Deliverables

Deliverable split

The workflow produces two artifact families:

• Public RFC deliverables — publishable documents for readers and implementers.
• Evidence artifacts — internal proof material used to build, audit, validate, and repair the public documents.

Public artifacts must stand on their own without exposing workflow internals. Evidence artifacts must preserve enough traceability to explain why the public text says what it says.

Public RFC deliverables

Public deliverables may include:

• RFC draft text.
• RFC section structure promoted into finished prose.
• Definitions and terminology.
• Normative requirements.
• Explanatory notes that belong in the RFC itself.
• User-facing or implementer-facing examples approved for publication.
• Appendices approved for publication.
• A public manifest listing publishable files only.

Public deliverables must not include:

• Internal claim IDs or matrix IDs.
• Source locators unless intentionally part of the specification.
• Discovery-aid queries, node IDs, edge IDs, or retrieval traces.
• Orientation-context maintenance notes.
• Scout notes.
• Prompt text.
• Tool transcripts.
• Review arguments, correction history, or escalation notes.
• Unresolved placeholders.

Evidence artifacts

Evidence artifacts may include:

• Run briefs.
• Context packets.
• Source-map snapshots, discovery output, and direct-source notes.
• Source scout reports.
• Claim ledgers.
• Capability matrices.
• Section maps and working notes.
• Example reports and rendered-text extraction notes.
• Mechanical validation reports.
• Review findings.
• Reader on-ramp reports.
• Prose-risk reports.
• Clean-room packets.
• Correction logs.
• Escalation logs.
• Clean-room proof reports.
• Final evidence manifests.

Evidence artifacts stay in the run-local work area unless a human explicitly chooses to promote a subset.

Deliverable matrix

Deliverable	Audience	Contains	Excludes	Pass criteria
RFC draft	Public readers and implementers	Finished prose, rules, examples	Workflow provenance	A clean-room reviewer can use it without hidden artifacts
Public manifest	Release owner and public reviewers	Publishable file list and status	Evidence-packet detail	Lists only public deliverables
Claim ledger	Reviewers and controller	Claim IDs, authority, evidence, status	Polished prose	Every public claim maps to an accepted row
Capability matrix	Reviewers and drafters	Required capabilities, coverage, gaps, cited public sections/examples	Unrelated source notes	Every required capability is covered, scoped out, open, or blocking
Example report	Reviewers	Inputs, outputs, rendered extraction method, validation status, claim links	Narrative filler	Examples match accepted claims and survive rendering
Mechanical report	Controller and drafters	Render/structure/reference checks and failures	Source re-summaries	Required checks pass or are explicitly escalated
Review report	Reviewers and controller	Findings, severity, disposition	Public-only packaging	All blocker findings are closed or escalated
Reader report	Controller and drafters	First-loss point, jargon-before-definition, opening-arc verdict	Hidden evidence	One isolated public-text review per target document
Prose report	Controller and drafters	Tell-cluster findings, allow/rewrite/upstream routing	Hidden-source citations	Wording-only defects are distinguished from upstream input defects
Clean-room packet	Controller and clean-room reviewer	Public-only review packet assembled for isolated proof	Hidden evidence and private notes	Matches the reviewer visibility rules exactly
Correction log	Controller	Failed gate, root artifact, rerun scope, disposition	Unrelated churn	Each correction repairs the earliest faulty artifact
Escalation log	Human decision-maker	Questions that change scope, semantics, or publication treatment	Routine review chatter	Only true decision points remain
Evidence manifest	Auditors and release owner	Retained evidence artifact list	Public-only manifest substitution	Reconstructs the proof chain

Stage outputs

• Scope outputs a run brief.
• Context assembly outputs a context packet.
• Source mapping outputs source-map records.
• Source scouting outputs bounded scout reports.
• Claim admission outputs accepted, rejected, inferred, and open claims.
• Capability mapping outputs required-capability coverage and gap records.
• Drafting outputs the section map and public draft source.
• Example validation outputs rendered-text-backed example reports.
• Mechanical validation outputs rendered public artifacts and validation reports.
• Specialist review outputs review findings and, when needed, escalation records.
• Reader on-ramp review outputs one isolated reader report per target document.
• Prose-risk audit outputs wording and route-upstream findings plus the clean-room packet.
• Correction routing outputs correction-log entries and any unresolved decision requests.
• Clean-room proof outputs a clean-room report.
• Final packaging outputs separate public and evidence manifests.

Packaging checklist

Before packaging:

• [ ] Public RFC files contain no workflow provenance.
• [ ] Evidence artifacts contain enough source references to audit public claims.
• [ ] Public and evidence manifests are separate.
• [ ] Scope, status, and exclusions are explicit.
• [ ] Every required implementer capability points to public sections, algorithms, or examples.
• [ ] Every public example has evidence-packet coverage.
• [ ] Every open question is excluded from public normative text or clearly scoped as future work.
• [ ] Any remaining human decision is captured in the escalation log or removed from the public dependency chain.
• [ ] The final packet states what is ready for promotion and what remains internal.