Extracts Factory Service Manual (FSM) section codes (e.g. 8A-9, 6E-92) from contents tables and body text and writes them onto the Qdrant manuals payload, giving a second, structural retrieval axis alongside the embedding-derived system tags. On telep-mainframe, code under ~/knowledgebase/kb-vectors/.

For Agents — what gets written

  • A finite-state parser pulls FSM codes (8A-9, 6E-92, …) from contents tables + text.
  • Qdrant payload gains: section_codes, section_names, section_prefixes on 490 pages.
  • Index file: kb-vectors/section_index.json = 1254 codes.

Prefix → system-tag map

Each code’s alpha-numeric prefix maps to a vehicle system:

Prefix(es)System
8A / 8B / 8CElectrical
6E / 6C / 6J / 6KFuel & Emissions
6AEngine
6BCooling
7A7FDrivetrain
3xSteering & Suspension
1A / 1BHVAC
9JBody
0xDiagnostics

Complementary to the text system-tags, not redundant

~61% agreement = two independent signals worth keeping both

The section-prefix system and the embedding-derived text system-tags (see 2026-08-06-kb-embedding-classifiers-system-tags-and-image-types) agree ~61% of the time. That partial agreement means they are complementary, not redundant — the section code is a hard structural signal from the manual’s own contents table, while the text tag is a soft semantic signal from the page prose. Where they disagree, both are worth surfacing (one may catch what the other misses).