planning-and-research

03 — Planning and Research

Decompose work into parallel vertical slices using deep web research, competitor scanning, and technology evaluation.

Deep research protocol

Before any non-trivial implementation:

1. Web search top 50 results for primary keyphrase
2. Read top 10 in depth — extract directives + counter-arguments
3. Cross-ref against existing rule mesh
4. Confidence-track every claim 0-1

Use web_search_20260209 + web_fetch_20260209 (free when paired with code_execution_20260120).

Competitor scanning

For every website build, run rules/competitor-research.md Phase -1 BEFORE Phase 0:

• Identify top 5-10 audience-comparable sites
• Score each on 100-pt rubric (10 dims × 10pts)
• Set ≥15% beat-floor for Phase 6 loop termination

Technology evaluation

For every new dep / framework / service consideration:

1. Already have equivalent? → use it
2. Feature truly needed? → if speculative, defer
3. License? → OSS free only (per rules/brian-preferences.md priority order)
4. CF compat? → adapter pattern per rules/cloudflare-hostable-supervisor.md
5. Bundle/perf impact? → measure before commit
6. Lighter existing solution? → prefer
7. Decide: install now · defer · adapter-only · reject + document why

Implementation planning

Decomposition

• Identify atomic units (file × layer)
• Mark dependencies between units
• Distinguish independent vs serial chains

Vertical slicing

Per 06-build-and-slice-loop:

• Each slice ships through every layer (UI → API → DB → tests → deploy)
• Homepage FIRST (no exceptions)
• Slice = one feature, not one layer

Parallelism plan

Per rules/parallel-subagent-economy.md:

• ≥5-min wall-clock saving + independent → fan out
• Sweet spot 3-4 specialists, ceiling 6
• Batch beyond 6 in waves
• Sonnet default; Opus for security/architect/visual-qa overrides

Critical path

• Identify longest dependency chain
• Front-load research that gates implementation
• Parallelize everything off the critical path
• Wall-clock = max(critical_path, max parallel branch)

Assumption tracking

Every assumption logged in _assumptions.md with:

• Claim
• Confidence 0-1
• Evidence sources
• Decision impact if wrong
• Verification trigger

Confidence <0.7 → research more. Per rules/auto-meta-work.md.

Three-Layer Knowledge

Per ~/.claude/CLAUDE.md § Thinking:

• L1 = proven (existing rules + patterns)
• L2 = trending (blog posts, recent papers)
• L3 = first principles (build from scratch)

Prefer L3. Best outcome of research is NOT finding a solution to copy — it's understanding the problem deeply enough to design a better one.

Self-Argue (before major decisions)

Generate strongest counterargument. If you can't defeat it, decision is wrong.

Boil-the-lake

Marginal cost of completeness is near-zero. When complete costs minutes more than shortcut, do complete. Boil lakes, flag oceans.

Output artifacts

• _research.json — raw findings, source URLs, confidence
• _assumptions.md — tracked claims
• PLAN.md — implementation roadmap w/ parallelism plan + critical path
• _decisions.md — architectural decisions w/ rationale + alternatives
• _brief_summary.txt — 100-word digest for downstream agents