Service Line

R&D / Evaluation

R&D / Evaluation Service Line

One-liner. Take a candidate (tool, framework, MCP server, plugin, market trend, methodology) and produce a structured evaluation with a Go / No-go / Hold decision.

When to use this line

  • "Is Computer Use ready for production?" → tool evaluation
  • "Should we adopt LangGraph for the factory?" → framework evaluation
  • "What's the competitive landscape for AI coding agents?" → market scan
  • "Watch this YouTube playlist and tell me what's relevant" → video intelligence
  • "Evaluate Maya RAG-as-wiki as a Confluence replacement" → architectural eval

Do not use this line for: building the talent (→ Talent Factory Build, often a follow-on if the eval is Go), redesigning a process (→ Consulting), mapping an architecture (→ EA).

Inputs

/request-create --service=rd (or /rd-intake) collects:

  1. Topic + source URL or reference
  2. Source type (tool / framework / MCP / plugin / paper / video / market)
  3. Driving question (what decision will this inform?)
  4. Decision deadline (when do we need to know?)
  5. Comparison set (what else are we comparing against?)
  6. Eval criteria (or use default RD pipeline rubric)

Standard production process

The 6-stage R&D pipeline (TFD-012).

1. Intake          /rd-intake             → RD-NNNN folder created
   ↓
2. Triage          relevance score
   ↓
3. Deep dive       hands-on or doc-driven evaluation
   ↓
4. Scorecard       rubric-driven scoring
   ↓
5. Decision        Go / No-go / Hold + rationale → TFD if structural
   ↓
6. Publish         to /research/evaluations and intranet

See /rd-evaluate, /rd-scan, /rd-status skills for tooling.

Deliverables

  • RD-NNNN/intake.md — initial capture
  • RD-NNNN/evaluation.md — deep dive notes
  • RD-NNNN/scorecard.md — rubric scoring
  • RD-NNNN/decision.md — Go / No-go / Hold + rationale
  • TFD entry if the decision is structural

Acceptance Criteria + DoD

  • Scorecard complete against the standard rubric
  • Decision documented with rationale (not just verdict)
  • Re-evaluation date set if Hold
  • Linked from /research/evaluations on the intranet
  • TFD authored if Go drives a factory-level change

Publishing target

Internal first. Lands on intranet under /research/evaluations. If Go → spawns a follow-on work order in the appropriate service line (Talent Factory Build, EA, Consulting).

Decision memos may be published to JCT portail client when the eval is client-facing (rare).

Worked examples

Eval Verdict Status
REQ-EXEC-016 — Computer Use Hold (re-eval 2026-05-28) Parked
REQ-EXEC-017 — Telegram channels MCP Go → TFD-015 Activated
RD-0003 — synthesis-strategique (factory needs beyond consumer Claude Code) Informs ongoing decisions Reference
/km-analyser-video outputs Various Continuous (playlist scan)

Lead role

Riley — R&D Analyst. Owns intake, triage, evaluation, scorecard. CEO (Oscar) owns Go / No-go decision on structural items.

Source-of-truth links

  • Skills: /rd-intake, /rd-evaluate, /rd-scan, /rd-status, toolkit:video-analyse
  • Pipeline TFD: company/decisions/TFD-012-rd-pipeline.md
  • Intranet view: /research/evaluations
  • Memory: project_rd-pipeline

Status

Active. Continuous scan running via /rd-scan and YouTube playlist intelligence.