SCIENTIA impact, readership, and citation-adjacent signals

This document is the single research anchor for extending SCIENTIA beyond novelty / prior-art toward impact and audience success proxies (what people read, cite, and amplify). It complements:

• SCIENTIA publication automation SSOT (automation boundaries),
• Novelty ledger contracts under contracts/scientia/ (finding-candidate, novelty-evidence-bundle),
• Tunable parameter seed: contracts/scientia/impact-readership-projection.seed.v1.yaml.
• SCIENTIA multi-platform ranking, discovery, and anti-slop SSOT (research 2026) — social vs scholarly ranking surfaces, ingest vs syndicate, and operator KPI sketches complementary to impact projection.

Non-goals: Vox does not claim to predict future citations authoritatively. The feasible product is an inspectable, contract-weighted projection used for prioritization, routing, and operator transparency, never as a hard publish/deny gate without human review.

Why this is orthogonal to novelty

A finding can be novel but low resonance (narrow tooling note) or high resonance but weakly novel (clear survey of known ideas). Publication policy needs both lenses without conflating them.

External landscape (what already does this)

Solid, citable references for implementation seeds:

1. Bibliometric APIs (observed counts, not forecasts)

   • OpenAlex: open work metadata, citation counts, open citation graph facets—good for post-hoc and comparable-work baselines.
   • Crossref / DataCite: DOI-level metadata; Crossref’s separate Event Data mention stream is sunset 2026-04-23 (see multi-platform ranking research §4.12 / Crossref blog). Useful for discoverability and persistence more than prediction.
   • Semantic Scholar: citation counts; highly influential citation labeling uses ML over full-text citation contexts (useful conceptually; Vox may only see API summaries without full text).

2. Citation prediction (research systems, heavy ML)

   • ForeCite (arXiv:2505.08941): causal LM–style forecasting of future citation rates on large biomedical corpora—illustrates that title/abstract + time + field carry signal; training such a model is not a near-term in-repo deliverable.
   • HLM-Cite (2024): hybrid LM workflow emphasizing core vs peripheral citations—relevant if Vox later does structured claim–evidence graphs.
   • Graph vs text benchmarks (e.g. EMNLP 2024 finding papers): edge-based (citation graph) vs node-based (text) tradeoffs depend on data scale and horizon—Vox should default to transparent features, not a black-box score.

3. Readership and attention (altmetrics)

   • Altmetric Attention Score and Dimensions integrations (see vendor docs): weighted mention counts across news, policy, social, blogs, etc. Not the same as scientific quality; strong early visibility signal.
   • Literature on altmetrics vs early citations (e.g. studies on Mendeley readership and Twitter features): useful for defining feature families if Vox ever ingests licensed altmetric feeds—not assumed available by default.

4. Venue and genre
   Journal tier, open access, and subfield norms shift baseline citation rates. Any projection must carry field_baseline / venue_tier / topic metadata to avoid naive global thresholds.

What Vox can feasibly implement (phased seeds)

Ordered for honesty about data access and SSOT weighting (impact-readership-projection.seed.v1.yaml):

Critique of recent in-repo novelty automation work

This section does not replace code review; it records architectural debt to fix while expanding toward impact projection.

1. Heuristic constants in Rust
   Significance axes, confidence decomposition, and overlap-to-novelty mappings use numeric literals in vox-publisher helpers. That optimizes for a fast first slice but violates the Dynamics preference (parameters should move with policy). Remediation: load weights and bucket thresholds from contracts/scientia/impact-readership-projection.seed.v1.yaml (or a split scientia-discovery-heuristics.v1.yaml if impact vs discovery tuning diverges).

2. Prior-art ≠ impact
   The federated bundle answers overlap; it does not, by itself, answer who will care. Remediation: extend stdout / MCP payloads with a ComparableWorksSummary (or separate impact_projection object) so operators see both panels.

3. Calibration telemetry today
   Current calibration envelopes emphasize latency and overlap. Remediation: add optional fields (behind schema version bumps) for projected audience tier and data completeness (missing_fields: [...]) when phase A ships.

4. Single source of truth
   Novelty contracts live under contracts/scientia/*.schema.json. Impact projection should follow the same pattern: schemas for stored artifacts, YAML seeds for tunables, this doc for rationale—avoid scattering magic numbers across scientia_discovery.rs and scientia_finding_ledger.rs long term.

SSOT maintenance rules

• New numeric policy for impact/readership → update the seed YAML + one line in this doc’s changelog (below).
• New external signal family → add to seed signal_families + document license/opt-in here.
• Shipped JSON shape → add or extend a JSON Schema under contracts/scientia/ and register in contracts/index.yaml.

Changelog