SCIENTIA multi-platform ranking, discovery, and anti-slop SSOT

This document synthesizes how major distribution surfaces rank and filter content, maps that landscape to Vox Scientia (outbound publication and planned inbound discovery), and proposes a single maintainable policy layer (manifest-centered metadata + contracts) so operators can add or subtract channels with minimal code churn.

Naming note: Internal references to “Vox Chianti” in planning conversations map to Vox Scientia for this repository.

See also

• Vox Scientia external discovery and monitoring (research) — inbound feeds, Socrates triage, hybrid dedup, digest agents.
• SCIENTIA impact, readership, and citation-adjacent signals — bibliometric and attention proxies; assist-only posture.
• SCIENTIA publication-worthiness and SSOT unification — standards-to-signals matrix, canonical metadata graph, automation boundary ledger.
• Vox RAG and autonomous research architecture — retrieval zones, corpora, Socrates gate (production SSOT for RAG).
• Tunable impact projection seed: contracts/scientia/impact-readership-projection.seed.v1.yaml.

1. Executive summary

Scientia faces a deliberate tension:

1. Anti-slop / “do not waste the reader” — limit what is promoted to humans and to the public internet so every outbound unit carries evidence, correct routing, and respect for community norms.
2. High-recall discovery — accept that the world produces more data than any team can read; the fix is sorting, deduplication, and provenance, not artificial scarcity of ingest.

Resolution (architecture): separate ingest volume from syndication volume. Ingest broadly into quarantine-capable stores and deduplicated indices; compile outbound posts and venue submissions from a canonical manifest graph with per-channel projection profiles (templates + policy + optional impact hints). Numeric tuning belongs in contracts/scientia/*.yaml and JSON Schemas where stored artifacts are versioned—not scattered as unexplained literals in Rust.

2. Information sufficiency and citation tiers

Public writing on “algorithms” mixes verifiable sources with marketing. This document uses explicit tiers:

Critical assessment: Tier A is sufficient to justify structural Scientia decisions (e.g. “Meta uses multiple rankers per surface,” “Scholar indexes PDFs with heuristic headers,” “arXiv moderates for scholarly standards”). Tier C dominates many web searches; any specific percentage (e.g. “CTR is 20% of YouTube rank”) should be treated as unverified unless traced to Tier A.

What we do not have without product-specific telemetry: per-tenant lift curves, per-channel A/B behavior, or legal/commercial constraints for each API. Those require operator data and counsel—not additional web search volume.

3. Platform clusters: signals, risks, Scientia posture

Posture legend: Ingest = pull into monitoring/quarantine/RAG; Syndicate = outbound post or venue handoff; Assist = human-in-the-loop or scoring only; Avoid = default off without explicit policy.

4. Deep research by distribution surface (expanded 2026 wave)

This section expands the summary table with first-party wording where available, then narrower technical or academic sources, then explicitly marks speculative creator-industry claims. Length is intentional: Scientia automation must respect materially different objective functions per surface.

4.1 Reddit (first-party: Home feed pipeline)

Tier A — Reddit Help (“Reddit’s Approach to Content Recommendations”): Reddit states that the logged-in Home feed mixes subscriptions with recommendations, and that personalized ordering uses:

• Content-related information: upvotes/downvotes, community, comment history, post type, age, flairs.
• Your activity: engagement history, time in communities, recent visits, subscriptions, onboarding topic interests, “show less” feedback.
• Account age: newer accounts may see more recommendations relative to subscriptions.
• Location setting: country preference.

Reddit describes a four-step pipeline: (1) candidate generation, (2) filtering (spam, seen-before, blocked), (3) predictive models for preference, (4) sort with diversity (“avoid too many similar posts in a row”). Logged-out Popular is described as showcasing popular recent posts by net upvotes, sometimes location-customized.

Implications for Scientia: “Hot” vs “New” vs “Top” remain user-controlled sorts inside a community; automated syndication must still defer to subreddit rules and moderator norms (not Reddit’s global ML). Inbound monitoring should treat vote/comment velocity as weak evidence of technical novelty—high votes correlate with entertainment or controversy.

4.2 YouTube (first-party: signals and responsibility)

Tier A — YouTube Blog (Goodrow, 2021): YouTube emphasizes that recommendations (homepage + Up Next) drive more viewership than subscriptions or search. The system learns from “signals” including clicks, watch time, survey responses, sharing, likes, and dislikes, with explicit narrative that click ≠ satisfaction (watch time added in 2012; valued watch time via surveys; models predict satisfaction for unrated views). For news and information, YouTube discusses authoritative vs borderline classification using human evaluators and public rater guidelines, with borderline demoted.

Official help pages (Google) complement this with consumer-facing descriptions of personalization and controls; treat help URLs as Tier A for product behavior, not for numeric rank weights.

Implications for Scientia: optimize for clarity of promise in title/thumbnail, early retention, and evidence-forward framing for technical talks. Do not treat Shorts and long-form as one projection profile.

4.3 Meta: Facebook and Instagram (first-party: transparency center + system cards)

Tier A — Meta Transparency Center: Meta documents separate ranking systems per surface (e.g. Instagram Feed, Instagram Explore, Instagram Search, Facebook Feed). Common pattern in the cards: gather inventory → integrity filtering → predictions → ranking → diversity / freshness controls. Explore documentation describes staged retrieval and ranking at high candidate counts; Search mixes multiple entity types (hashtags, audio, Reels, profiles).

Implications for Scientia: any “post to Instagram” automation must declare which surface the copy targets; Reels-first video vs static Feed post vs carousel document are different distribution contracts.

4.4 X (Twitter): open archive vs current stack

Tier A (historical): Twitter released recommendation source as twitter/the-algorithm (candidate generation, ranking, mixer concepts documented in repo and accompanying commentary).

Tier B / moving target: Post-rebrand X, independent reporting and third-party repos (e.g. xai-org/x-algorithm documentation mirrors) discuss newer ML ranking stacks. Treat these as engineering curiosity, not stability contracts, unless pinned by your legal/compliance review of current Terms and API fields.

Implications for Scientia: prefer single canonical URL threads; avoid duplicating long manifest text across tweets (fragmentation + edit drift).

4.5 LinkedIn (first-party engineering blog)

Tier A — LinkedIn Engineering: LinkedIn has published multiple articles on dwell time, feed funnel architecture, and retrieval/ranking passes (e.g. posts on dwell time and “next generation” feed engineering). These establish semantic retrieval + multi-pass ranking as the mainstream architecture for large professional graphs.

Implications for Scientia: long-form research updates should be written as native posts with structured headings; bare link drops underperform and read as spam to both humans and rankers.

4.6 TikTok (first-party transparency)

Tier A — TikTok Transparency / Newsroom: TikTok’s public pages describe For You personalization using user interactions (likes, shares, follows, watch length, completions), video information (captions, sounds, hashtags), and device/account settings (language, country, device) at lower weight. They explicitly note some non-factors in their public FAQ (e.g. follower count not directly used as a recommendation input in the way many creators assume).

Implications for Scientia: short video is a different production and integrity surface; default Avoid unless you operate a vertical video pipeline with separate moderation.

4.7 Hacker News (first-party FAQ + open ranking folklore)

Tier A — Hacker News FAQ: ranking is not “higher karma users rank higher”; flags, vouching, software penalties, and moderation exist alongside a gravity curve over votes and time.

Tier B — Long-standing reverse engineering posts (e.g. classic “How HN ranking works” articles) remain useful for intuition but should not override the FAQ for product decisions.

Implications for Scientia: keep ManualAssist as the default posture; treat HN as a high-context, low-forgiveness channel.

4.8 Google Scholar (first-party inclusion guidelines)

Tier A — Scholar inclusion documentation: Scholar indexes scholarly works meeting PDF and bibliographic header heuristics; inappropriate genres (news, editorials) are out of scope. Ranking inside Scholar is not fully specified publicly at the same granularity as consumer social feeds; expect relevance + citation + venue signals at a high level.

Implications for Scientia: invest in clean PDFs, structured metadata, and persistent DOIs rather than keyword stuffing.

4.9 PubMed and NLM retrieval (peer-reviewed + official help)

Tier A/B — PubMed “Best Match”: NLM has published peer-reviewed and technical bulletin material describing a two-stage pipeline (retrieval + learning-to-rank rerank) for relevance sorting. This is the canonical pattern for scientific text retrieval at national-library scale.

Implications for Scientia: for biomedical topics, PubMed complements OpenAlex; unify DOI/PMCID in the manifest graph to avoid duplicate cards.

4.10 Semantic Scholar (AI2) graph and recommendations API

Tier A — Semantic Scholar API docs: AI2 documents graph endpoints, fields (including citation and “influential citation” concepts in API summaries), and a Recommendations API for “papers like this” / list-based positives and negatives.

Implications for Scientia: ideal for assist-only expansion of prior-art packets—never a publish gate by itself.

4.11 OpenAlex, ORCID, and persistent identity

Tier A — OpenAlex documentation: CC0 graph, works/institutions/topics, citation facets, filters, and (as of documentation evolution) semantic search beta—verify current capabilities in docs before locking contracts.

Tier A — ORCID trust and visibility: ORCID explains visibility levels (Everyone / Trusted parties / Only me) and trust markers from member organizations vs self-assertion.

Implications for Scientia: ORCID and ROR-style affiliations belong in the canonical contributor graph, not retyped per social post.

4.12 Crossref Event Data sunset and replacement (critical for “attention” plans)

Tier A — Crossref blog (March 24, 2026): Crossref will sunset the Event Data API on April 23, 2026 (historical access on request). Rationale: shift toward integrity and structured relationships; low usage. Replacement emphasis: a data citations API endpoint surfacing dataset links from member metadata (beta; feedback solicited).

Implications for Scientia: any roadmap item that assumed Crossref Event Data as a live web-mention firehose must be rewritten. Attention/altmetrics-style monitoring should plan around surviving licensed vendors, first-party platform analytics, or curated feeds—not deprecated Crossref Event streams.

4.13 Bluesky and composable feeds (protocol + first-party blog)

Tier A — Bluesky blog on custom feeds: Bluesky describes algorithmic choice via third-party/custom feeds rather than a single opaque ranker.

Tier B — Ecosystem tooling: community frameworks (e.g. SkyFeed / feed builders) show how declarative rules can combine engagement, graph filters, and ML similarity—useful as patterns for Scientia inbound selectors, not as dependencies.

Implications for Scientia: subscribing to a small allowlisted set of expert feeds can beat generic firehoses for ML research surfacing.

4.14 Mastodon and the fediverse (open source + docs)

Tier A — Mastodon docs (trends APIs) and server source: trending surfaces exist with documented endpoints; implementation details (e.g. reblog/favorite scoring, decay) live in server code paths discussed publicly.

Implications for Scientia: useful for open-community announcements; not a substitute for arXiv/DOI persistence.

4.15 Discord discovery (first-party support + developer docs)

Tier A — Discord Support / Developers: Discovery is governed by eligibility, community health, and directory/search UX—not a global “For You” optimized for off-platform URLs.

Implications for Scientia: keep research artifacts on DOI/repo surfaces; use Discord only as optional community mirror with human moderators.

4.16 EU Digital Services Act and researcher access (regulatory Tier A/B)

Tier A — Primary law and EU Commission materials: the DSA imposes transparency, risk, and researcher-facing obligations on Very Large Online Platforms and Very Large Online Search Engines (thresholds defined in the regulation). Practical researcher access flows are being operationalized via Commission-level FAQ pages (e.g. algorithmic transparency centre FAQs).

Tier B — Legal commentary: law firms and NGOs summarize Articles on recommender transparency, non-profiling feeds, and ads repositories—useful for checklists, not for implementation literals.

Implications for Scientia: when syndicating to VLOPs, expect disclosure strings, opt-outs, and audit logs to become part of the distribution projection metadata—not optional marketing footers.

4.17 Information quality and “slop” (research framing, not platform docs)

Independent of any one ranker, scientometrics and HCI literature (not exhaustively cited here) consistently warns that engagement maximization ≠ epistemic quality. Scientia’s existing direction—Socrates triage, inbound preflight, quarantine—aligns with treating engagement as a diagnostic, not a truth label.

5. End-to-end flow (canonical SSOT → channels → inbound)

flowchart TB
  subgraph canonical [Canonical_SSOT]
    Manifest[Publication_manifest_and_metadata_graph]
    Contracts[contracts_scientia_schemas_and_YAML_seeds]
  end
  subgraph outbound [Outbound_compile]
    Publisher[vox_publisher_syndication]
    Channels[Twitter_Reddit_HN_YouTube_RSS_Forge]
  end
  subgraph inbound [Inbound_discovery_planned]
    Feeds[RSS_Atom_feed parsers]
    SocialRead[Read_only_social_APIs]
    Search[vox_search_SearXNG_and_hybrid_memory]
    Gates[Socrates_preflight_quarantine]
  end
  Manifest --> Publisher
  Contracts --> Manifest
  Publisher --> Channels
  Feeds --> Gates
  SocialRead --> Gates
  Search --> Gates
  Gates --> Manifest

Code anchors today: UnifiedNewsItem and SyndicationConfig in crates/vox-publisher/src/types.rs; publisher orchestration in crates/vox-publisher/src/lib.rs; SearXNG query URL in crates/vox-search/src/searxng.rs with defaults embedded from contracts/scientia/searxng-query.defaults.v1.yaml via crates/vox-search/src/searxng_defaults.rs and optional VOX_SEARCH_SEARXNG_ENGINES / VOX_SEARCH_SEARXNG_LANGUAGE overrides in crates/vox-search/src/policy.rs.

6. SSOT proposal: projection profiles

Extend the canonical publication metadata graph (see publication-worthiness doc, Deliverable 2) with distribution projection profiles:

1. identity / evidence / policy blocks remain canonical—adapters do not fork truth.
2. Each channel (Twitter, Reddit, LinkedIn, YouTube, …) references a projection_profile_id resolved from contracts/scientia/ (YAML) rather than from ad hoc env vars.
3. A projection profile specifies:
   • Template (max length, thread vs single, video vs text).
   • Allowed claims (which manifest fields may appear in public text—no uncertain metrics presented as facts).
   • Surface (for Meta: feed vs reels vs story as distinct profiles).
   • Posture (syndicate_once, manual_assist, ingest_only).
   • Throttle (min spacing, max items per day)—operator-tunable without rebuild.

This mirrors the existing idea of compiling Crossref / arXiv / social from one graph; it only makes the social side as explicit as the bibliographic side.

7. Measurement framework: useful vs noise

These are research-level KPI definitions for operators and future telemetry—not implied as shipped dashboards.

8. Automation boundary ledger (alignment)

Publication-worthiness research defines actions that must remain never_automate without explicit human accountability. Multi-channel syndication inherits those boundaries:

• No automatic deny of a manuscript based solely on projected social “virality.”
• No automatic bypass of ethics / disclosure / citation gates because a channel prefers shorter copy.

Cross-reference: Deliverable 1 table and never_automate ledger language in scientia-publication-worthiness-ssot-unification-research-2026.md.

9. Balancing the two problems (design recap)

10. Works cited and link registry (Tier A emphasis)

Use this table as a maintenance checklist when URLs rot or products rebrand. Prefer archived copies for long-lived policy citations where possible.

11. Changelog