ADR: design cases and regional LCI scenarios without duplicating plants¶

Status: accepted. Code: domains/assessment/lca/scenarios.py, the Scenario model of lca/lca.yaml, the Case model of process/flowsheet.yaml. Commands: cheme assess lca-scenarios, cheme assess lca-cases, cheme assess lca --case.

Context¶

A process design (e.g. the process library's 500 t/d GOX air separation unit) should be usable for several operating cases and several geographical backgrounds without copying the engineering project, and without presenting an unvalidated capacity or a swapped grid factor as a new plant. Four things must stay apart:

the technology/configuration (which physical process);
the design or operating case (capacity, specifications, operating point: which foreground inventory is valid);
the background scenario (region, year, electricity procurement, background data: how the foreground is contextualised);
evidence: the calculation maturity and design maturity of the design and model, and separately the coverage, provenance and review of the inventory. A region swap never raises the calculation maturity or the design maturity.

What already existed¶

concept	existing entity
reference design / technology	the plant repository; `lca.yaml technology`; `lci_dataset` `id`, `technology`; the process library's `process.yaml`
design / operating case	the flowsheet's `cases:` (stream states per case); `lca.yaml case` picks one
unallocated foreground	`lca/dataset.yaml` (`kind: lci_dataset`, per 1 h of operation), derived by `inventory.py`
background scenario	`lca.yaml scenarios:`: factor ids (`factors:`) or Brightway databases (`databases:`) replaced
functional unit, multi-output	`functional_unit`, `allocation.py` (allocation applied after the unallocated dataset)
coverage	gaps (never zeros), `missing:` mappings, classification, geography and year checks
staleness, publication	`lca/manifest.json` fingerprints; the source registry and publication policy
maturity	`cheme maturity` (design deliverables), independent of the LCA

So no new top-level technologies/ or scenarios/ entity is needed. The design case is a flowsheet case; the regional scenario is an lca.yaml scenario. What was missing was metadata and a per-scenario output.

Decision¶

Additive optional fields, existing files. No field is required; existing process.yaml, flowsheet.yaml, lca.yaml and lci_dataset files load and compute as before.

On a flowsheet case (process/flowsheet.yaml cases:):

field	meaning	unset
`capacity`	the declared capacity of the case, e.g. `500 t/d GOX`	not declared
`provenance`	`given`, `calculated`, `assumed` or `extrapolated`	not declared
`applicability`	`{range, source, limitations}`: an envelope with its source, not a scaling law	none claimed
`source`	what the operating point rests on	—

On a scenario (lca/lca.yaml scenarios:), besides id, title, factors, databases, source:

field	meaning	unset
`location`	the geography the background represents: makes it a regional scenario, checked against this geography	the study's location; former behaviour
`year`	the background's reference year	the study's year
`electricity`	`grid-average`, `residual-mix`, `contractual` (needs `source`: the instrument), `on-site`	not stated
`case`	a separately computed flowsheet case: the foreground changes with it	the study's case: foreground unchanged
`missing`	flow key (globs) -> why no background of this region exists: a gap with its quantity	—
`proxy`	flow key (globs) -> why the background stands in for this region: marked `PROXY`	—

Rules:

A regional scenario's contributions are checked against its own geography and year (cheme.lca-scenario-geography, -year, info -unchecked). A background kept from the baseline (the ASU's UK grid in an NL scenario) is named in a warning, never relabelled. The baseline mapping's proxy: note describes the baseline's geography: in a regional scenario it does not silence the check, and a background the scenario replaces is judged by its own status. Only the scenario's own proxy: declaration discloses a foreign background.
A scenario that cannot find a regional background declares it missing: the exchange stays in the inventory with its quantity and is a gap. A replacement factor that does not exist is an error. Nothing falls back silently to another region or to zero.
A physical change (ambient temperature, cooling, compressor work, capacity) needs its own computed flowsheet case. A scenario naming case: is a design-case scenario; one without is a background sensitivity whose foreground is the baseline's. Nothing scales a design to a capacity it was not computed for; cheme assess lca-cases only sets existing cases side by side, and a case marked extrapolated is labelled as not predictive.
Scenarios without location behave exactly as before (including the mapping's proxy marker).

One record per scenario. cheme assess lca-scenarios --write writes lca/scenarios/<scenario>.json (schema cheme/lca-scenario/v1) for the baseline and every scenario; scenarios never overwrite each other. A record holds:

id: <dataset id>/<case>/<scenario>, stable while the plant's flowsheet title, the case id and the scenario id are; scenario (kind baseline, background or design-case, geography, year, electricity, source, the replacements, missing, proxy); process (dataset id, title, technology); design_case (capacity, provenance, applicability, extrapolated);
three layers, each labelled layer:
foreground (foreground): every exchange of the unallocated dataset in its basis (1 h of operation), its relation to the baseline (baseline, unchanged, design-case) and a fingerprint of the exchanges;
linked_inventory (linked-inventory): each non-product exchange with its quantity and unit, its background in this scenario and the baseline's mapping, whether the scenario replaced it, geography, year, status, source_id and state (linked, proxy, missing, unmapped, free, released, gap, partial), and data_year where the factor declares one;
impacts (characterized-impact): per functional unit (functional_unit: product, amount, unit, reference_amount {value, unit} in the reference product's unit, and kg, the same in kg, which is null for an energy product such as exported electricity or steam), with the allocation method and shares, the categories, totals, classification and reasons, and comparable/dropped (dropped_by_category): the exchanges the baseline characterises that are gaps in this scenario. A category's total is null when such an exchange is missing or when all its contributions are gaps: a total without the electricity is not a lower footprint, and a category with no data is not zero. Otherwise a total follows core's convention: it excludes the gaps, which are listed with their quantities, and partial_categories names the categories where that is so. characterised_sum keeps the sum of what was characterised, gaps excluded, for analysis only. comparable covers the gaps only; whether the backgrounds measure the same thing is comparability below;
comparability (lci-comparison-basis.md): verdict (baseline, like-for-like, proxy, not comparable), reasons and compared (each replaced background pair with its differing or undeclared dimensions), and each linked exchange's scope (null: not declared). A scenario can be comparable (no gap) and still not comparable on its verdict, e.g. a direct-emission grid factor against a life-cycle one;
gaps, notes, warnings (the scenario's own: its characterisation and, for a design-case scenario, its own case's inventory; never another scenario's; and the study's general ones, such as the flowsheet's boundary check cheme.boundary-*), publication (the scenario's own policy) and inputs (the run manifest's file fingerprints, mapping hash, background profiles and sources, software, and a hash of the scenario definition). No time, commit or absolute path: a rerun gives the same file, and --check fails when an input, a factor, a profile, a database identity or the software changes, or when a record's scenario was removed from lca.yaml (--write deletes such a record).

The record keeps the per-hour foreground and the per-functional-unit impacts apart: an unallocated per-hour inventory is not a kg-oxygen footprint. Allocation stays the study's choice (allocation:) and is stated where it is applied.

The same rule applies outside the records. results.md shows such a scenario's cell as "not comparable (no background for ...)", or "no total" when every contribution is a gap, and marks a total "(not comparable)" when the scenario's verdict is not comparable. cheme assess lca --json and GET /api/lca add not_comparable: for each scenario and category, the dropped exchange keys, or null. results.csv keeps its numeric rows, with the gaps in gaps.csv. cheme export-static withholds lca/scenarios/** together with the other LCA result files whenever the results may not be published.

A design case's own dataset¶

A scenario record characterises a design case, but lca/dataset.yaml is the unallocated dataset of the study's case only. A plant with two computed designs exports the other one with --case (ADR 0045):

cheme assess lca --dataset --case hot   # the YAML, to stdout
cheme assess lca --write --case hot     # lca/datasets/<dataset id>-hot.yaml
cheme assess lca --check --case hot     # exit 1 when that file is missing or stale

The dataset is derived from case hot's stream states in the same way as lca/dataset.yaml, with the study's scope. Its identity names the case, so two datasets of one process never share an id, a title or a uuid:

id: <dataset id>-<case> (the case id made an identifier);
title: <title>, case <case>;
uuid: uuid5(<base uuid>, "case:<case>"), from the uuid dataset: uuid: pins, or else the one cheme derives for lca/dataset.yaml. The same base and case give the same uuid.

The reference product, functional unit, basis and dataset: documentation are the study's, except what describes the study's dataset alone: a stated review becomes unreviewed (nobody reviewed this dataset), superseded_by is dropped, and parent is the study dataset's uuid. The case id is made an identifier in lower case (--case B writes <dataset id>-b.yaml); two declared cases that give one identifier (Hot and hot) are the error cheme.lca-case-id-clash, and neither gets a dataset. The header comment says the file is generated and names the command that rewrites it. The provenance (generator, case and the sha256 of each input file) records what it was derived from. --write --case writes only this file, never over a dataset in lca/datasets/ that it did not generate for that case (cheme.lca-dataset-not-generated). --check --case reports cheme.lca-dataset-stale with the file and the hint to rerun. Run it in a kind: command gate. cheme assess lca --check without --case checks the study's outputs only, and reports a stale lca/dataset.yaml with the same finding. cheme validate does not read lca/datasets/: the check of a committed case dataset is --check --case.

A case the flowsheet does not declare exits 2 and lists the declared cases. --case naming the study's own case is the same as no --case, so lca/dataset.yaml is unchanged. Data quality (lca/manifest.json), the allocation choice, the results and the scenario records remain those of the study's case.

Why this avoids duplication¶

The engineering sources (P&IDs, flowsheet, design basis) exist once. A regional background is a few lines in lca.yaml and, where needed, factors in lca/factors.yaml; a different operating point is a flowsheet case of the same project. Consumers (the process library, the Process Atlas, Brightway jobs) read the records; no folder is copied per region or capacity.

Consequences and limits¶

cheme assess lca-scenarios is separate from cheme assess lca --write, so projects with former scenarios gain no new outputs and their lca-current gates stay as they are. Projects that publish scenarios commit lca/scenarios/*.json and check them with --check.
The packaged open data are UK (DESNZ). Regional factors for other countries must come from the project with their source and licence; until then a scenario declares missing or proxy. The core test fixture uses synthetic NL/DE factors, marked synthetic.
Findings of the shared dataset resolver (dataset: backgrounds) are not yet tagged per scenario and appear in every record.
With backend brightway, the records name activities and replaced databases; the per-exchange geography is what the activity mapping gives.
Not done here: a scale-up model, a capacity optimiser, a facility registry. A specific facility (coordinates, measured data) is not modelled: location is background applicability, not a site.

Example¶

# lca/lca.yaml
scenarios:
  - id: nl-grid-2025
    title: Netherlands grid average 2025
    location: NL
    year: 2025
    electricity: grid-average
    factors: {"desnz-ghg-2026:electricity-gb": electricity-nl}
    missing: {"product:CW": "no NL cooling water factor yet"}
    source: "<the published NL grid factor and its licence>"
  - id: hot-climate
    case: hot          # a computed flowsheet case, not a scaled one
    source: "flowsheet case hot: 40 C ambient"

cheme assess lca-scenarios                 # summary per scenario
cheme assess lca-scenarios nl-grid-2025 --json
cheme assess lca-scenarios --write         # lca/scenarios/<id>.json
cheme assess lca-scenarios --check         # exit 1 when a record is stale
cheme assess lca-cases                     # existing cases side by side, per unit of product
cheme assess lca --write --case hot        # the hot case's own dataset, lca/datasets/

The fixture tests/data/lca-scenarios-asu shows all of it on an ASU-like plant.

Publishing from the process library¶

The process library's catalog builder reads lca/scenarios/*.json of each process and exports the design case, the scenarios and their evidence; --public keeps only records whose publication.publishable is true and withholds the impact numbers of the others (geography, year, states and gaps stay, since they hold no restricted numbers).