Download the exact captured file
---
name: emem-eudr-due-diligence
description: Produces a signed, Annex II shaped Due Diligence Statement under the EU Deforestation Regulation (EU 2023/1115) for one or more production plots with emem, and explains what it proves and what it leaves to the operator. Checks each plot against the JRC GFC2020 forest baseline and Hansen Global Forest Change loss year at the 2020-12-31 cut-off, cross-checks JRC TMF, and grades how well the inputs support the verdict. Use when a supplier, importer or auditor asks whether a cocoa, coffee, soy, palm, rubber, cattle or wood plot is deforestation-free, or needs per-cell evidence for an Article 12 record.
---
# emem-eudr-due-diligence
> **Network use.** The commands in this skill send and receive JSON (and, where a step says so, an image or a raster file) to `https://emem.dev` only, the service the plugin's MCP server connects to. Nothing they download is executed. This skill ships no scripts. Its commands read no environment variable, key or credential, and the only local files they read are ones an earlier step saved from `https://emem.dev`.
`POST /v1/eudr_dds` (MCP `emem_eudr_dds`) evaluates plots against the
Regulation's forest definition (10 % canopy, 0.5 ha, 5 m) and returns a
signed envelope 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)), and
the response says so in `legality_disclaimer` and
`degradation_disclaimer`. Pass both on.
## The request
Call `emem_eudr_dds` with:
```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}]}
```
`emem_eudr_dds` is outside the short tool list the plugin's MCP server
shows. The server runs it by name through `tools/call`. If your host
calls only listed tools, use the REST route.
Save the result's JSON text as `dds.json`. The call can take minutes,
so send it with a `task` param and poll `tasks/get`. A statement larger
than the MCP wire budget (about 24 KB) is refused, not cut, because a
partial statement still looks whole. When that happens, or when you have
no MCP, send the same body to `POST https://emem.dev/v1/eudr_dds` and
allow up to 300 s for the answer.
Per plot: `geometry_geojson` (a Polygon for plots over 4 ha; a Point is
allowed for 4 ha or less, except cattle, per Article 2(28); or
`{"bbox":[w,s,e,n]}`), `country_of_production` (ISO 3166-1 alpha-3),
`commodity_hs` (HS-6 or longer, checked against Annex I) and
`quantity_kg`. Optional: `request_visual_evidence: true` adds a
year-by-year Sentinel-2 NDVI and Sentinel-1 backscatter timeline with
scene links, and several minutes of upstream reads. Top level:
`cut_off_date` (default `2020-12-31`), `forest_baseline_override`
(`jrc_gfc2020_v4`, `hansen_only`, `both`; `jrc_gfc2020_v3` is still accepted), `max_cells_per_plot`
(auto-derived from polygon area when omitted). The full schema is
`GET https://emem.dev/v1/schemas/eudr_dds.json`.
A cold plot can take minutes and can fail with `source_fetch_failed`
naming the slow tile. Warm it first with a recall on one cell of the
plot:
1. Call `emem_locate` with the plot's coordinates, and keep the
`cell64` it returns:
```json
{"lat": 6.8276, "lng": -5.5471}
```
2. Call `emem_recall` with that cell and the forest bands. Read `band`
and `value` of each entry in `facts[]`:
```json
{"cell": "<the cell64 from step 1>",
"bands": ["jrc_gfc2020.forest_2020", "forest_change.lossyear", "forest_change.treecover2000"]}
```
Over REST, the same bodies go to `POST https://emem.dev/v1/locate` and
`POST https://emem.dev/v1/recall`.
## Reading the response
```sh
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**: `pass`, `fail`, `not_in_scope` (cleared at or
before the cut-off, so not forest at the cut-off), `indeterminate`, or
`below_mmu` (failing area under the 0.5 ha floor, counted as
compliant). `per_cell_verdicts` carries the signed fact cids behind
each; quote them in the Article 12 record.
- **`forest_baseline_dataset`**: the dataset versions actually read, by
name and source. `forest_baseline` names the JRC version its facts
were read from (`jrc_gfc2020_v4`), and the fact derivations carry it
too (`jrc_gfc2020_v4_pixel@1`); quote the versions from this field.
- **`tmf_cross_check`**: JRC TMF deforestation year read on every cell
and compared with Hansen after the cut-off (`both`, `hansen_only`,
`tmf_only`, `neither`, `agreement`). It is not counted in the verdict.
A `tmf_only` loss sets `review_required: true`, because TMF often sees
degradation-to-clearance that Hansen misses.
- **`verdict_support.level`**: `strong`, `moderate` or `weak`, from how
complete the inputs are (share of cells with both baselines, TMF
agreement, borderline canopy, and sample spacing against the 71 m side
of a 0.5 ha square), each shortfall listed in `limits`. It is not a
calibrated probability.
- **`traces_nt_envelope`**: the TRACES NT shaped statement.
`statementOfComplianceReviewRequired` says when a person must look
first. `ddsReferenceNumber` is an emem placeholder; TRACES NT issues
the real one.
A statement is held for review when any plot has
`tmf_cross_check.review_required`, when any plot has
`verdict_support.level` `weak`, or when the verdict ran without the JRC
baseline (`forest_baseline_computed` is `hansen_only_jrc_unavailable`).
The top-level `statement_of_compliance_signable` and the envelope's
`statementOfComplianceSignable` use one rule, so they always agree: each
is true only for an overall `pass` that is not held for review.
`statement_of_compliance_held_because` (envelope:
`statementOfComplianceHeldBecause`) lists every cause that applies:
`verdict_<label>` when the overall verdict is not `pass`,
`tmf_cross_check_review_required`, `verdict_support_weak` and
`jrc_baseline_unavailable_hansen_only`. It is empty when the statement
is signable.
## Recorded on 2026-09-28
The Point plot above (cocoa, Côte d'Ivoire) failed once with
`source_fetch_failed` on a cold Hansen tile, then answered in a later
call:
```json
{"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": "strong", "limits": [],
"tmf": {"agreement": null, "review_required": false}}]}
```
The one cell read `jrc_forest_2020: 0`, `hansen_treecover_2000: 20`,
`hansen_lossyear: 0`, with three fact cids, and the receipt verified
offline. Two things in that answer to pass on rather than smooth over:
TMF read no cells (`cells_read: 0`, `agreement: null`), and
`verdict_support` stayed `strong` anyway, because an unread TMF does not
lower the level. And one Point cell is about 9.5 m of a plot of up to
4 ha.
## Verify and keep
The response carries a signed `receipt` over the per-cell fact cids. On
a large plot it lists the first ones by cid and says so
(`receipt_fact_cids_capped`, `receipt_fact_cids_total`). Every sampled
cell's facts decide the verdict and are stored, because each plot's
`per_cell_rows_cid` addresses all of them, inlined or not
(`facts_computed` and `facts_persisted`, per plot and in total). Every
fact cid in the response, and every one behind `per_cell_rows_cid`,
resolves at `/v1/facts/<cid>`.
### Per-cell rows
Each plot carries `per_cell_verdicts`, `per_cell_rows_cid` and the
paging fields.
- `per_cell_detail`: `rows` (default) inlines `per_cell_verdicts`.
`anchor` leaves the rows out and keeps `per_cell_rows_cid`. Ask again
with `rows` to expand it.
- `per_cell_offset`: send it to page the rows in canonical order, one
inline cap at a time. Follow `per_cell_next_offset` until it is
`null`.
- `per_cell_rows_cid`: blake3 hex over one line per sampled cell,
`<cell>|<verdict>|<fact_cids joined by ,>` and a newline, over every
cell. The joined pages recompute it. It is not part of the signed
receipt, but the fact cids inside it are signed.
- Coverage is `sampled_polygon_fraction` (sampled area over polygon
area). Never use a count of the rows as coverage. Without
`per_cell_offset`, the inlined rows over the cap are a sample:
failing cells first, then the first of the rest.
Verify the receipt offline with
[`emem-verify-receipt`](../emem-verify-receipt/SKILL.md), and keep the
fact cids (or a bundle of them, `POST /v1/memory_bundle
{"fact_cids":[...]}`) in the operator's record. For a lab certificate or
land title that belongs in the same file, see
[`emem-document-evidence`](../emem-document-evidence/SKILL.md).
## What to tell the user
- The verdict, the plot's `verdict_support.level` and its `limits`.
- Whether review is required, and why (`tmf_cross_check.review_cells`).
- A Point geometry samples one cell (about 9.5 m) of a plot of up to
4 ha; the response adds a `sampling_caveat` saying so.
- Legality and degradation are not assessed. The obligations apply from
30 December 2026 for large operators and 30 June 2027 for micro and
small enterprises (`regulation_status_note`); the cut-off date is
unchanged.