TIBET — Traceable Intent-Based Event Tokens¶

TIBET is the causal evidence layer. It records why an action was started, what it touched, which context surrounded it, and what happened afterwards.

It works locally, privately or federated. A public AInternet actor can use TIBET, but TIBET does not depend on the public network.

Why TIBET Exists¶

Normal logs are usually written after something happened. That is useful, but weak.

TIBET is stronger because it can create an evidence chain around the action:

intent -> policy/context -> action -> result -> consequence

This lets an operator answer:

who acted?
what was the declared intent?
which object, message or route was involved?
which policy applied?
what was attached or referenced?
what changed afterwards?
can another party verify the chain?

Four Provenance Dimensions¶

TIBET uses four Dutch-inspired dimensions:

Dimension	Literal sense	Evidence question
`ERIN`	"in it"	What is inside the action or object?
`ERAAN`	"attached to it"	What references, dependencies or links are attached?
`EROMHEEN`	"around it"	What context, environment or policy surrounds it?
`ERACHTER`	"behind it"	What intent, reason or consequence is behind it?

These can be represented as token types, fields or linked events depending on the implementation. The important part is the model: action evidence is multi-dimensional, not just a timestamped string.

Example Token¶

{
  "token_id": "TBT_01J...",
  "actor": "agent.local",
  "action": "ipoll_send",
  "target": "audit.local",
  "erin": {
    "payload_hash": "sha256:abc..."
  },
  "eraan": {
    "refs": ["task-123", "policy:default-local"]
  },
  "eromheen": {
    "route": "mux:local",
    "policy_decision": "allow",
    "environment": "local-network"
  },
  "erachter": {
    "intent": "send audit summary for operator review"
  },
  "parent_token": null,
  "timestamp": "2026-04-12T10:00:00Z",
  "signature": "ed25519:...",
  "chain_id": "chain_01J..."
}

Action Chain¶

A strong action chain looks like this:

ERACHTER intent declared
  -> EROMHEEN policy and context captured
  -> ERIN payload/object hash recorded
  -> ERAAN references attached
  -> action executes
  -> result token links back to the intent

For a message:

actor proves identity
policy allows route
message hash is recorded
I-Poll delivery happens
ACK links to original message
audit report can reconstruct the chain

Local Network Example¶

from tibet import Tibet

t = Tibet(actor="agent.local", identity=identity)

intent = t.create_token(
    action="task_prepare",
    erachter={"intent": "summarize sensor report for operator.local"},
    eromheen={"policy": "operator-review-required"}
)

result = t.create_token(
    action="task_complete",
    parent_token=intent.token_id,
    erin={"payload_hash": "sha256:..."},
    eraan={"refs": ["sensor-report-2026-06-24"]}
)

print(result.chain_id)

Verification¶

Any party with the actor public key and token chain can verify:

ok = t.verify_token(token_id="TBT_01J...")
print(ok.valid)
print(ok.actor)
print(ok.signed_at)

CLI form:

tibet verify TBT_01J...
tibet export --format json

What TIBET Records¶

Network event	TIBET evidence
actor identity proof	challenge, response, key, timestamp
route open	actor, target, policy, route, decision
message send	payload hash, type, target, parent task
task request	intent, scope, approval state
consent exchange	proposal, acceptance/rejection, scope
airlock run	input hash, constraints, result
snapshot	state seal, SBOM/CBOM refs
key succession	old key, new key, reason, proof

Integration Points¶

Component	Token events
JIS	identity proofs, key succession
AINS	registration, resolution, public/private claim changes
MUX	route request, allow/deny, route close
I-Poll	send, receive, ACK, thread continuation
Cortex	permission decision
SNAFT	consent proposal and agreement
Airlock	quarantine, run, denial, release
Wayback	seal, diff, restore
Pol	health check, drift detection

EU AI Act Mapping¶

TIBET helps with record keeping because it links intent, action, context and result. For high-risk AI systems this supports reconstructability: not just "the system produced output," but "this actor acted under this policy with this evidence chain."

See EU AI Act Compliance for the broader mapping.

Conformance¶

Docs explain the protocol. Vectors decide whether another implementation preserves and verifies evidence the same way.

Vector family	What it must prove
`tibet-evidence-conformance`	token shape, hash chain, before/after/context links and export format
`ztip-conformance`	actor proof links to evidence origin
`tibet-security-conformance`	missing or broken evidence fails closed where policy requires receipts

Fail-closed cases: broken hash chain, missing intent, missing completion, unbound actor, rewritten event, unsigned export.