Automation and guarded writes

This page is for operators with complete input files, not a first-use tutorial. Complete identity checks and local validation first. Filenames below represent task-specific artifacts, not downloads provided here. Do not run an entire command block as an exercise.

This page uses the globally installed tiangong-lca short command; see getting started for installation.

From preparation to verification

1. Define the target account, dataset IDs/versions, allowed field changes, and output directory.
2. Validate complete inputs; resolve failed rows, evidence gaps, and blockers first.
3. Use --dry-run where the command supports it and save the report. A dry run is not a write or proof of access or final results.
4. Present the actual scope, risks, and approval details to the operator. Obtain explicit approval before real writes.
5. Independently read back identity, owner, content, and counts. Accepted, queued, and partially completed requests are not completed acceptance.

Preview a draft write

tiangong-lca process save-draft --input ./patched-processes.jsonl --out-dir ./process-save-draft --dry-run --json
tiangong-lca lifecyclemodel validate-build --run-dir ./lifecyclemodel-run --json

Use status, counts, issues, files, and artifact paths to decide the next step, not terminal prose alone. For actual batch process-draft writes, provide --target-user-id and retain authentication, draft-owner checks, and readback. This guide does not provide a write approval to copy blindly.

Quality checks before writing

For data-production pipelines, the CLI also provides quality-gate commands that are useful before writing data, publishing data, or handing work to a reviewer:

tiangong-lca process identity-preflight --input ./process-preflight.json --out-dir ./process-preflight --json
tiangong-lca flow identity-preflight --input ./flow-preflight.json --out-dir ./flow-preflight --json
tiangong-lca process build-plan validate --input ./process-build-plan.json --out-dir ./process-build-plan --json
tiangong-lca flow build-plan validate --input ./flow-build-plan.json --out-dir ./flow-build-plan --json
tiangong-lca publish run --input ./publish-request.json --dry-run --json

• identity-preflight compares a target process or flow with candidate data and reports whether automation can reuse it, should route it to manual review, or should block new creation.
• build-plan validate checks whether a process or flow build plan includes identity decisions, evidence bindings, naming plans, unit_of_analysis decisions, and the required reference-flow or flow-property fields.
• dataset evidence-search plan/run plans field-level public evidence retrieval and records external search results; the CLI owns the query matrix, budget, result normalization, and evidence declaration artifacts, while human or agent workflows still own source judgement.
• publish run --dry-run reports publish ruleset results before a real write or publish step.

These commands write machine-readable reports under outputs/ or reports/ in the selected --out-dir. For automation, read fields such as status, blockers, issues, files, and artifact paths instead of relying on terminal text.

Data and evidence boundaries

Schema-invalid rows remain blocked before writes and are recorded in failures.jsonl or the validation report.

dataset save-draft blocks account-local drafts of reference-only foundations such as Unit Groups and Flow Properties by default. --allow-account-local-support or TIANGONG_ALLOW_ACCOUNT_LOCAL_SUPPORT=1 is only for explicitly authorized migration or repair, not a routine bypass.

dataset classification apply --type location may create a missing parent and field only when target_path explicitly identifies a schema-derived location field. Ambiguous paths remain blocked. dataset evidence-search run saves plans, normalized results, reports, and necessary evidence declarations; it does not replace human source judgment.

For LCI/LCIA packages, read Data publishing. For account cleanup and repair, follow Data maintenance; do not bypass safeguards through raw SQL or privileged keys.