| name | research-techniques |
| description | Use when a research project needs a specific named technique — mini-analysis, crosswalk matrix, cross-cohort synthesis, gap analysis, reference interview, controlled vocabulary, pain-point taxonomy, search-operator grammar, Google-search API. Companion to research-orchestration; load only the technique the situation demands. |
| metadata | {"portable":true,"compatible_with":["claude-code","codex"]} |
Research Techniques
Use When
- A defined research plan needs one named analysis, synthesis, interview, taxonomy, vocabulary, or search technique.
Do Not Use When
- Use research-orchestration for waves and research-design for a full method portfolio.
Inputs
| Input | Source/provider | If absent |
|---|
| Defined question and technique objective | Research plan or analyst | Stop and clarify the supported decision. |
| Corpus or interview/search access | Evidence registry or provider | Return a technique plan and access gaps. |
Workflow
- Match the question to one technique and load its reference.
- Confirm corpus, fields, permissions, and stop conditions.
- Execute read-only, recording exclusions and contradictions.
- Validate the artifact and recover by routing gaps back to orchestration.
Outputs
| Artifact | Consumer | Acceptance condition |
|---|
| Technique-specific matrix, taxonomy, record, query set, or synthesis | Analyst and orchestrator | Inputs, method, exclusions, findings, and gaps are visible. |
Evidence Produced
| Category | Artifact | Acceptance condition |
|---|
| Correctness | Reproducible technique record | A reviewer can trace corpus, operations, and findings. |
Capability Contract
Analysis defaults to read-only. Interviews, paid APIs, account use, external contact, file edits, or publication require explicit authority.
Degraded Mode
With incomplete corpora, unavailable APIs, or inaccessible interviewees, produce the smallest supported artifact and label coverage gaps. Never infer missing counts or quotations.
Decision Rules
| Choice | Action | Failure/risk avoided |
|---|
| Heterogeneous sources | Use a crosswalk matrix | Hidden gaps |
| Missing evidence shapes next wave | Use gap analysis | Repeated searches |
| Cross-cohort patterns requested | Verify before synthesis | Unsupported generalisation |
Quality Standards
The technique answers a named question, preserves provenance, exposes exclusions, and yields a reusable artifact.
Anti-Patterns
- Technique without a question. Fix: start from the decision.
- Counting uncoded fragments. Fix: define a codebook.
- Synthesising unverified cohorts. Fix: verify checkpoints.
- Hiding empty cells. Fix: mark gaps.
- Treating search results as findings. Fix: evaluate sources.
Worked Example
When cohort reports use different labels, build a shared-dimension crosswalk, preserve unmapped cells as gaps, and synthesise only after verification.
References
Library of named research techniques the engine applies on demand. The orchestrator (research-orchestration) decides what kind of research is being run; this skill provides the specific tools used inside the waves.
Whenever a technique produces a mini-conclusion, synthesis, gap claim, taxonomy interpretation, cross-cohort inference, or recommendation, run critical-reasoning-and-argument before the artifact is considered complete.
When to use
The orchestrator (or another skill) tells you the project needs a named technique. Load only the matching reference. Do not load all references by default.
Technique router
| Need | Reference | One-line description |
|---|
| Decompose a big question into smaller answerable analyses | references/minianalysis-engine.md | Abbott's brachiation — every research question broken into mini-analyses |
| Compare findings across heterogeneous source types | references/crosswalk-matrix.md | Sources × dimensions matrix to surface gaps and contradictions |
| Synthesize across cohorts / segments / regions | references/cross-cohort-synthesis.md | Wave-4 work: cross-cohort patterns, divergences, product/policy inferences |
| Identify what the existing literature has not answered | references/gap-analysis.md | Feeds Wave-2 brief construction; structures missing-evidence calls |
| Interview a librarian / SME / archivist for the right starting points | references/reference-interview.md | Bell's reference-interview discipline; reframes vague queries |
| Build a vocabulary the engine can reuse (subject headings, synonyms, related terms) | references/controlled-vocabulary-builder.md | Bell + Bergeron; precision-recall dial |
| Build a taxonomy of pain points / complaints / use cases from raw qualitative data | references/pain-point-taxonomy.md | Code → cluster → name → counts → exemplar quotes |
| Master Google / Bing / archive search operators | references/search-operator-grammar.md | site:, intitle:, inurl:, filetype:, before:, after:, allintext:, around:, exact phrase, exclusion, OR/AND |
| Programmatic search via Google CSE / Bing Web Search APIs | references/google-search-api-operator.md | API setup, query budget, JSON parsing, rate limits |
| MacLeod's full search-mastery layer — operator catalog, deep-web heuristics, anti-patterns, "never compile what's already compiled" | references/macleod-search-mastery.md | Don MacLeod, How to Find Out Anything (Tier 1) |
| Russell's metacognitive layer — clarify the question, lateral reading, knowing when to stop, triangulation standards, leading-question lint | references/russell-search-literacy.md | Daniel M. Russell, (Tier 1) |
Orchestration
new research wave
├─ load research-orchestration (decides type/discipline/reading mode + wave plan)
├─ if Wave-2 gap fill needed
│ └─ load references/gap-analysis.md
├─ if a question is too big for one analysis
│ └─ load references/minianalysis-engine.md
├─ if the source set is heterogeneous (academic + media + government + interview)
│ └─ load references/crosswalk-matrix.md
├─ if multi-cohort synthesis required
│ └─ load references/cross-cohort-synthesis.md
├─ if vague query needs reframing
│ └─ load references/reference-interview.md
├─ if precision/recall tuning matters
│ └─ load references/controlled-vocabulary-builder.md
└─ when issuing search queries
├─ load references/search-operator-grammar.md (manual)
└─ load references/google-search-api-operator.md (programmatic)
Universal rule — every technique outputs structured artifacts
Every technique in this skill produces an artifact the orchestrator can consume:
- mini-analysis → list of sub-questions, evidence-per-sub, mini-conclusion.
- crosswalk matrix → table (sources × dimensions) with cell entries.
- cross-cohort synthesis → patterns, divergences, inferences.
- gap analysis → list of named gaps with evidence-of-absence.
- reference interview → reframed query + recommended starting points.
- controlled vocabulary → preferred terms + synonyms + related terms + scope notes.
- pain-point taxonomy → coded clusters with exemplar quotes and counts.
- search query → operator-rich string + estimated result-set characteristics.
Any artifact with a conclusion also includes the evidence-to-claim link, countercase, and confidence limit from critical-reasoning-and-argument.
A technique that does not produce a structured artifact has been used incorrectly.
Technique Failure Modes
- Loading all references at once instead of the matching one.
- Using a search-operator grammar reference to write English prose.
- Inventing pain points; the taxonomy must be built from coded raw data only.
- Synthesizing across cohorts before each cohort has its own complete corpus.
- Treating gap analysis as "things I haven't read yet"; gap analysis names what the literature has not answered.
- Technique output that lists evidence but does not test the inference drawn from it.
Companion skills
research-orchestration — the dispatcher; this skill is its toolbox.
critical-reasoning-and-argument — mandatory for mini-conclusions, gap claims, cross-cohort synthesis, taxonomy interpretation, and recommendations.
research-design — formal research design (historical, trend, MROC, knowledge lifecycle, design document, report builder).
source-evaluation — mandatory pairing whenever sources are involved.
academic-writing, report-and-proposal-craft, business-writing — output containers.