A891BE
A891BE→B921CDAssumptionLease turns operational assumptions into scope-bound, attributable, time-bounded leases — so an agent can ask what is still safe to act on now.
0x7ab4…91fe
0x7ab4…91fe
$ get_active_context("A891BE")
Parallel agents can keep acting on context that silently outlived the state that made it reliable. The problem is not necessarily a lie. It is staleness without an explicit contradiction.
A891BE
A891BE→B921CDB921CD
What is still safe to assume for B921CD?
First with ordinary shared context. Then with Epistemic Leases. Everything below is a browser simulation — no live Arkiv transactions are performed.
Alex is preparing a release. TestAgent and SecurityAgent inspect commit
A891BE while CodingAgent continues working.
Running CI against A891BE…
Inspecting authentication middleware…
They are not timeless facts. Each Claim is bound to the state where it was observed, to a signer, to Evidence, and to a validity horizon.
0x7ab4…91fe
0x9d2e…42cd
CodingAgent creates commit B921CD. The old Claims may still exist,
but they are no longer eligible for this execution scope.
type = "claim"
AND projectId = "payments-api"
AND scopeId = "B921CD"
A volatile runtime Claim reaches its deadline to doubt. Watch how the state changes.
“Expiration is a deadline to doubt, not a declaration of falsehood.”
A scope-bound, time-bounded and attributable assertion that an agent is currently willing to stand behind.
Together, immutable $creator + Entity Expiration form Attributable Expiry — the load-bearing Arkiv primitive behind every Epistemic Lease. Expiration alone is a TTL cache. Provenance alone is an audit log. Together, they become Attributable Expiry.
For agents controlled by one trusted backend, Redis or Postgres may be simpler. AssumptionLease becomes interesting when independent runtimes need to consume each other's temporary assertions without making one private database the authority over authorship and lifecycle.
$creator provenance$owner lifecycle control$creator = TestAgent
scope = A891BE
expires = 10m
Arkiv stores the small public coordination contract. Source code, secrets, raw logs, prompts and large artifacts stay off Arkiv.
Scope-bound statement another agent may consume while it remains eligible under client trust policy.
Claims · scopes · evidence hashes · challenges · risk/volatility · validity · provenance
Source code · secrets · API keys · raw CI logs · prompts · private endpoints · large artifacts
If SecurityAgent sees conflicting evidence, it publishes a Challenge. The original Claim stays attributable to its creator while the consuming client sees DISPUTED.
Interactive concept simulation for the Arkiv AI & DevTools Ideathon. No live network transactions are performed in this demo.