WEBMIND RESEARCH PIPELINE

A bounded pipeline for observation, experiments, evidence, and human-approved change.

WebMind connects the Glyphd corpus, executable benchmarks, ProofStatus, and approval controls. It reads identified inputs, produces candidate work, dispatches tests, collects receipts, and waits for human approval before canonical changes.

CURRENT STATE

READY FOR ACTIVATION

The repository contains the workflows, schemas, approval boundary, and public surfaces. Scheduled execution depends on GitHub Actions capacity and the listed runtime configuration.

10proposal classes
10opportunity classes
7submission channels
0automatic canon writes
PIPELINE STAGES

Explicit transitions from observation to accepted evidence.

01

Observe

Read canonical papers, claim ledgers, benchmark contracts, results, failures, and open questions.

02

Propose

Generate a bounded set of candidate upgrades, applications, hypotheses, demonstrations, code changes, and experiments.

03

Test

Dispatch deterministic or self-hosted work and collect exact configuration, logs, metrics, hashes, and limitations.

04

Classify

Keep INVALID, FAILED, INCONCLUSIVE, and AWAITING_REVIEW distinct from a promoted result.

05

Approve

Present consequential proposals and submissions in ProofStatus admin for a human decision.

06

Record

Commit accepted artifacts and receipts into the corpus as versioned canonical state.

HUMAN CONTROL

Human approval governs changes, submissions, and claims.

Approvals

Every proposal states what it changes, expected value, risks, related projects, evidence state, deliverables, and the experiment that could prove it wrong.

Competitions

An MCP search connection discovers open competitions, prizes, grants, calls, and programs, then matches them to existing work or proposes a legitimate new project structure.

Submissions

Competition, publication, university, scientific, and corporate packets progress through IDEA, DRAFTING, READY FOR REVIEW, APPROVED TO SEND, and SUBMITTED.

Benchmarks

Hosted checks and private self-hosted experiments produce result packets. Completion is not PASS; the evidence gate and human acceptance remain separate.

RUNTIME CONFIGURATION

Scheduled execution depends on the listed environment and Actions secrets.

GitHub Actions secrets

  • OPENAI_API_KEY
  • WEBMIND_REPO_TOKEN

Optional: WEBMIND_MCP_SEARCH_URLWEBMIND_MCP_SEARCH_TOKENWEBMIND_MODEL

ProofStatus Vercel environment

  • WEBMIND_ADMIN_PASSWORD
  • WEBMIND_SESSION_SECRET
  • WEBMIND_REPO_TOKEN

The ProofStatus admin System tab reports the runtime configuration currently available to the pipeline.

NON-NEGOTIABLE BOUNDARY
  • Canon changes only through reviewed, receipted commits.
  • The system may propose; it may not self-accept.
  • A failed or inconclusive experiment remains part of the record.
  • Competition and publication discovery never authorizes submission.
  • Plain English precedes implementation detail in every approval packet.
  • Proofstatus is the evidence and control surface; Glyphd Labs is the research corpus and execution environment.

Read the evidence doctrine →