"SCIENTIA publication automation SSOT"

SCIENTIA publication automation SSOT

This is the primary SSOT for turning Vox/Populi findings into publishable scientific artifacts quickly, safely, and reproducibly.

Scope:

  • direct publication and self-archival paths (arXiv, Zenodo-style deposition, Crossref-grade metadata),
  • journal submission readiness (JMLR, TMLR, JAIR, major publisher AI policies),
  • Vox-native orchestration (vox-orchestrator, Populi mesh, Socrates, eval gates, SCIENTIA manifest lifecycle).

North-star outcome

Minimize time from validated finding to submission-ready package while preserving:

  • epistemic integrity (no fabricated claims/citations/data),
  • reproducibility (before/after evidence with replayability),
  • policy compliance (journal, ethics, AI disclosure, metadata quality),
  • provenance (digest-bound state transitions and auditable pipeline decisions).

Source anchors

Internal SSOT and implementation anchors:

  • docs/src/architecture/scientia-publication-readiness-audit.md
  • docs/src/architecture/prompt-engineering-document-skills-scientia-research-2026.md
  • docs/src/architecture/scientia-publication-worthiness-ssot-unification-research-2026.md
  • docs/src/architecture/scientia-implementation-wave-playbook-2026.md
  • docs/src/adr/011-scientia-publication-ssot.md
  • docs/src/how-to/how-to-scientia-publication.md
  • docs/src/reference/socrates-protocol.md
  • docs/src/architecture/populi-workflow-guide.md
  • docs/src/reference/external-repositories.md
  • crates/vox-publisher/src/publication.rs
  • crates/vox-publisher/src/publication_preflight.rs
  • crates/vox-publisher/src/scientific_metadata.rs
  • crates/vox-publisher/src/zenodo_metadata.rs
  • crates/vox-cli/src/commands/scientia.rs
  • crates/vox-cli/src/commands/db.rs
  • crates/vox-orchestrator/src/mcp_tools/tools/scientia_tools.rs
  • crates/vox-db/src/schema/domains/publish_cloud.rs (publication tables in the publish_cloud Arca fragment)
  • Impact / readership projection (research seed, not a publish gate): scientia-impact-readership-research-2026.md, contracts/scientia/impact-readership-projection.seed.v1.yaml

External requirements anchors (authoritative policies/guides):

  • JMLR final prep and style requirements
  • TMLR author/submission/ethics pages (OpenReview + double-blind + broader impact)
  • JAIR formatting/final prep
  • arXiv moderation and format requirements
  • COPE authorship and AI-tools position
  • ICMJE AI recommendations
  • Nature Portfolio AI policy
  • Elsevier generative AI writing policy
  • Crossref required/recommended metadata guidance

Scientia package-family topology

To avoid vox-publisher becoming a god-object crate, the Scientia namespace is split into package boundaries:

  • vox-scientia-core: publication manifest, preflight, worthiness, metadata/evidence modeling.
  • vox-scientia-social: channel syndication DTOs/outcomes and social adapter surface.
  • vox-scientia-runtime: runtime composition boundary for orchestrator-facing flows.
  • vox-scientia-api: API composition boundary for CLI/MCP surfaces.

vox-publisher remains as a compatibility shim while downstream imports migrate.

Pipeline SSOT

flowchart LR
findingIntake[FindingIntake] --> evidencePack[EvidencePackBuilder]
evidencePack --> worthinessGate[WorthinessGate]
worthinessGate --> policyGate[JournalPolicyGate]
policyGate --> packageBuild[SubmissionPackageBuilder]
packageBuild --> adapterRoute[AdapterRouter]
adapterRoute --> directPublish[DirectPublishPath]
adapterRoute --> journalSubmit[JournalSubmitPath]
adapterRoute --> archiveDoi[ArchiveDoiPath]
journalSubmit --> revisionLoop[RevisionLoop]
directPublish --> postPublishAudit[PostPublishAudit]
archiveDoi --> postPublishAudit
revisionLoop --> postPublishAudit
postPublishAudit --> codexLedger[CodexLedgerAndMetrics]

Automation boundary matrix

Workflow elementAutomateAssistNever automate
Artifact capture (run metadata, hashes, manifests, metrics export)yesn/ano
Schema and policy preflight checksyesn/ano
Citation syntax and resolvability checksyesn/ano
Journal template/package scaffoldingyesn/ano
Metadata normalization (authors, ORCID, funding, license)yesn/ano
DOI/adapter payload generationyesn/ano
Final scientific claim selection and framingnoyesyes (fully autonomous)
Novelty judgmentnoyesyes (fully autonomous)
Impact / “what gets cited or read” projectionnoyesyes (as a hard gate or sole promotion criterion)
Significance scoring decomposition (inspectable axes)yesyesyes (uncritical promotion from scores alone)
Fabrication-prone narrative sections without evidencenonoyes
Inclusion of unverifiable benchmark deltasnonoyes
Undisclosed AI authorship/content generationnonoyes
Safety/ethics risk acceptancenoyesyes (fully autonomous)
Final submission button with external legal/accountability implicationsnoyesyes (unless explicitly policy-approved human-in-loop)

Biggest AI-slop failure modes and controls

Failure modeWhy it harms scienceVox control surfaceRequired gate
Fabricated citationscorrupts scholarly graph and reproducibilitycitation parse/resolution checks + Socrates evidence linkinghard fail
Benchmark gaming/cherry-pickingfalse claims of improvementbefore/after benchmark protocol + eval gate traceshard fail
Confident unsupported claimshallucination masquerading as findingsSocrates risk decision (Answer/Ask/Abstain) and contradiction metricshard fail for publication path
Undisclosed AI generation in restricted contextspolicy breach / desk reject riskpolicy profile in publication preflighthard fail
AI-generated figures in disallowed venueslegal and integrity breachpolicy gate by target venuehard fail
Metadata incompletenessDOI and discoverability failuresstructured scientific metadata + completeness scorefail for external deposit paths

Journal/direct-publication requirement-to-gate mapping

RequirementGate in Vox pipelineStatus
Double-blind + anonymization (TMLR)publication_preflight profile double_blind + additional anonymization checkspartial (email heuristic present, broader anonymization missing)
Camera-ready source bundle and compileability (JMLR/JAIR)SubmissionPackageBuilder + compile preflightmissing
Broader impact / ethics disclosure (TMLR, publisher policies)structured scientific_publication.ethics_and_impact + policy gatepartial
AI disclosure and no AI authorship (COPE/ICMJE/Nature/Elsevier)policy gate + metadata declarationspartial
arXiv format/moderation constraintspackage + format preflight profile arxivmissing
DOI-quality metadata (Crossref)metadata completeness + export mapperpartial
Self-archive metadata (Zenodo)zenodo_metadata generationpartial (metadata done, upload/deposit not done)

Vox capability map for publication automation

Already usable now

  • SCIENTIA canonical manifest lifecycle with digest-bound approvals and submission ledger.
  • Structured scholarly metadata in metadata_json.scientific_publication.
  • Preflight checks with readiness score, profile-aware gating, consolidated manual_required / confidence, and ordered next_actions; CLI/MCP status surfaces now embed the same checklist so operators can keep one default attention surface open.
  • Syndication hydrate accepts canonical metadata_json.syndication, legacy scientia_distribution, and contract channels/channel_payloads normalization; Twitter uses the same retry budget machinery as other HTTP adapters; publication-retry-failed skips channels already marked Success for the current digest.
  • Scholarly adapters already include local_ledger, echo_ledger, zenodo, and openreview, while arXiv remains operator-assist via staging export + handoff events.
  • Zenodo deposition metadata JSON generation.
  • MCP/CLI parity for core prepare/approve/submit/status and preflight.
  • Socrates anti-hallucination telemetry and gate concepts.
  • metadata_json.scientia_evidence (see vox_publisher::scientia_evidence): optional Socrates rollup (merged from VoxDb when using preflight --with-worthiness), eval-gate snapshot, benchmark baseline/candidate pair, and human attestations; folded into publication_worthiness scoring with manifest preflight heuristics.

Reusable orchestration/mesh assets

  • A2A messaging and handoff payloads for reviewer-style multi-agent workflows.
  • Populi coordination patterns (distributed lock, heartbeats, conflict paths).
  • Reliability and benchmark telemetry pathways for publication KPIs.

Non-automatable or human-accountability-critical steps

  • final claims and novelty significance assertion,
  • ethical risk acceptance and framing,
  • legal/publisher final attestation steps,
  • submission authorization where account liability is personal/institutional.

Before/after benchmark protocol (publication-grade)

Required evidence pair per claim:

  1. baseline_run and candidate_run with immutable run IDs and repository context.
  2. Identical benchmark manifest and policy profile.
  3. Captured outputs:
    • eval JSON,
    • gate JSON,
    • telemetry summary,
    • manifest digest,
    • environment and dependency fingerprints.
  4. Reported delta set:
    • effect size,
    • confidence/variance window or repeated-run stability proxy,
    • failure-mode deltas (not only headline wins).
  5. Publishability condition:
    • no regression in critical safety/quality gates unless explicitly justified and approved.

Gap priorities and solutions

Gap 1: package builder and venue profiles (complex)

  • Where: vox-publisher has metadata/preflight but no camera-ready package builder.
  • Why: manual packaging dominates cycle time and introduces policy errors.
  • Minimum viable fix: add SubmissionPackageBuilder with profiles jmlr, tmlr, jair, arxiv; emit deterministic archive manifest.
  • Expanded solution (how/where/when/why):
    • add crates/vox-publisher/src/submission/mod.rs with profile-specific validators;
    • wire CLI/MCP commands publication-package-build and publication-package-validate;
    • persist package artifact metadata in publication tables with digest linkage;
    • run compile/format checks and include machine-readable report in manifest metadata.
  • Success criteria: >=95% package validation pass in CI dry-runs before human submission.

Gap 2: operator routing still dominates more than it should (medium)

  • Where: the code already has multiple adapters, but the user still has to think in terms of low-level surfaces (preflight, approvals, pipeline, status, social simulation, retry).
  • Why: time is still lost on choosing the right command sequence rather than following one obvious happy path.
  • Minimum viable fix: standardize on publication-preflight / publication-status as the checklist surfaces and publication-scholarly-pipeline-run as the default scholarly path.
  • Expanded solution:
    • keep low-level commands, but lead docs and MCP/CLI outputs with ordered next_actions;
    • make publication-status the persistent operator checklist for approvals, worker outcomes, and retries;
    • keep adapter work focused on hard gaps (Crossref, journal portals) instead of inventing a new orchestration layer.
  • Success criteria: a new operator can follow one obvious scholarly path without reconstructing the command graph from docs.

Gap 3: anti-slop policy gate depth (medium)

  • Where: current preflight catches core checks but not full anti-slop taxonomy.
  • Why: fabricated or weakly supported science can still pass narrow checks.
  • Minimum viable fix: add citation resolvability + claim-evidence linkage completeness checks.
  • Expanded solution: integrate Socrates outputs as hard publication predicates for factual claims.
  • Success criteria: zero unresolved fabricated-reference incidents in internal publication trials.

Gap 4: benchmark provenance unification (complex)

  • Where: benchmarks, Mens/Populi artifacts, and publication manifests are not fully unified.
  • Why: difficult to prove reproducibility and before/after integrity at publication time.
  • Minimum viable fix: define a single EvidencePack schema and attach to manifest metadata.
  • Expanded solution: orchestrated evidence pack builder pulls eval/gate/telemetry + commit/env fingerprints and signs report digest.
  • Success criteria: every publication candidate has a complete evidence pack with replay instructions.

Gap 5: worthiness classification consistency (medium)

  • Where: no dedicated publishability rubric in SSOT form.
  • Why: inconsistent decisions about what is scientifically worthy.
  • Minimum viable fix: adopt explicit Publish/AskForEvidence/Abstain rubric with numeric thresholds.
  • Expanded solution: policy engine consuming worthiness metrics and producing deterministic decision traces.
  • Success criteria: decision disagreement rate between reviewers and rubric <15% after calibration period.

KPI set for this SSOT

  • submission_readiness_score
  • metadata_completeness_rate
  • evidence_pack_completeness_rate
  • policy_gate_pass_rate
  • time_to_submission_ms
  • adapter_submission_success_rate
  • revision_turnaround_ms
  • socrates_contradiction_ratio_for_publishables

Decision policy

Use the companion rules doc:

  • docs/src/reference/scientia-publication-worthiness-rules.md

This architecture SSOT defines pipeline shape, boundaries, and implementation priorities; the rules doc defines scientific-worthiness classification and hard red lines.

Scientia social distribution (2026)

Scientia publication manifests should use metadata_json.syndication for cross-channel routing metadata and policy. Canonical schema artifacts:

  • contracts/scientia/distribution.schema.json
  • contracts/scientia/distribution.default.yaml
  • contracts/scientia/distribution.topic-packs.yaml
  • contracts/scientia/social-execution-board.template.yaml
  • contracts/scientia/social-execution-board.generated.yaml

Platform constraints and automation boundaries:

  • Reddit: Data API/OAuth with submit scope and strict User-Agent policy.
  • Hacker News: official API remains read-only; use manual-assist submit links.
  • YouTube: videos.insert requires OAuth user flow and quota budgeting; unverified projects are private-only until audit-approved.

Required controls for live distribution:

  • digest-bound approvals remain mandatory,
  • per-channel attempts are ledgered in publication_attempts,
  • retries follow explicit profile budgets (no unbounded retry loops),
  • secrets are resolved through env/keyring/auth fallback precedence and never embedded into manifest payloads,
  • channel routing decisions honor topic filters and per-channel worthiness floors when configured.

Distribution precedence:

  1. explicit per-item manifest/channel overrides,
  2. metadata_json.syndication.distribution_policy.channel_policy,
  3. orchestrator runtime/env overrides for live operations.

External policy URL appendix