Digital Research Engine

Digital Research Engine Router

This router turns a user request into the minimum skill stack required to produce research that is traceable, well-reasoned, and release-ready. It is the entrypoint for the engine, not a replacement for sub-skills.

Non-negotiable foundation

Always load these before research work:

1. skills/source-evaluation/SKILL.md

2. skills/source-evaluation/references/evidence-discipline.md

3. skills/anti-ai-slop/SKILL.md

4. docs/continuous-improvement/portfolio-kaizen-standard-2026-08.md for every engine or research-product audit, improvement plan, or book-study integration.

All engines may route here for current-source verification. This engine also audits its own research outputs: use skills/00-meta-initialization/references/kaizen-engine-and-product-audit.md, publish the current audit at a maximum of 65/100, and produce a remediation plan targeting 95/100.

Every sub-agent prompt must include the hard-constraint clause from evidence-discipline.md verbatim.

Route by user intent

Running example

Use docs/world-class-exemplars/running-example.md as the shared example across sub-skills: a Chwezi Core Systems research engagement on informal cross-border trade digitisation in East Africa. It includes executive, academic, OSINT, data, and policy variants without relying on invented statistics.

Standard operating sequence

1. Intake: define audience, decision, research type, output family, risk level, and stop conditions.
2. Evidence: load source-evaluation and evidence-discipline; assign source tiers at collection time.
3. Design: choose method, cohorts, search plan, and data needs.
4. Collection: run waves; log sources, search strings, exclusions, and unresolved gaps.
5. Verification: run source-verification and the verifier tool before synthesis.
6. Reasoning: run critical-reasoning and analytic-tradecraft where claims, judgments, or recommendations appear.
7. Synthesis: build the claim graph, answer-first message, and evidence pack.
8. Output: use the format skill matching the deliverable and apply anti-slop gates.
9. Release: run quality gates, citation density dashboard, reviewer notes, and final manifest.

Stop conditions

Stop and ask for clarification when the request lacks the decision/audience, asks for private surveillance, asks for fabricated or unsourced claims, requires current legal/regulatory facts that cannot be verified, or asks to publish a claim whose source trail fails verification.

See also

• docs/skill-authoring-standard.md
• docs/pathing-model-engine-vs-projects.md
• docs/quality-gates/release-blocking-gates.md
• docs/source-registers/research-standards-register.md
• templates/research-evidence-pack-template.md
• examples/research-types/README.md