feat(seed): expand industrial process catalogue with per-process subcategories

[mrivas00]

Context

Two reviewers flagged the same gap in Alcance 1: the industrial-process catalogue only covers cement, glass, steel and zinc, so a company running any other process has nowhere to declare it.

Gabriela: si la empresa realiza procesos industriales distintos a acero, fundición de metales, cemento o vidrio (por ejemplo industria química), no hay dónde incorporarlo.

Gianluca: en emisiones directas, ¿no es muy restrictivo solo acero, cinc y vidrio?

They are right, and the standards say why. The current set of four is an arbitrary subset of the IPCC IPPU tree: it includes glass and zinc, which the GHG Protocol's own sector table does not list, and omits lime, aluminium, chemicals and pulp & paper, which it does (GHG Protocol Corporate Standard, Appendix D "Industry Sectors and Scopes", pp. 92–93).

Nothing here needs code. The catalogue is seed data plus the maintainer screens, and no migration has ever touched methodology content — the one data migration in this repo, 20260819120000_unify_custom_factors, relabels a source string on captured inventory lines.

Two options, and why there are two PRs

Neither standard requires a subcategory per material. The GHG Protocol's own Alcance 1 taxonomy has a single "process emissions" bucket (ch. 6, p. 41), and subdividing by source type is explicitly optional information (ch. 9, p. 63). The IPCC, meanwhile, structures IPPU as a two-level hierarchy — branch (2A mineral, 2B chemical, 2C metal, 2H other) then leaf (2A1 cement, 2A2 lime, …), with an explicit Other (please specify) leaf on every branch (2006 GL Vol. 3, Fig. 1.1, p. 1.5).

So the real question is only where the boundary between subcategory and dimension falls. Both options cover the same IPPU leaves.

	Option A — IPPU branches	Option B — one subcategory per process (this PR)
Branch	feat/mati/industrial-processes-ippu-branches	feat/mati/industrial-processes-per-process
Subcategories in Alcance 1	11 → 11	11 → 18
…of which industrial	4 → 4	4 → 11
Shape	subcategory = branch, Proceso → Material o método cascade	subcategory = process, one flat dimension
Existing four subcategories	replaced	untouched
Dimension values / seeded IPCC factors	79 / 41	66 / 33
Adding a process later	a dimension value in the maintainer UI, no release	new subcategory + explanation + release
Reporting granularity	per branch (2A / 2B / 2C / 2H)	per process
Reduction-plan initiatives	per branch — cement and glass share a bucket	per process, exact
Explanation guides	4	11
Installed deployments	the four existing subcategories need rework via the maintainer screens	purely additive

The costs are real on both sides and neither is free:

• A collapses the per-process breakdown in reports (carbon_inventory_subtotals_view aggregates by subcategory) and blunts initiative targeting, because getSuggestedReductionPlan filters only by subcategoryId. ReductionPlanInitiative already has dimensionValue1Id / dimensionValue2Id columns, so scoping initiatives to the process later is a query change, not a schema change.
• B keeps that granularity but grows Alcance 1 to 18 subcategories, 11 of them industrial. useSubcategoryPreselectionData lists every subcategory of the methodology (recommended ones merely sort first), so every user — a bakery, a law firm — scrolls past all of them.

Recommendation: A, and this PR is the alternative. A matches the IPCC hierarchy, keeps the subcategory count flat, and turns "there is nowhere to put my process" into a maintainer-screen edit rather than a recurring release. This PR is the right call if the priority is not touching data that already exists in RD/demo — it is purely additive, so no existing line, factor or initiative changes shape.

Worth naming its long-term cost honestly: it leaves mixed levels of the IPPU tree side by side as siblings. Cemento (leaf 2A1) sits next to Química (branch 2B) and Cerámica y otros carbonatos (leaf 2A4) — a distinction that is invisible to a user and permanent once inventories reference it.

These two PRs are mutually exclusive — merge one and close the other. Option A: #608

What this PR does

Keeps Cemento, Vidrio, Acero and Cinc exactly as they are and adds seven subcategories:

Subcategory	IPPU	Dimension	Factors
Procesos industriales - Cal	2A2	Tipo de cal	3
Procesos industriales - Aluminio	2C3	Tecnología de celda	2
Procesos industriales - Química	2B	Proceso (16 values)	none
Procesos industriales - Ferroaleaciones y otros metales	2C2, 2C4, 2C5	Producto o proceso (15 values)	13
Procesos industriales - Papel y celulosa	2H1	Proceso	none
Procesos industriales - Cerámica y otros carbonatos	2A4	Proceso	none
Procesos industriales - Otros	2D, 2E, 2G, 2H2	Proceso	none

• Open-ended dimensions end in a bare Otro, matching the convention established on feat/mati/machinery-catalog-otro-option (the dimension name supplies the context, and the capture dropdown sorts that exact literal last). Closed taxonomies — lime types, aluminium cell technology — get none, since Otro carries no methodological meaning there.
• Adds reduction-plan initiatives and one explanation guide per new subcategory.
• Allowed measurement units on the seven are g, kg and ton, matching every other mass-denominated subcategory in the methodology.

Emission factors

All values are IPCC 2006 Guidelines, Volume 3 (IPPU) Tier 1 defaults, read from the published PDFs, in kg/ton:

Process	Source	Values
Cal	Table 2.4	high-calcium 750 · dolomitic 770 · hydraulic 590
Aluminio	Table 4.10	Prebake 1600 · Søderberg 1700
Ferroaleaciones	Table 4.5	2500 · 3600 · 4000 · 4800 · 1300 · 1500 · 1400 · 5000 · 1300 / 1600
Plomo	generic factors by furnace	ISF 590 · DS 250 · secondary 200 · default 520

Química, Papel y celulosa, Cerámica y otros carbonatos and Otros seed no defaults, deliberately. The IPCC values there are either per-gas (N₂O for nitric and adipic acid) or depend on the carbonate content of the specific feedstock, so a single seeded kgCO₂e figure would be misleading. Those require a custom factor, and each explanation walks through deriving it.

Dolomitic lime uses 0.77, the value IPCC Table 2.4 recommends for developing countries, rather than 0.86.

Ferrochromium is offered twice. Table 4.5 lists it as 1.3 (1.6 with sinter plant) — sintering chrome ore releases CO₂ of its own — so seeding only 1.3 understated a plant with its own sinter line by 23%. Split into Ferrocromo (sin planta de sínter) 1300 and Ferrocromo (con planta de sínter) 1600, with the guide explaining which applies. Every other value in this PR was re-read against the published IPCC PDFs and matched its table exactly; the audit of the four pre-existing subcategories, and the four corrections it produced, is in #619

⚠️ Aluminium seeds only the anode-consumption CO₂ factor, which is all Table 4.10 covers. PFC emissions from the anode effect are excluded. The explanation states this and instructs a separate line, because the seeder cannot express a per-gas split — EmissionFactor.gasDetails exists and is validated against the total, but seedEmissionFactors hardcodes it to {}.

Factor normalization moved out

The kg/ton normalization of the four existing subcategories now lives in its own PR cut from main, and this PR is based on that branch: #619

It is a behaviour change — steel and zinc process factors were understated by 1000×, and cement and glass now resolve in tonnes rather than kilograms — and it is the one part of this work that is not purely additive. It also applies regardless of which catalogue shape is chosen, which is why it was split out: it fixes main on its own. Read it first; the reasoning and the impact on existing inventories are spelled out there.

Installed deployments get it through a migration

Purely additive to the seed data, but on a database that already holds the catalogue the seed cannot install it: seedEmissionFactors runs createMany({ skipDuplicates: true }) over every factor, and skipDuplicates deduplicates nothing here — the partial unique index covers (subcategory, dimension_value_1, dimension_value_2, source), one of those dimension columns is NULL on most factors, and Postgres treats NULLs as distinct in a unique index. A reseed would duplicate every existing factor and then abort on its own row-count check, so RD and demo would never receive these subcategories.

20260824130000_add_industrial_process_subcategories installs them: subcategory rows with their explanation guides, allowed units, dimensions and values, the IPCC factors in both kg/ton and kg/kg, and the reduction-plan initiatives. The SQL is generated from tools/seed/src/data/base rather than hand-copied, so it cannot drift from the seed. Guarded by NOT EXISTS throughout, and on an empty database — an integration-test container — the category lookup matches nothing and every statement inserts zero rows.

Verified by replaying it against a seeded database inside a transaction, on top of the correction migration from #619: the resulting 70 factors and 70 dimension values across the 11 industrial subcategories match the base dataset exactly, and applying both migrations twice leaves the same state.

Verification

• pnpm format:check, pnpm lint, pnpm type-check — all clean.
• Both datasets parse against the real FullMethodologyDataSchema: 30 subcategories, 11 industrial, 66 dimension values, 33 factors, identical in base and testing.
• Checked against the constraints that only bite at seed time or in the editor: unique (subcategory, code) and (subcategory, position) on dimensions, unique (dimension, value) on values, unique (subcategory, dv1, dv2, source) on factors, every factor's rate-unit denominator present in the subcategory's allowed units, every parentValue and factor reference resolving, and one explanation file per subcategory.
• Not run: the seed itself against a live database.

Out of scope, found on the way

1. The factor seed is not idempotent. emission_factor_unique_subcategory_dims_source covers (subcategory, dimension_value_1, dimension_value_2, source), and one of those dimension columns is NULL on most rows — Postgres treats NULLs as distinct in a unique index, so createMany({ skipDuplicates: true }) deduplicates nothing. Running the seed twice duplicates those factors and then trips its own row-count check. (An earlier revision of this section claimed 25 of 255 factors were being dropped by that index; they are not — the seeded database holds all of them, counted directly.) Adding the rate unit to the index and making the NULL columns non-distinct would fix both.
2. gasDetails is never seeded. The column supports a per-gas breakdown and is validated against the total, but the seeder writes {}. Populating it would let aluminium carry its CO₂/PFC split and satisfy the per-gas reporting both ISO 14064-1 and the GHG Protocol require.
3. No new sector recommendations. Chemical subsectors still do not recommend the chemical process subcategory. Deliberately left out so the two PRs differ only in structure; worth a follow-up so the feature is discoverable rather than merely present.
4. Overlap with refactor/mati/remove-factor-propio-source. That branch relabels the custom factor source to Otro. The explanation guides here say "Factor propio", matching main today — whichever lands second needs a pass over the new guides.

CI

22 of 24 checks pass. Audit and Trivy Image Scan (api) fail on deepmerge-ts 7.1.5 (CVE-2026-40345), a transitive dependency of @prisma/client. Pre-existing on main and unrelated to seed data — already addressed on fix/mati/deepmerge-ts-cve.