Cross-Agent Planning Graph Visibility

Mechanism

Cross-agent planning graph visibility is the mechanism by which one agent lets trusted peer agents observe parts of its speculative landscape without surrendering control of its plan. In a multi-agent coordination context, an agent may selectively expose portions of its planning graph to trusted peer agents through a policy-governed visibility interface. The exposed portions are read-only copies of selected planning graph branches. They are not the live branches themselves, and the peer never gains write access to the exposing agent's planning graph.

The copies are transmitted with their speculative markers intact. This is the property that makes the exposure safe. A planning graph branch is speculative content, structurally separated from verified execution memory, and the speculative marker is what records that separation. By preserving the marker on every exposed copy, the interface lets a peer agent observe the exposing agent's speculative landscape without contaminating its own verified execution memory. The peer reads the exposed branches as another agent's projections, not as established facts about the world.

The Policy-Governed Visibility Interface

The visibility interface is governed by the exposing agent's own policy configuration, not by a central coordinator and not by the receiving peer. Exposure is a decision the exposing agent makes about its own state, under its own policy constraints. The agent that holds the plan decides what, to whom, how deep, and for how long the plan becomes visible.

The policy configuration specifies four parameters for any exposure. First, which branches may be exposed: a deployment may permit only eligible branches to be exposed, or may additionally permit introspective or delegable branches. Second, which peer agents are authorized to receive exposed branches. Third, the maximum exposure depth, which fixes how many levels of the planning graph subtree are visible to the receiving peer. Fourth, the exposure duration, which fixes how long the exposed copy remains accessible before automatic revocation.

Bounded Depth and Automatic Revocation

The maximum exposure depth bounds visibility structurally. The exposing agent can reveal the top levels of a planning graph subtree, where the coordination-relevant shape of its plan lives, while withholding the finer speculative detail beneath. Depth is a property of the exposure, set per the exposing agent's policy, rather than a property the peer can negotiate.

The exposure duration bounds visibility in time. An exposed copy is not a permanent grant. It remains accessible only for the configured duration, after which it is automatically revoked. Because the exposed branches are read-only copies rather than references into the live planning graph, revocation withdraws the peer's view without disturbing the exposing agent's own plan, which continues to evolve independently of whatever copies were shared.

Coordinated Speculative Reasoning

The purpose of the interface is to enable coordinated speculative reasoning across agents that keep their own planning authority. A peer agent that can observe an exposing agent's speculative landscape can align its own planning graph construction with that landscape. Concretely, the peer can identify complementary branches that fit with the exposing agent's projected actions, avoid redundant speculation that would duplicate work the exposing agent is already pursuing, and detect potential conflicts between the agents' respective projected futures before either agent commits.

All of this happens without centralizing planning authority in a single coordinator. Each agent still constructs and owns its own planning graph. Coordination emerges from the mutual observation of independently generated speculative landscapes, not from a central authority that assembles or imposes a combined plan. This distinguishes the visibility interface from executive-graph aggregation, which is a separate mechanism in which an executive engine collects planning graphs across a scope to synthesize a system-level structure.

Lineage Recording on Both Sides

Every exposure is recorded in lineage on both sides of the transaction. Each exposure event is recorded in the exposing agent's lineage, so the exposing agent retains a governance record of what it revealed, to which peer, and when. The receiving agent's lineage records the receipt of exposed speculative content, and it records that receipt with the appropriate speculative marker.

Recording the speculative marker on the receiving side is the second line of defense against contamination. The marker travels with the copy in transit and is then captured in the receiver's lineage, so the receiving agent's own governance machinery treats the received content as speculative and prevents its inadvertent promotion into verified state. The two-sided lineage record means an auditor can reconstruct which agents observed which speculative landscapes, while the markers ensure observation never silently becomes belief.

Composition With the Forecasting Engine

Cross-agent visibility composes with the rest of the forecasting engine. It operates on the same planning graph branches the forecasting engine constructs and classifies, and it reuses the branch classifications, eligible, introspective, and delegable, to decide what is exposable. The speculative markers it preserves are the same markers the containment layer uses to keep speculative content separated from verified execution memory; the visibility interface extends that separation across agent boundaries rather than weakening it.

The interface is distinct from, and complementary to, the other multi-agent constructs in the forecasting engine. Executive-graph aggregation builds a synthesized system-level plan by collecting agents' planning graphs within a scope. Planning graph delegation transfers a delegable substructure from a parent agent to a child agent with re-scoped context. Cross-agent visibility transfers neither authority nor a writable structure: it shares a read-only, time-bounded, depth-bounded view so peers can reason in light of one another's plans while each retains its own.

Disclosure Scope

Cross-agent planning graph visibility, comprising the policy-governed visibility interface through which an agent selectively exposes read-only copies of selected planning graph branches to trusted peer agents with their speculative markers intact, the policy configuration that specifies which branches may be exposed, which peer agents are authorized, the maximum exposure depth, and the exposure duration before automatic revocation, the resulting coordinated speculative reasoning in which peers align planning, identify complementary branches, avoid redundant speculation, and detect conflicts without centralizing planning authority, and the recording of each exposure in the exposing agent's lineage and the receipt in the receiving agent's lineage with the appropriate speculative marker, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart) at Section 4.23. This article describes that disclosed mechanism. The scope extends to embodiments in which the exposable branch classes, authorized peer sets, exposure depths, and exposure durations vary by policy, provided the exposed branches remain read-only copies carrying speculative markers and each exposure is recorded in lineage.