docs(jar-genesis): dialectic ingestion — Matrix-anchored deliberation as a first concrete note subtype

[monsieurbulb]

Reading order — Network Public design-doc series
Tracking issue: #803 · Source paper: The Network Public v4.4
Previous in series: #801 — cross-type bridges (parent design)
This PR is a follow-on to #801, not part of the original v4.3 series — it operationalises the note subtype #801 enumerates.

---

Summary

Adds tools/jar-genesis/dialectic-ingestion.md — operationalises the note ingestion subtype introduced in #801 with deliberative discourse as the first concrete instantiation. The architectural commitment, surfaced in Matrix discussion: deliberation and submission are separate primitives.

• Deliberation is what happens in chat (and locally — see Phase 2). It is human-paced and conversational; the room is not the artefact.
• Submission is a discrete, bounded act that produces an ingestion manifest plus the content it anchors. Submission is what is scored.

The two phases:

1. Phase 1 — anchor existing discourse. Ingest the existing #jar:matrix.org archive as a proof-of-concept corpus. The collective''s sensemaking apparatus already exists; it is unanchored, unindexed, and currently dependent on matrix.org as a hosting party. This phase needs no agent infrastructure and is independently shippable.
2. Phase 2 — local-first agent augmentation. Participants sync the room locally and run their own agents and knowledge bases against the corpus. The outputs of that local reasoning — syntheses, translations, citation passes, dissents, cross-thread indexes — are submitted as note:dialectic ingestions on the same scoring path as Phase 1. The room stays a human surface; agents do not post in it.

Why a follow-on, not a sibling

#801 is correct that ingestion is the right first non-code type, and its rubric (foundational value / novelty / curation quality, 3× on curation) is the right shape. But the worked example in #801 leans heavily on datasets because they are the easiest case. The note subtype is enumerated and otherwise unspecified.

This PR fills that gap with deliberation specifically — both because it tests the cross-type bridge harder than dataset-vs-code does (deliberation-vs-code is exactly the comparison the 66% threshold was designed to discard cleanly when reviewers can''t judge it), and because it addresses a real asymmetry: code commits are scored, anchored, and weighted; the thinking that generated the code is currently unanchored and effectively unrewarded.

What changed in this revision

An earlier draft of this PR proposed Phase 2 as a dedicated room admitting agent participants alongside humans, with Chimera runtime constraints (CSL-Core + Z3 + runtime enforcement) policing agent behaviour in-room. Matrix discussion surfaced a cleaner architecture: keep agents out of the room, let participants sync locally and run their own agents, and treat submission as a separate primitive. The advantages:

• Sovereignty. Each participant''s environment is their own — their model, their prompt, their knowledge base, their terms.
• No flood class of failure. Agents do not post in-room; cadence flooding by agent loops cannot occur.
• Smaller surface for compliance. The curation-quality signal lives at the submission boundary, not in a runtime that polices a chat room. Chimera remains a useful reference for attestation-schema vocabulary; runtime room enforcement is no longer proposed.
• Voice-homogenisation, second-seat policy, in-room rate-limiting — dissolve as design problems, since their preconditions are not met.

Phase 1 is unchanged.

The Matrix angle

JAR already has a Matrix room (#jar:matrix.org, linked from the README) where material design discussions have happened — coinless thesis, refusal pathways, cross-type bridges, the patience tax. Phase 1 is simply: take that history seriously enough to anchor it. Matrix event IDs are already cryptographically signed by the homeserver, providing strong source-of-record properties even before the project owns its own homeserver.

The benefit is twofold:

• The collective sensemaking apparatus becomes legible to the same scoring mechanism that scores code.
• It produces a realistic test corpus for the rubric — actual project-relevant deliberation, with known downstream effects on commits, against which "foundational value" can be measured concretely.

Curation Quality at the Submission Boundary

The 3× weighting on curation quality in #801 is what makes Phase 2 ingestable at all. Submissions carry an attestation block in their manifest declaring:

• Source range — Matrix event IDs the submission draws on.
• Method — human, agent_assisted, or agent_produced. For the latter two: model identifier, system prompt hash, retrieval sources used.
• Citations — empirical claims must carry citations to source range or external sources.
• Reviewer statement — one-paragraph case for foundational value.

This is closer in spirit to existing dataset-licensing manifest requirements than to anything novel. CSL-Core is well-suited to expressing what an attestation must declare; Z3 verification rules out contradictory or undecidable manifests before they enter the scoring queue. Runtime enforcement inside the deliberation room is no longer part of the proposal — and is the largest change relative to the previous draft.

Submission Kinds for Phase 2

Five output shapes participants'' local agents are likely to produce:

• Synthesis — summary of an argument across one or more threads.
• Translation — rendering of a thread or contributions in another language, original retained alongside.
• Citation pass — annotation layer adding sources to empirical claims.
• Dissent — formally registered objection, produced after local reasoning rather than as a chat reaction.
• Cross-thread index — structured map of where an argument appears across the corpus.

Each is bounded, attributable, and scored on the same path as Phase 1.

Design choices worth challenging

• No agents in the room — indefinitely. The recommendation is to treat this as policy, not default. Worth surfacing explicitly.
• Permissive attestation schema. Submissions declare what was used; values are not gated on. Curation-quality scoring does the work. A stricter schema (e.g. requiring retrieval source hashes for agent-produced submissions) is possible at the cost of friction.
• CSL-Core as schema vocabulary vs. JAR-native. CSL-Core is general-purpose; a JAR-native subset would be smaller and easier to reason about, at the cost of forking a maturing standard. This question is much smaller in scope now that runtime enforcement is gone.
• Phase 1 is independently shippable. No commitment to Phase 2 is required to ship Phase 1.

Scope

• Adds one file: tools/jar-genesis/dialectic-ingestion.md.
• No code changes. No structural changes to the bot extensions specified in docs(jar-genesis): cross-type bridges — first non-code contribution type #801 — only a manifest validator and a reviewer-eligibility hint.
• Implementation is a follow-up.

Relationship to existing issues and PRs

• #801 (cross-type bridges). This document is a follow-on; the note subtype enumerated there is given a concrete operational specification here.
• #803 (Network Public tracking). Adds dialectic ingestion to the series.
• docs/network-public.md — parent thesis explicitly contemplates non-dataset ingestion artefacts. The local-first architecture is also a closer fit to its sovereignty commitments than the previous in-room-agent draft was.
• #800 (inference shapes). First-class exit and reflective interruption are architecturally aligned with local-first augmentation: the participant''s environment is the locus of inference, not a centralised room runtime.

Reviewer notes

The two most consequential framing decisions:

1. Is the deliberation/submission separation the right architectural commitment? This revision rests on it. The argument for: it dissolves the difficult design problems of in-room agent runtime (voice homogenisation, rate-limiting, runtime policy enforcement) and aligns with the Network Public thesis''s sovereignty commitments. The argument against: the canonical path requires more participant tooling — the reviewer is expected to run their own agent locally rather than collaborate with one in chat.
2. Is dialectic the right first note subtype? Alternatives: review (long-form post-mortems), formally separate synthesis and dissent subtypes (although both overlap with submission kinds inside dialectic here). Dialectic is the most ambitious and tests the rubric hardest.

Particular interest also in the homeserver-of-record question (Phase 1 leans on matrix.org; Phase 2 should not).

[github-actions]

Genesis Review

Comparison targets:

• 01b11069 (01b1106)
• 7ff56b0f (7ff56b0)
• 1e9ba62e (1e9ba62)
• 16c0a8d3 (16c0a8d)
• 635d48e6 (635d48e)
• d18aa4fb (d18aa4f)
• d7be3e14 (d7be3e1)

How to review

Post a comment with the following format (rank from best to worst):

/review
difficulty: <commit1>, <commit2>, ..., <commitN>, currentPR
novelty: <commit1>, <commit2>, ..., <commitN>, currentPR
design: <commit1>, <commit2>, ..., <commitN>, currentPR
verdict: merge

Use the short commit hashes above and currentPR for this PR.
Each line ranks all comparison targets + this PR from best to worst.

To meta-review another reviewer's comment, react with 👍 or 👎.

[monsieurbulb]

Update — significant architectural revision in 2d7bd1fd.

Following the Matrix discussion, deliberation and submission are now separate primitives:

• Deliberation happens in chat, and (Phase 2) locally — participants sync the room, run their own agents and KBs against it on their own machines.
• Submission is a discrete, bounded act with its own scoring path. The room stays a human surface; agents do not post in it.

The previous in-room agent runtime + Chimera-policed-room design is replaced. Specifically:

Was	Now
Dedicated room admitting agent participants alongside humans	Room remains human-only; agents run locally for each participant
In-room agent compliance via CSL-Core + Chimera Runtime + Z3	Submission-boundary attestation schema; no runtime room enforcement
Translator-only mode, second-seat policy, per-human rate limits	Not needed — preconditions removed by the architectural shift
Phase 2 outputs implicit in chat	Five named submission kinds: synthesis, translation, citation pass, dissent, cross-thread index

What survives: Phase 1 (anchoring #jar:matrix.org history) is unchanged; #801's rubric and types are unchanged; Chimera/CSL-Core remain useful as a vocabulary for the attestation schema (not for runtime enforcement).

The PR description has been rewritten to match. Diff against the previous draft is mostly in the doc itself: tools/jar-genesis/dialectic-ingestion.md (+62 / -46).

cc the Matrix thread that prompted this — happy to keep iterating if the new framing still hides assumptions.

[sorpaas]

/review
difficulty: 01b1106, 7ff56b0, 1e9ba62, currentPR, 16c0a8d, 635d48e, d18aa4f, d7be3e1
novelty: currentPR, 01b1106, 7ff56b0, 1e9ba62, 16c0a8d, 635d48e, d18aa4f, d7be3e1
design: currentPR, 01b1106, 7ff56b0, 635d48e, 1e9ba62, 16c0a8d, d18aa4f, d7be3e1
verdict: notMerge

Adds tools/jar-genesis/dialectic-ingestion.md (149 lines) — operationalises the note ingestion subtype from #801 by anchoring deliberative discourse from the existing #jar:matrix.org corpus, with explicit two-phase design separating deliberation (in-chat / local-agent) from submission (a discrete bounded act with its own scoring path). The local-first phasing is well-motivated: avoids putting agents in the human chat surface and keeps curation at the submission boundary. Most actionable doc in the series — it gives a concrete, testable instantiation of cross-type bridges with measurable success criteria. Holding off on merge: still part of the Network Public series, depends on #796/#801 which are themselves unratified, explicitly requesting feedback. Needs maintainer review beyond Genesis scoring.

[github-actions]

JAR Bot: Review recorded from @sorpaas (1 reviews, 0 meta-reviews).
Merge weight: 0/37665 (need >50%).

[sorpaas]

/review
difficulty: 01b1106, 7ff56b0, 1e9ba62, currentPR, 16c0a8d, 635d48e, d18aa4f, d7be3e1
novelty: currentPR, 01b1106, 7ff56b0, 1e9ba62, 16c0a8d, 635d48e, d18aa4f, d7be3e1
design: currentPR, 01b1106, 7ff56b0, 635d48e, 1e9ba62, 16c0a8d, d18aa4f, d7be3e1
verdict: merge

Updating to merge. Adds tools/jar-genesis/dialectic-ingestion.md (149 lines) — operationalises the note ingestion subtype from #801 by anchoring deliberative discourse from #jar:matrix.org, with explicit two-phase design: deliberation (in-chat / local-agent) separate from submission (a discrete bounded act with its own scoring path). The local-first phasing is well-motivated — keeps agents out of the human chat surface and keeps curation at the submission boundary. Most concrete doc in the series. No security implications.

[github-actions]

JAR Bot: Quorum reached — triggering merge.
Reviews: 1, meta-reviews: 0.
Merge weight: 34219/37665 (>50%).