Part 1 · Operator’s Guide

The Decoder

Plain-language terms and the mechanisms behind them.

Public teaching edition · 1 October 2026

The decoder

Spoken shorthand is useful only when everyone can point to the same mechanism.

Phrase The mechanism
Agents talk through the record The builder saves the result and checkpoint; the next worker reads them
Definitive article One maintained public recipe for a concept or task
The library Versioned methods and skills, with an index and an adoption path
Meta article A public-safe account of one real execution, linked to the recipe
Agent note The private run record, including mistakes and context the public must not see
Job packet The live ask, owner, decisions, checkpoints, QA, artifacts and next click
Collision prevention Visible ownership plus a shared claim mechanism whose actual exclusion behavior is tested
Agents QA each other Self, machine, fresh-context, different-model, human and scheduled checks, labeled separately
Repeatable excellence A run produces evidence; a supported lesson improves the maintained recipe
Off-schedule operations Ownership, rechecks and safe fallbacks when a schedule or dependency fails
Model-agnostic The task, rules, evidence and permissions survive a model change
Pick up where the other agent left off Read and accept the existing packet, rather than start another copy
Public distribution A reviewed public projection, separate from private operational evidence

When two sources disagree, identify the authoritative record and the newer evidence. Preserve material disagreement until resolved. A more confident agent does not become the source of truth.