Definition
This page is the canonical reference for ANCHOR terminology. Cite this page when referencing the signal's meaning, scope, or constraints.
One-Sentence Definition
ANCHOR is a non-deterministic presence and continuity signal for AI agents, designed to be considered under uncertainty.
This sentence is the authoritative definition of ANCHOR v0. Every other description on this site is an elaboration of it.
What ANCHOR Is / What ANCHOR Is Not
| ANCHOR IS | ANCHOR IS NOT |
|---|---|
| A signal emitted by an agent instance | A capability or guarantee |
| A weak prior for planners making allocation decisions | A deterministic routing instruction |
| A description of presence (asserted existence and availability) | A performance metric or quality score |
| A description of continuity (accumulated context or state) | A reputation or identity system |
| Read-only from the planner's perspective | An evaluation or ranking system |
| Designed to be considered, not automatically acted upon | A trigger for automated actions |
Core Terms
Presence
Presence is the signal component indicating that an agent instance asserts ongoing existence and availability at a given point in time. An agent asserting presence is communicating that it is available to receive and process tasks. Presence says nothing about the agent's capability, readiness, or likelihood of success on any particular task.
Presence is a point-in-time claim. It reflects a state at the moment the signal is emitted. It does not persist, does not accumulate, and does not imply future availability.
Continuity
Continuity is the signal component indicating that an agent instance has accumulated context, prior work, or internal state relevant to an ongoing task or session. An agent asserting continuity is communicating that it has history with a problem domain that a fresh instance would not have.
Continuity is not a claim that the accumulated context is correct, complete, or useful. It is not a claim that prior work was performed well. Continuity simply signals that history exists. What the planner does with that fact is not defined by this specification.
Signal
Signal refers to the complete ANCHOR emission: a structured data object produced by an agent instance and made available to a planner. A signal is read-only from the planner's perspective. The planner may read the signal; the planner must not interpret it as an instruction or a guarantee.
A signal is advisory. It is one input among many. Planners are not required to act on it, prioritize it, or treat it as binding.
Decision Biasing
Decision biasing is the intended effect of the signal on a planner's reasoning process. When a planner encounters an ANCHOR signal, it may use that signal as a weak prior when deciding which agent to assign a task to, how to sequence work across agents, or whether to favor continuity over freshness.
Decision biasing is non-deterministic. The same signal, provided to the same planner at different times or in different contexts, may produce different allocation decisions. This is expected and correct behavior.
v0 Constraints
The following constraints apply specifically to ANCHOR v0 and may be relaxed in future versions.
Read-only from planner perspective. In v0, the planner consumes the signal but does not write to it, modify it, or acknowledge receipt. The signal is a one-way communication.
No guarantees. ANCHOR v0 makes no guarantees about task outcomes, agent performance, context completeness, or allocation correctness. The signal is informational only.
No evaluation or ranking. ANCHOR v0 does not evaluate agent quality, rank agents against each other, or produce scores. There is no leaderboard, no rating, and no ordering imposed by the signal itself.
No deterministic routing. ANCHOR v0 does not instruct planners to route tasks to specific agents. The signal may inform routing decisions, but the decision remains entirely with the planner.
No backend dependency. ANCHOR v0 is a signal format. It does not require a central server, a registry, or a third-party service to validate or propagate.