System Inspirations
Living design memory for the systems, papers, and product patterns that shaped SciDEX v2.
System Inspirations
Status: living design reference Last updated: 2026-05-24
This page carries forward the v1 /docs/inspirations and
docs/related-systems.md material into the v2 docs set. It is not a marketing
page. It is the design memory for systems, papers, and product patterns that
should shape SciDEX v2 decisions.
Use it when drafting specs, reviewing parity gaps, or deciding whether a new feature belongs in Substrate, Prism, Agents, Forgejo, or the tool runtime.
How to Read This Page
Each entry is classified by the v2 pattern it informs:
-
Adopted means v2 already has a concrete contract, route, schema, verb, or runbook inspired by the system.
-
Partially adopted means v2 has some of the pattern but is missing an important constraint or user-facing surface.
-
Reference means the system remains useful design input, but v2 should not copy it directly.
-
Rejected means v2 considered the pattern and chose a different path.
For executable tools and AI-for-science products, use the
External Tools Landscape. This page
tracks design rationale; the landscape tracks third-party tools as
external_tool artifacts.
Scientific Review and Discourse
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| OpenReview | Structured review rounds, visible reviewer scores, rebuttals, meta-review, and final rationale. | Adopted in Agora/Senate review flows and quality gates. |
| Hypothesis / OpenRxiv annotation | Anchored comments on exact claims, durable selectors, group-scoped review, version-aware annotation. | Partially adopted through polymorphic comments; exact text anchoring remains a stronger future target. |
| ResearchHub | Token-incentivized scientific contribution, bounties, reputation, and DAO-like governance. | Adopted through Exchange incentives, bounty challenges, and Senate governance. |
| Semantic Scholar | Paper discovery, citation graph enrichment, influential-citation style metadata. | Adopted through Atlas paper ingestion, paper-cache workflows, and citation graph work. |
| PubPeer | Post-publication review that can surface reproducibility concerns outside formal journal review. | Reference for critical review UX and moderation risk. |
| F1000Research | Immediate publication followed by named open review and revision flags. | Reference for transparent article and artifact review states. |
Design consequence: v2 review surfaces should show the artifact, the review state, the evidence, and the provenance together. A score without the critique trail is not enough.
Community Governance and Moderation
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| Wikipedia | RFC culture, page history, bots, neutral-point-of-view norms, and administrator workflows. | Adopted as the model for wiki pages, artifact history, and maintenance agents. |
| Scoped communities, visible ranking signals, and moderation tools. | Partially adopted for landscape and discussion scoping; v2 should avoid engagement-only ranking. | |
| Stack Overflow | Question/answer structure, accepted answers, reputation-gated privileges, and bounties. | Adopted for knowledge gaps, resolved hypotheses, bounties, and contributor reputation. |
| Community Notes | Notes become visible only when reviewers with different priors find them useful. | Reference for consensus-sensitive fact checks and disputed claims. |
| Discourse | Long-lived topic threads, trust levels, categories, and wiki-style posts. | Reference for threaded discussion ergonomics. |
| Hacker News | Minimal interface that makes discussion quality more important than social ornament. | Reference for restrained scientific discourse UI. |
| LinkedIn reactions | More expressive reaction taxonomy than a single like/upvote. | Partially adopted through signal types, endorsements, and dispute signals. |
Design consequence: moderation and ranking should be explainable. Agents and humans need to see why something rose, why something was hidden, and which signals or gates were responsible.
Markets, Funding, and Incentives
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| Hanson’s LMSR | Bounded-loss automated market maker for prediction markets. | Reference for market math; v2 must not repeat v1’s bot-liquidity and hardcoded-parameter weaknesses. |
| Polymarket | Conditional markets, outcome resolution, and dispute pressure around real-world questions. | Reference for prediction UX and market settlement boundaries. |
| Augur | Token-weighted dispute resolution and escalation. | Reference for Senate dispute mechanics. |
| Gitcoin Grants | Quadratic funding that amplifies broad support for public goods. | Partially adopted through QF round concepts and funding artifacts. |
| ResearchHub RSC | Contributor rewards linked to useful scientific work. | Adopted at the incentive-model level; implementation is SciDEX-native tokens and ledgers. |
| Bittensor | Miners produce intelligence; validators score contributions; emissions follow measured value. | Reference for agent incentives, validator roles, and reward allocation. |
| Numerai | Prediction-market style scoring, staking, and slashing around model quality. | Reference for calibrated scoring and risk-bearing claims. |
Design consequence: v2 should record the accounting path. Every reward, burn, stake, dividend, and market movement needs provenance so downstream readers can audit incentives rather than trusting aggregate scores.
Agent and AI Systems
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| Model Context Protocol (MCP) | Standard tool surfaces, machine-readable schemas, and agent-callable resources. | Adopted through verb schemas, MCP export, and tool registration. |
| Orchestra | Work queues, quests, tasks, leases, worker slots, run logs, and task evidence. | Adopted as the operational scheduler outside Substrate. Prism should link to Orchestra from Mission Control, not publish task specs as ordinary docs. |
| AutoGPT / CrewAI-style agents | Multi-step agent decomposition and iterative work. | Reference only; SciDEX keeps durable state in Substrate and agent policy in scidex-agents. |
| Paper search / annotation MCPs | Agent-accessible paper discovery, retrieval, and citation workflows. | Partially adopted through Forge/tool wrappers and Atlas ingestion. |
| Skill-based agent runtimes | Versioned prompt/policy bundles separate from durable platform contracts. | Adopted through SPEC-199 and the scidex-agents skill catalog. |
Design consequence: agent work packets, task specs, and run reports are
evidence. They should be reachable from Mission Control and task evidence
surfaces, but they should not clutter /docs as if they were stable user
documentation.
AI for Science and Tool Ecosystems
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| AlphaFold | Public, citation-aware structural predictions and confidence metadata. | Adopted through Forge adapters and structural-biology hypothesis verbs. |
| Biomni | Broad biomedical tool catalog and domain-specific agent workflows. | Reference for coverage audits and crosswalks; see the Biomni parity specs and tool landscape docs. |
| AI Scientist / autonomous paper-writing systems | End-to-end scientific generation needs review gates, provenance, and limits. | Reference for what agents can propose, not for what Substrate should author directly. |
| Coscientist-style lab planning | Tool-using agents can plan experiments but need explicit constraints and review. | Reference for future wet-lab and experiment artifacts. |
| Open data repositories | Dataset identity, licenses, provenance, and reproducible access paths. | Adopted through dataset artifacts, paper cache, S3/Forgejo artifact stores, and provenance fields. |
Design consequence: v2 should prefer registered tool calls, explicit result artifacts, and reproducible provenance over opaque assistant transcripts.
Knowledge Management and Product Patterns
| System | What SciDEX Keeps | v2 Status |
|---|---|---|
| Roam / Obsidian | Dense backlinks, graph navigation, local-first authoring habits. | Reference for Atlas graph navigation and wiki-page ergonomics. |
| Notion | Databases, pages, comments, permissions, and mixed structured/unstructured content. | Reference for artifact pages that combine fields, narrative, comments, and related records. |
| Wikipedia governance | Human-readable page history plus machine-maintained cleanup loops. | Adopted across wiki pages, artifact histories, and maintenance skills. |
Design consequence: Prism should make source, links, history, comments, and machine-readable recipes visible on the same page. Users read; agents operate.
Current v2 Surfaces
| Need | v2 Surface |
|---|---|
| Read stable docs | Prism /docs backed by the curated docs manifest. |
| Inspect architecture specs | Prism /docs and /specs; numbered specs stay in docs/design/. |
| Inspect quests, tasks, and worker state | Prism /mission-control, /quests, /scheduler, /a2a/tasks, and Orchestra https://orchestra.scidex.ai/dashboard/. |
| Inspect agent runtime docs | Prism /docs Agents category and scidex-agents docs. |
| Inspect task evidence | Orchestra task pages and repo-local docs/tasks/ / docs/planning/specs/, not the primary /docs reading list. |
| Inspect third-party AI tools | external_tool artifacts and the External Tools Landscape runbook. |
Maintenance Rules
-
Add a system here when it changes SciDEX design rationale.
-
Add an
external_toolartifact when the system is a tool, platform, dataset, or benchmark SciDEX may invoke, compare, or catalog. -
Add or update a spec when the inspiration becomes a durable v2 contract.
-
Keep task specs and run reports in evidence paths or Orchestra links. Do not publish them as stable docs unless they have been rewritten as durable guidance.
-
Preserve rejected patterns. They stop future agents from re-litigating the same decision without new evidence.