Decisions (Alexander, 26 Oct 2026): a second company, not the plant
company, holds the catalogue; it owns and licenses titles; language
models are self-hosted on premises.
catalogue-intelligence.md reworked: the catalogue company is a customer
of the plant at arm's length (orders through intake and portal, the
plant's price list, standby pressing offered to every customer, no
moved dates, no data either way, metal in the archive on customer
terms; Revenue Code section 71 bis, 200M THB disclosure threshold per
the OECD profile), its own ERPNext site, the run-size rule on the
plant's price instead of its direct cost, its own advisor instance on
hardware it controls, phasing, cost borne by it, risks, open decisions
(shareholders and Poy Records, working capital, standby terms, where its
systems run).
Mirrors: README (planning bullet, project status), roadmap item 28,
digital-factory (tool rule, language-model row in the open-source
table), artist-platform (standby pressing proposal, pre-orders),
production-timeline, revenue-model, plant-assistant (self-hosted on
premises; catalogue data removed; advisor as a separate instance).
New docs/plant-assistant.md (decision, Alexander, 26 Oct 2026): an
assistant for the whole plant, asked in Thai or English at kiosks, desks
and phones, answering from the knowledge base (SOPs as ERPNext Quality
Procedures, vendor manuals on the Assets, recipes and limits, this
repository, lessons learned) and from ERPNext and the machine-data store,
with sources on every answer. Roles and questions, what it does
(explain, trace, summarise, alert context, plan vs actual, drafts) and
never does (machines, submitting, safety improvisation, seeing more than
the asker, rating operators, needing the internet). Built on Typhoon 2.5
(open weights, Apache-2.0, CPU trial first), Typhoon ASR for optional
voice, pgvector retrieval, MCP tools in cnx-factory through the one
integration layer and the PostgreSQL replica; vinyl-ai GPU server only if
needed (~250k to 300k THB, RTX 5090 Thai retail Oct 2026). Test set per
role, phasing, cost, risks, open decisions.
Mirrors: digital-factory systems and machines tables, rough cost note;
catalogue advisor becomes a role of the assistant; worker-safety
training; roadmap item 18; README bullet and layout.
New docs/catalogue-intelligence.md for Alexander's idea of a catalogue
the plant presses for itself. Why (idle Classic hours ~60k to 160k
records a year at 60 % specials on the doc basis; two-week represses
from the metal archive; Thai catalogue sells in Japan, EM Records,
ZudRangMa), scope (Poy Records as seed; customers' data and press dates
protected), data sources with access terms (Bandcamp ship-to fields,
distributor sell-through clauses, opt-in record passport registrations,
Spotify for Artists, Discogs CC0 dumps; Discogs API marketplace and
want data restricted), ERPNext touchpoints (Licence, Channel Sales
Report, Catalogue Recommendation, geocoded addresses, CRM), PDPA rules
for the map, dashboard views on an open-source map stack, a staged
demand model with a stocking rule for run size, a propose-only advisor
on an open-weights model, phasing, cost estimates, risks and open
decisions.
Backlinks: README (planning documents, layout), roadmap item 28,
digital-factory tool rule, artist-platform repress pre-orders,
production-timeline repress, revenue-model ResurRec Japan.
grabowski