They show engineering potential, not a production service guarantee.
Frontier read, heuristic action, validation, and JSON from an already-updated Bo state. Two runs of 10,000 calls each.
Final signed-receipt update, Bo decision, deterministic QUIN action, and JSON. Range across seven tested topologies and two runs of 1,000 calls per topology.
What this does not measure: trained-model inference, earlier history setup, external evidence retrieval, disk persistence, network transport, process startup, or concurrent service load. Sustained mixed multi-worker sub-ms latency is not established.
THE WORK AHEAD 05 / 05
ZEUS is a research system taking shape. Here is the line between what exists and what still has to earn its place.
Bo verifies and replays typed, signed chronology in local prototypes. Output levels keep a compact decision separate from full proof materialization.
SCOUT signs run summaries for a read-only comparison lane. Sugar Bush fetch returns have been audited, but are not yet a live Bo feed.
QUIN has a working bounded loop and trained source-choice experiments. The current learned candidates have not passed the complete safety and field-validation gates.
Faster decisions are useful.
Knowable limits make them trustworthy.