Micro-switch keyword catalog v1
Date: 2026-10-12
Acceptance result
- Requested keywords: 10,000
- Generated keywords: 10,000
- Normalized unique keywords: 10,000
- Strict direct-topic accepts: 10,000
- Existing-database duplicates: 0
- Paid calls used by catalog generation: 0
- Database mutations: 0
- Articles or plans generated: 0
Distribution
- Industrial machine applications: 5,184
- Motion and access systems: 1,611
- Specific appliance, vehicle, transport, medical, office, and control applications: 2,205
- Actuator topics: 173
- Mechanical and electrical parameters: 239
- Contact and circuit topics: 222
- Environmental topics: 175
- Failure topics: 191
Every row contains an approved direct micro-switch term and passes the strict topic classifier. Keyboard switches, household lighting wiring, Nintendo, guitar electronics, brand navigation, commercial-intent wording, product models, and normalized duplicates are excluded.
The generated CSV and JSONL are runtime deliverables under pipeline/state/outputs/keyword-catalog-v1/. They are keyword candidates only. They do not assert that every application uses the same switch design, and they do not bypass later search-volume, evidence, planning, or article-content checks.
An asynchronous DataForSEO round using 20 direct micro-switch seeds returned 224 rows, of which 31 were net-new. Those rows were imported into the local raw keyword store, but the deterministic catalog excludes all keywords already present in that store, so the delivered 10,000 are net-new relative to the database.
Semantic clustering result
The catalog was imported into the pipeline, expanded with 2,000 additional strict direct-topic candidates, embedded locally with all-mpnet-base-v2, and clustered at cosine threshold 0.93. Semantic matches are also required to have the same reader task and at least 0.50 lexical subject overlap so that unrelated applications are not joined solely by vector similarity.
- Selected direct-topic keywords: 12,461
- Independent keyword clusters: 10,658
- Keywords consolidated into an existing cluster: 1,803
- Singleton clusters: 9,174
- Multi-keyword clusters: 1,484
- Largest cluster: 12 keywords
- Mean cluster size: 1.1692
- Embedding provider cost: $0 (local Mac model)
- Clustering threshold: 0.93 cosine similarity
The runtime deliverables are under pipeline/state/outputs/keyword-clustering-v1/:
keyword-cluster-primary.csv: one primary keyword per cluster; this is the main independent-topic list.keyword-cluster-membership.csv: every keyword and its assigned cluster.keyword-clusters.jsonl: cluster-level records with all member keywords.keyword-cluster-summary.json: aggregate counts.
This stage does not generate article plans or articles. The 10,658 cluster-primary keywords are the candidate input for later evidence checks and content planning.