Lantronix Phase 1 Foundation Build

Executive summary

The durable phase-1 foundation is now on disk. The build creates a stable workspace path, first-class schemas for Company, Product, Customer Segment, and Evidence / Source, and a deterministic recast spec for the legacy all_enriched.json dataset that preserves lineage without inventing product structure.

The right next slice is a narrow importer plus pilot migration: seed the customer segment taxonomy, migrate 10 to 15 US-priority companies into Company and Evidence records, then add explicit Product records only where named products can be sourced from company pages.

What was built

Path	Purpose
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/README.md`	phase-1 foundation overview and folder contract
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/schema/company.schema.json`	Company schema, close to legacy but extended for phase-1 needs
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/schema/product.schema.json`	Product schema for first-class product records
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/schema/customer-segment.schema.json`	Customer Segment schema, US-first and NDAA-aware
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/schema/evidence-source.schema.json`	Evidence / Source schema for traceable claims
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/docs/recast-mapping-legacy-uas-companies.md`	deterministic legacy field mapping and unresolved-field rules
`/Users/vinny/.openclaw/workspace/data/lantronix-uas-workbench-phase1/records/...`	target record folders for future importer and curated updates

Build choices made in this slice

Kept the Company schema close enough to the legacy dataset to allow deterministic recast with low ambiguity.
Made Product first-class in schema, but blocked auto-generation from free text.
Added Customer Segment linkage through a fixed vertical-to-segment crosswalk.
Added Evidence / Source as a required first-class record type so imported claims can be tracked as legacy seed rows until upgraded.
Left strategic relevance, platform domain, revenue band, channel model, and detailed product structure unresolved on purpose.

What remains

Remaining item	Why it was deferred
pilot importer script	outside this slice, but now unblocked by schema and mapping
customer segment seed records	straightforward next step once importer exists
first-pass product records	needs explicit source-backed product names, not free-text splitting
Lantronix strategic relevance rubric	requires a defined scoring rule tied to the engagement
report templates	better built after sample records prove the new model

Recommended next slice

Build one deterministic importer for a pilot set.

Scope:

seed the 16 customer segment IDs from the mapping crosswalk
migrate 10 to 15 US-priority legacy companies into normalized Company records
create legacy-seed-row and company-website Evidence records for each migrated company
add Product records only for companies whose product names are explicit on their site or source material
review the pilot output against the three report goals: competitive landscape, technical positioning, target customer segments

This is the fastest path to proving the model without repeating the failed broad migration path.

Risks and guardrails

Legacy descriptions and notes are not report-grade proof by themselves.
Product structure is the main ambiguity in the old dataset, so it stays source-gated.
The current model is US-first by intent. Non-US records can still exist, but should not drive the pilot set.
NDAA posture should stay explicit and conservative. unknown is better than over-claiming compliance.

Validation summary

I read the source spec, current legacy schema, and a representative sample of all_enriched.json before writing this foundation. After writing, I read back the created files to confirm that:

folder structure and record paths are consistent
field names align across schemas and the mapping doc
the recast rules match actual legacy fields present in the dataset
unresolved fields are called out instead of silently inferred
the next slice is unblocked without attempting a full migration now