vortx-ai/emem/emem-eudr-due-diligence

emem-eudr-due-diligence

Network use. The commands in this skill send JSON to https://emem.dev and receive JSON from it. https://emem.dev is the service that the MCP server of the plugin connects to. The commands do not execute what they download. This skill ships no scripts. Its commands read no environment variable, key or credential. The only local files that they read are files that an earlier step saved from https://emem.dev.

The endpoint tests each plot against the forest definition of the Regulation: 10 % canopy, 0.5 ha and 5 m. It returns a signed envelope that an operator can keep as evidence. It covers the geolocation and deforestation parts of Annex II.

It does not cover legality (Article 9(1)(b)) or forest degradation (Article 2(7)). The response says so in legality_disclaimer and degradation_disclaimer. Give both to the user.

Step 1: warm a cold plot

A cold plot can take minutes. It can also fail with source_fetch_failed, which names the slow tile. To prevent this, read the forest bands on one cell of the plot first.

Get the cell64 of the plot point. Send this body to POST https://emem.dev/v1/locate and read cell64 in the answer:

{"lat": 6.8276, "lng": -5.5471}

This returned defi.zb44d.zaedf.bujI on 7 November 2026. Read the three forest bands at that cell. Send this body to POST https://emem.dev/v1/recall:

{"cell": "defi.zb44d.zaedf.bujI", "bands": ["jrc_gfc2020.forest_2020", "forest_change.lossyear", "forest_change.treecover2000"]}

For each entry of facts[] in the answer, read band, value and unit.

On 7 November 2026 it returned jrc_gfc2020.forest_2020 0, forest_change.lossyear 0 and forest_change.treecover2000 20. The MCP tools emem_locate and emem_recall take the same bodies.

For a Point, the statement samples one cell in a box that reaches 0.0001 degrees on each side of the point. That cell can be a neighbour of the cell from locate. On 7 November 2026 the statement read defi.zb44d.zae9f.bujI. The warm-up still helps, because it loads the source tiles, and neighbour cells share them.

Step 2: send the request

Send this body to POST https://emem.dev/v1/eudr_dds and save the answer as dds.json:

{
  "operator": {"name": "Example Cocoa Importer"},
  "plots": [{
    "plot_id": "CIV-001",
    "geometry_geojson": {"type": "Point", "coordinates": [-5.5471, 6.8276]},
    "country_of_production": "CIV",
    "commodity_hs": "180100",
    "quantity_kg": 1200
  }]
}

Allow up to 300 s for a cold plot. After step 1, this call answered in less than 1 s on 7 November 2026.

Over MCP, call emem_eudr_dds with the same body. The plugin's MCP server does not list this tool, but tools/call runs it by name. If your host calls only listed tools, use the REST route. A long call over MCP can use a task param and poll tasks/get. The MCP server refuses a statement larger than its wire budget (about 24 KB). It does not cut it, because a partial statement looks complete. In that case, use the REST route.

Fields for each plot

Fields at the top level

The request schema is EudrDdsReq in the OpenAPI document at GET https://emem.dev/openapi.json. The response schema is at GET https://emem.dev/v1/schemas/eudr_dds.json.

Step 3: read the response

jq '{verdict: .due_diligence_statement.verdict,
     signable: .traces_nt_envelope.statementOfComplianceSignable,
     review_required: .traces_nt_envelope.statementOfComplianceReviewRequired,
     baseline: .forest_baseline, datasets: .forest_baseline_dataset,
     plots: [.per_plot_results[] | {plot_id, verdict, failing_area_ha,
             support: .verdict_support.level, limits: .verdict_support.limits,
             tmf: .tmf_cross_check | {agreement, review_required, post_cutoff_loss}}]}' dds.json

Plot verdicts

per_cell_verdicts gives the signed fact cids behind each cell. Put them in the Article 12 record.

Other fields

These inputs set verdict_support.level (verdict_support in lib.rs):

A value between the two columns gives moderate. Three more cases give moderate:

• The plot has one sample cell, for example a Point.
• The plot is less than 0.5 ha.
• TMF read no cell, and the plot is within 30 degrees of the equator.

When the endpoint holds the statement for review

The endpoint holds the statement for review when one of these is true:

• A plot has tmf_cross_check.review_required: true.
• A plot has verdict_support.level weak.
• The JRC baseline was not available. forest_baseline_computed is then

hansen_only_jrc_unavailable.

statement_of_compliance_signable at the top level and statementOfComplianceSignable in the envelope use one rule, so they always agree. Each is true only when the overall verdict is pass and the endpoint does not hold the statement for review.

statement_of_compliance_held_because (in the envelope: statementOfComplianceHeldBecause) lists each cause that applies. It is empty when the statement is signable.

Result on 7 November 2026

The Point plot in step 2 (cocoa, Côte d'Ivoire) returned this at 01:06 UTC. A second call at 03:31 UTC returned the same values:

{"verdict": "pass", "signable": true, "review_required": false,
 "baseline": "jrc_gfc2020_v4",
 "datasets": [{"name": "JRC Global Forest Cover 2020", "version": "V4"},
              {"name": "Hansen Global Forest Change", "version": "v1.13 (2025)"}],
 "plots": [{"plot_id": "CIV-001", "verdict": "pass", "failing_area_ha": 0.0,
            "support": "moderate",
            "limits": [{"severity": "qualifies", "why": "one point sample: a clearance elsewhere on the plot is not seen"},
                       {"severity": "qualifies", "why": "the plot is 0.01 ha, under the 0.5 ha minimum mapping unit (Art. 2(4)), so no clearance of that size fits in it and none can fall between samples"}],
            "tmf": {"agreement": 1.0, "review_required": false,
                    "post_cutoff_loss": {"both": 0, "hansen_only": 0, "neither": 1, "tmf_only": 0}}}]}

The one cell read jrc_forest_2020 0, hansen_treecover_2000 20, hansen_lossyear 0 and jrc_tmf_deforestation_year 0, with four fact cids. The receipt signs five fact cids and verified. The datasets lines above leave out the source URLs.

On 28 September 2026 the same plot first failed with source_fetch_failed on a cold Hansen tile. A later call then answered with support strong, while TMF read no cell. The code now lowers the level to moderate for one sample cell and for an unread TMF in the tropics.

Step 4: verify and keep the evidence

The response has a signed receipt over the per-cell fact cids. On a large plot the receipt signs only the first cids, sorted by cid. The response then shows receipt_fact_cids_capped and receipt_fact_cids_total.

Verify the receipt on the server. Build the body from the saved answer:

jq -c '{receipt}' dds.json > verify_req.json

Send verify_req.json as the body of POST https://emem.dev/v1/verify_receipt. Read valid, signature_valid and merkle_proof_valid in the answer.

On 7 November 2026 all three were true. To verify offline, without trust in the responder, use `emem-verify-receipt`.

The responder stores the facts of every sampled cell, also the cells that the response does not show. facts_computed and facts_persisted give the counts, for each plot and in total. Each fact cid in the response resolves at GET https://emem.dev/v1/facts/<cid>, and so does each cid behind per_cell_rows_cid.

Per-cell rows