Discovery, evaluation, learning, and change proposal — analyzed by Philippe, researched by Riley, reviewed by Francois.
The Talent Factory has 10 R&D-related skills (3 core, 4 toolkit, 3 knowledge management) and has completed 10 evaluations in the REQ-EXEC series. The individual tools are well-designed. The problem is the pipeline between them.
All 10 evaluations were CEO-triggered. Zero proactive scans have ever been saved. The scans/ folder is empty. Accepted evaluations produce recommendations that vanish — no traceability to implementation, no outcome tracking, no feedback loop.
Recommendation: Expand from 4 informal sub-processes to a 6-stage gated pipeline with a Technology Radar, adding three missing stages (Decide, Prototype, Validate) and fixing the broken connections between existing stages. The kanban-style dashboard becomes the single source of truth.
The current R&D lifecycle has 4 informal sub-processes. Each has its own tools and artifacts, but the handoffs between them are manual, ad-hoc, and largely dependent on the CEO remembering to take action.
13 gaps identified across all 4 sub-processes, assessed by dimension. High-severity gaps break the pipeline; medium gaps cause friction; low gaps are inefficiencies.
| ID | Sub-Process | Gap | Severity |
|---|---|---|---|
| G1 | Discovery | No discovery cadence — /rd-scan has never saved a digest | High |
| G2 | Discovery | Watch list is a dead end — no mechanism to revisit | Medium |
| G3 | Discovery | Video discovery is fully manual | Low |
| G4 | Discovery | No single intake funnel — 5 entry points, 5 artifact formats | High |
| G5 | Evaluation | Evaluation is one-shot — parked items have no re-eval trigger | Medium |
| G6 | Evaluation | No evaluation queue — CEO must remember what needs evaluating | Medium |
| G7 | Learning | Dashboard copy/paste can't trigger skills from browser | Medium |
| G8 | Learning | Dashboard is 138KB static HTML — no data separation | High |
| G9 | Learning | Video insights stay in flashcards — don't flow to factory | Medium |
| G10 | Learning | kb-ingest exists but is disconnected from R&D output | Medium |
| G11 | Change | No traceability from evaluation to backlog implementation | High |
| G12 | Change | Manual routing — CEO is single point of failure | High |
| G13 | Change | No completion tracking — did the recommendation actually get built? | High |
Francois identified 4 framework-level gaps that Philippe's process analysis did not surface:
10 opportunities scored on Impact (1-5) × Feasibility (1-5). Plotted on a priority matrix to guide phased implementation.
Opportunities placed by their priority score. Quick wins go first.
O1 Kanban R&D dashboard (5×4=20)O8 Learning extraction pipeline (4×3=12)O6 Data-separated dashboard (5×3=15)O4 Eval-to-backlog traceability (4×5=20)O2 Auto-route eval results (4×4=16)O5 Watch list mechanism (3×5=15)O10 Completion tracking (3×5=15)O9 Dashboard triggers skills (5×2=10)O7 Scan cadence enforcement (3×5=15)O3 Discovery intake funnel (4×3=12)Expand from 4 informal sub-processes to a 6-stage gated pipeline. Three new stages fill the gaps identified by Francois's framework review. A Technology Radar provides strategic direction.
Maintained by Riley, reviewed quarterly. Defines what the factory is actively looking for — not just scoring what it finds.
Quadrants: Techniques · Platforms · Tools · Frameworks
What changes: Single entry funnel normalizing all 5 discovery sources into one queue. All items enter through /rd-intake (new skill) regardless of source.
Owner: Riley
Gate 0: Is this relevant to the factory's technology radar categories? Yes → Stage 1. No → Log and skip.
Fixes: G1 (no cadence), G4 (no intake funnel), G6 (no evaluation queue)
What changes: Nothing — /rd-evaluate's 5-dimension scorecard works well.
Owner: Riley
Gate 1: Score ≥ 3.0 → Stage 2. Score 2.0-2.9 → Hold (schedule re-eval in 30 days). Score < 2.0 → Kill.
Fixes: G5 (one-shot evaluation — Hold items now get re-eval triggers)
What changes: CTO Clara formally records Go/Hold/Kill decisions with rationale. Lightweight — just new fields in request.md frontmatter: decision: go|hold|kill, decision_date:, decision_by:.
Owner: CTO Clara
Gate 2: Does this fit the current portfolio? Go → Stage 3 or 4 (depending on complexity). Hold → Re-evaluate in 30 days. Kill → Archive with rationale.
Notifications: N-RD-DECIDE → Riley (feedback), Pablo (if production impact), Francois (if framework impact), Elena (if architecture impact)
Fixes: G12 (CEO bottleneck — Clara owns the decision, not CEO)
What changes: Explicit experimentation stage for complex items. Simple adoptions (e.g., "use this MCP server") skip directly to Stage 4.
Owner: Varies — Riley for research spikes, engineering role for technical PoCs
Gate 3: Did the prototype produce sufficient evidence? Yes → Stage 4. No → back to Stage 2 (re-decide with new data).
Artifact: Prototype findings document in the REQ-EXEC process/ folder
Framework basis: ISO 56002 concept validation, SAFe Continuous Exploration
What changes: Adds impact assessment checklist and mandatory traceability link. Pablo is now in the loop for production-impacting changes.
Owner: Riley (proposes) + Pablo (assesses production impact)
Gate 4: Has impact been assessed against: (a) production pipeline, (b) framework library, (c) deployed talents, (d) backlog capacity? All assessed → create CI-XXX or REQ-XXX with references: REQ-EXEC-NNN.
Fixes: G11 (no traceability), G12 (manual routing)
What changes: Closes the feedback loop. Did the recommendation actually get implemented? Did it work?
Owner: Riley
Gate 5: Outcome recorded on originating REQ-EXEC. Field: outcome: implemented|partial|abandoned|superseded.
Feedback loop: Outcome data feeds into technology radar updates and future evaluation calibration.
Fixes: G13 (no completion tracking)
How the redesigned R&D process connects to other factory departments. New notification IDs to register in TFD-0008.
Phased implementation. Quick wins first, then the dashboard, then the full pipeline automation.
outcome: and decision: fields to REQ-EXEC request.md. Add references: to /ci entries. Create watch list file. Schedule weekly /rd-scan.This report proposes a 6-stage R&D pipeline, a Technology Radar, 7 CI items, and 4 decisions. Review the analysis above and record your decision.