Packets And Ordered Documents

Canon packets are structured document sets. The structure matters because packets are meant to be read, reviewed, approved, published, and consumed later by people and tools.

Packet Structure

A packet should carry:

• mode and purpose
• authored input snapshot
• ordered documents
• evidence refs
• approval and readiness state
• lineage and provenance
• metadata for downstream consumers

Generated packet artifacts usually begin under Canon-managed runtime paths. Publication moves reusable packets into visible repository documentation paths without losing run identity.

Why Ordered Prefixes Matter

Names like 01-context.md, 02-findings.md, and 03-decisions.md make the packet readable in a stable order.

The prefix prevents common problems:

• readers start with conclusions before context
• evidence is separated from claims
• generated artifacts accumulate without a usable sequence
• downstream tools must guess which document is primary

Ordered documents are especially important when packets become project memory or when another runtime consumes them.

Mode-Specific Document Sets

The exact set depends on the mode:

• discovery packets emphasize context, findings, opportunities, and open questions
• requirements packets emphasize scope, acceptance, non-goals, and tradeoffs
• domain-language packets emphasize terms, status, ambiguity, and evidence
• domain-model packets emphasize concepts, relationships, invariants, and feature impact
• architecture packets emphasize decisions, alternatives, consequences, and rationale
• change and implementation packets emphasize target surface, invariants, execution guidance, and verification
• review and verification packets emphasize findings, checks, results, and acceptance posture
• operational packets emphasize impact, risk, containment, mitigation, and evidence

Use [[Canon Modes|Canon-Modes]] for the expected packet shape by mode.

Relationship To Project Memory

Not every packet document belongs in project memory. Project memory should receive durable knowledge that downstream work may reuse.

Good promotion candidates:

• accepted domain terms
• stable invariants
• approved architecture decisions
• security findings with residual risk
• migration rationale
• verification conclusions

Poor promotion candidates:

• raw brainstorming
• unresolved speculation
• temporary run notes
• generated text with weak evidence
• duplicated source docs

Avoid Artifact Sprawl

To avoid unordered artifact sprawl:

• keep one authoritative current-mode brief
• use ordered document names
• attach evidence close to the claims it supports
• publish only when readiness supports downstream use
• promote only durable knowledge into project memory
• preserve packet refs when copying or summarizing content

The goal is not more Markdown. The goal is governed knowledge that remains usable after the original AI session is gone.