Performance problems surface under real load, exactly when they cost the most. Workloads land on a foundation where the same inputs produce the same outputs, every time, placed where they fit, with the record to prove it.
When something fails, "we can't reconstruct what happened" is a regulatory finding, not an excuse. The Bunny Lab is not an infrastructure scheduler. Routing is a feature. Accountability is the product: deterministic execution, a context graph that keeps its history, and a trust model that assumes some machines are not yours.
DETERMINISTIC KERNELS
One export. Zero imports. Same result, every run.
A kernel is a WASM component with one export and zero imports; the host declares every capability it gets
Deterministic engine, metered execution, epoch limits: no stop-the-world pause at market open
Same inputs, same outputs, same timing
AWSGCPAzureOCINutanixvSphereProxmox30D
Trust ledger
Executions4,812
Audits passed241 / 241
Divergences0
Idle TimeMap · node fleet-ba-112
free (harvestable)host-reserved (p95)
THE CONTEXT GRAPH
Facts with history, not a config file.
A ledgered fact graph: every fact identified by its hash, appended, never overwritten
Two-axis bitemporal queries: what was true, and what we believed at the time
A genuine contradiction surfaces as inconsistent; it is never silently resolved. Agents stage, humans approve
30D
Context graph
Context graph
Subject
Predicate
Value
Valid from
Recorded
billing.proration
runs_on
pool fin-news-prod
2026-08-22
2026-08-22
billing.proration
owner
payments
2026-01-10
2026-01-10
fin-news-prod
trust_class
trusted
2026-07-01
2026-07-01
billing.proration
p95_budget
250 ms
2026-08-01
staged
PLACEMENT
Evaluate where it fits.
Supported execution options compared on workload demand, hardware benchmarks, capacity, price and reliability
Cost per run, not price per hour: an hourly rate never says how long the job takes
Benchmarks carry their provenance, measured or estimated; unknown chips are excluded, not guessed
TRUST POOLS
Mixed-trust fleets, verified by replication.
Every node is enrolled with its own service account and lives in a pool with a trust class
Untrusted executors are verified by replication; the verification plan is part of the job
Restricted data classes only place where policy allows
TRUST LEDGER
A record you can replay.
Every execution recorded with the digests of its inputs and outputs
Under regulatory review an unreconstructable state is a finding. Here it doesn't exist
The DR test passes, and you can replay what the system did on Tuesday
AWSGCPAzureOCINutanixvSphereProxmox30D
Trust ledger
Executions4,812
Audits passed241 / 241
Divergences0
Idle TimeMap · node fleet-ba-112
free (harvestable)host-reserved (p95)
30D
Context graph
Context graph
Subject
Predicate
Value
Valid from
Recorded
billing.proration
runs_on
pool fin-news-prod
2026-08-22
2026-08-22
billing.proration
owner
payments
2026-01-10
2026-01-10
fin-news-prod
trust_class
trusted
2026-07-01
2026-07-01
billing.proration
p95_budget
250 ms
2026-08-01
staged
BEFORE YOU START
Before you start
Is this a scheduler?
No. Routing is a feature. Accountability is the product: reproducible results, a context graph with history, and a record of every execution.
What is a kernel?
A WASM component with one export and zero imports, content-addressed, executed by a deterministic engine with metered limits. Nothing reaches the host that was not declared.
Where does the record live?
In your environment, signed and immutable, readable by your auditors.
What's next
Start with the 48-hour assessment or bring one workload. We agree the scope in the first conversation.
