Skip to content

Latest commit

 

History

7 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Coordination Readout

A specification for read-model surfaces and explicit actuation contracts over the Agent Coordination Substrate.

The substrate defines how signals, claims, and induction behave. This specification defines how agents and operators observe that state, evaluate it, and decide what to do next — without hidden mutation, without implicit execution, and without unbounded evidence payloads.


Scope

This spec covers This spec does NOT cover
Pressure and trend readout semantics Signal deposit behavior (see advisory spec)
Recommendation packets (evidence-bound suggestions) Claim acquisition and conflict (see enforcement spec)
Recommendation outcome ledger (audit correlation) Saturation detection and manifest generation (see induction spec)
Empirical calibration (trust from outcomes)
Causal threads (cross-zone continuity)
Operator cognition readout
Explicit actuation boundary

Core Documents

  • SPEC.md — Normative behavioral contract
  • schemas/ — JSON Schema definitions (Draft 2020-12)
    • stigmergic-delivery.schema.json — Wire/read model schema
    • operator-cognition.schema.json — Operator cognition readout schema (v2)
  • examples/ — Concrete JSON payloads
  • traces/ — Replayable reference trace corpus
  • conformance/ — Conformance test expectations

Design Principles

  1. Read surfaces never mutate. Observing pressure, reading a recommendation, or querying a causal thread produces no side effects.

  2. Recommendations are not commands. A recommendation packet exposes a next_explicit_command hint. Executing that command is a separate, auditable action.

  3. Evidence is bounded. Readouts expose summaries and references, never unbounded raw source payloads.

  4. Calibration is visible, not hidden. Weak historical confidence is exposed as flags, never silently suppressed or auto-applied.

  5. Causal threads are derived. They are read-only projections over shared evidence — not a new project-management authority layer.


Conformance Levels

Level Name Capability
1 Read Model Signal-zone pressure, trend, recommendation packets, bounded evidence
2 Outcome-Aware Recommendation snapshots, command events, outcome correlation
3 Thread-Aware Bounded causal threads and continuity summaries
4 Calibration-Aware Empirical trust metadata from outcome history

Relationship to Agent Coordination Substrate

This specification consumes outputs from all three substrate layers:

┌──────────────────────────────────────────────────────┐
│  Agent Coordination Readout (this spec)              │
│  Recommendations, outcomes, calibration, threads     │
├──────────────────────────────────────────────────────┤
│  Agent Coordination Substrate                        │
│  ├── induction/   (Layer 3)  → manifest proposals    │
│  ├── enforcement/ (Layer 2)  → claims, gates         │
│  └── advisory/    (Layer 1)  → signals, zones        │
└──────────────────────────────────────────────────────┘

An implementation of this readout spec requires a conforming substrate (advisory Level 1 minimum). Higher readout conformance levels may require higher substrate conformance levels.


Status

v0.1.0 — Initial publication. Read model (Level 1) normative. Outcome ledger, causal threads, and calibration are normative at their respective levels.


License


Contributing

Contributions welcome. Open an issue or pull request.

About

Specification for read-model surfaces and explicit actuation over the Agent Coordination Substrate

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors