Wireframe — Training Data

Diagnostic dashboard for the labor-line training pool. Shows total examples, breakdown by source (seed vs. refined), sparse categories that need more data, and recent refinements that were promoted. Useful for answering "is the model getting smarter for my org?"

{{ name }} — Training Data

House Doctors NoCo ▾ D Dan ▾

Org Profile Pricing & Defaults Integrations Members Branding Training Data Billing

Training Data

Historical labor lines this org's AI uses for retrieval. Approved refinements automatically add new examples (with embeddings) to the pool.

Total Labor Lines

494

▲ 7 this month

From Seed

487

initial HCP import · Jan 2026

From Refinements

growing each approval

Categories

3 sparse · need more data

Coverage by Category

Drywall & Patching 82

Painting 71

Plumbing 59

Carpentry 50

Electrical 46

Flooring 36

General Handyman 33

Fencing Sparse 16

Multi-Trade 32

Diagnostic / Other Low 7

Sparse Tasks · < 5 examples each

Category	Task	Examples	Avg Hrs	Suggestion
Diagnostic / Other	Mold remediation prep	2	3.5	Add more examples on next refinement
Fencing	Gate hardware replacement	3	1.2	High variance · refine to stabilize
Electrical	Smart switch retrofit	4	0.8	New trend · monitor

Recently Promoted from Refinements

Date	Category	Task	Hours	$	Source Estimate
1 min ago	Drywall & Patching	Drywall patch + sand 4x6 hole	5.0	$615.70	HD-20260501-1234
yesterday	Painting	Repaint trim + caulk seams	3.5	$430.99	HD-20260430-0987
2 days ago	Plumbing	Vanity swap + faucet hookup	4.8	$591.07	HD-20260429-0716

Notes

SCOPED Each org sees only their own training pool. House Doctors' 494 lines are theirs alone — customer #2 starts with whatever seed data we provide them (or empty).
SEED OPTIONS When onboarding a new org, options for the initial pool: (a) start empty, (b) seed with a generic anonymized industry pool, (c) import from their existing HCP estimate history. Decision deferred to onboarding design.
SPARSE TASKS Calls the existing internal.laborLineExamples.thinTasks query (per the handoff doc). Useful for prioritizing where Dan should focus refinements to improve future estimates.
RE-EMBED "Re-embed all" is an admin action that re-runs the embedding model over the entire pool. Useful when changing embedding models or after a bulk import.