CMMMech

An evidence-backed knowledge base for critical minerals and materials, with microbial extraction, transformation, and recovery mechanisms alongside their wider resource and material context.

The knowledge base provides a LinkML schema, an offline strict validator, tests, CI, and three primary-literature pilot records. Each record has a timestamped scientific review; synthetic test fixtures remain separate.

Element record	Curated microbial mechanism	Evidence boundary
Cobalt	Acidophilic consortium bioleaching of spent-battery black mass	Laboratory solubilization, not isolated cobalt product
Neodymium	Biosorption by dried Chlorella vulgaris biomass	Chemically pretreated magnet leachate; inactive biomass
Palladium	Resting Desulfovibrio desulfuricans cells reduce Pd(II)	Defined soluble salt assay; hydrogenase causality remains inferred

These are element records. Experimental salts, alloys and secondary resources are specified in mechanism context, not equated with mineral species. See the pilot assessment and record reviews.

Scope

• Distinguish elements, mineral species, mineral groups, commodities, and secondary resources rather than treating those as synonyms.
• Record criticality by jurisdiction, list name, and edition, with a source. There is no universal or timeless is_critical flag.
• Represent mechanisms such as bioleaching, biosorption, biomineralization, redox transformation, separation, recovery, and recycling.
• Mark microbial involvement explicitly; connect identified organisms and experimental context to evidence for the particular mechanism.
• Preserve uncertainty and evidence that supports, partially supports, or contradicts a mechanism. Validation is not scientific verification.

Local Checks

uv sync --extra dev --locked
just check
uv run cmmmech validate
uv run cmmmech validate path/to/record.yaml
uv run cmmmech validate --require-records

Without just, use uv run python scripts/check.py. The same command is used by CI. This local gate uses no research provider, model, credential, or network lookup after dependencies are installed. CI also runs the separate canonical artifact check, which reads the pinned public CLAW revision:

bash scripts/check_vendored_sync.sh

Each record belongs in data/records/<slug>.yaml. See the record guide and the packaged LinkML schema. Validation rejects unknown fields, duplicate YAML keys, malformed references, dangling source references, and duplicate record identifiers. An empty corpus is reported explicitly; --require-records makes it an error. The shared local/CI gate now uses this flag so accidental loss of the whole corpus fails just check.

Use the local record-review skill for new or changed scientific records. Review artifacts identify the reviewed commit and working changes, hash the record bytes, and retain unresolved limits.

Sources

Use the local source-discovery skill to investigate CMM data sources, reassess known sources, or identify coverage gaps. It produces a dated search log, source assessment and ranked curation handoff under research/sources/, keeping discovery leads separate from verified evidence.

Criticality is a designation made by a named authority at a particular time. Useful discovery sources include the USGS critical-minerals programme and the EU critical raw materials overview. These are entry points, not imported datasets or an endorsement of a frozen list. Curate microbial mechanisms from the underlying research, preserving organism, substrate, conditions, and the distinction between laboratory results and field or industrial deployment.

Integration Status

CLAW fleet admission registers CMMMech with its measured native capabilities. The 18 applicable governed artifacts are pinned in scripts/.vendored_canon_ref; CI verifies their bytes and file modes against that immutable public CLAW revision. Use CLAW's supported synchronizer for updates, never hand-edit governed copies.

The native schema keeps inline curation_history events separate from history_refs, which link canonical sidecar sessions. just new-history creates a validated sidecar without rewriting a scientific record; attach its printed path in a reviewed edit. Both canonical timestamp guards remain unchanged. History paths reject symlinks, and CI compares committed sessions with the trusted event base to prevent rewriting or deleting earlier provenance. See the history guide.

The native merge queue is configured for main through CLAW's reviewed plan/apply/check process. Its receipts and readiness evidence distinguish configuration, writer App token readiness, and actual queue execution. The PR shepherd remains manual and comment-only; its configuration and available secret names were checked without invoking its model. Other unverified fleet adapters remain disabled. No research provider is invoked by the native or canonical-byte validation gates.

The included CI handles pull requests, pushes to main, and merge_group events. Operational merge evidence is tracked in issue #3. The dated integration audit records the earlier pre-admission state, not the current deployment. Coordinated fleet convergence is tracked in CLAW issue #580.

License

Project-authored code and schemas use BSD-3-Clause. Project-authored data and narrative documentation use CC-BY-4.0. Third-party materials retain their own terms. See LICENSE, LICENSE-CODE, and LICENSE-DATA.