METHODOLOGY · DATA & EVIDENCE
Every figure is traceable.
EUDR Clear screens plots by sending them to FAO Open Foris Whisp, which intersects them with openly published datasets. No black boxes: each result names its datasets and records the engine and Whisp versions with the analysis time.
THE STACK · SIGNATURE
One plot. Named datasets. One indicative result.
Every live screening sends your exact geometry to FAO Open Foris Whisp, which intersects it with the datasets it maintains — Hansen GFC, JRC TMF, RADD and GLAD among them — and returns hectare figures, forest-type indicators and a risk rating. We show each figure under its own name, the rating, and the rule that combined them.
NO BLACK BOXES · NO HIDDEN WEIGHTING · SCROLL TO STACK
HANSEN GFC · GLAD/RADD · FAO WHISP
EC JOINT RESEARCH CENTRE · DEFORESTATION & DEGRADATION
LAYER 1 · JRC TMF
Tropical moist forest change
EC JOINT RESEARCH CENTRE · DEFORESTATION & DEGRADATION
LAYER 2 · WHISP
FAO Open Foris Whisp
FOREST INDICATORS + RISK RATING · QUERIED AT EACH LIVE SCREENING
LAYER 3 · HANSEN GFC
Tree-cover loss 2001–present
HANSEN/UMD/GOOGLE/USGS/NASA · ANNUAL LOSS YEARS · 30 M
LAYER 4 · GLAD/RADD
Post-2020 disturbance
GLAD/UMD (CC BY 4.0) · RADD/WUR (CC BY 4.0) · HECTARES INSIDE THE PLOT · CARRIED AS EVIDENCE
Indicative screening for decision-support only — not a compliance determination, certification, or legal advice. Due-diligence obligations remain with the operator (EUDR Art. 10).
DATASETS · VERSIONS · LICENCES
The whole evidence base, on one table.
| SOURCE | VERSION | WHAT IT CONTRIBUTES | UPDATE CADENCE | LICENCE / ATTRIBUTION |
|---|---|---|---|---|
| FAO Open Foris Whisp | Whisp version recorded per screening | Hectare figures, forest-type indicators (Ind_01–Ind_11) and a per-commodity risk rating | Queried at each live screening | FAO Open Foris — open methodology; attribution per licence terms |
| Hansen Global Forest Change | Not exposed by the Whisp API | Tree-cover loss year 2001–present, 30 m pixels — hectares-by-year inside the plot | Annual release | Hansen/UMD/Google/USGS/NASA |
| GLAD & RADD disturbance | Not exposed by the Whisp API | Post-2020 disturbance in hectares inside the plot as submitted (RADD_after_2020, GLAD-L_after_2020, GLAD-S2_after_2020) — carried as evidence and reflected in Whisp’s post-2020 disturbance flag (Ind_04) | Rolling updates | GLAD/UMD (CC BY 4.0) · RADD — Wageningen University & Research (CC BY 4.0) |
| JRC Tropical Moist Forest (JRC TMF) | Not exposed by the Whisp API | Tropical moist-forest deforestation and degradation after 2020 (TMF_def_after_2020, TMF_deg_after_2020), in hectares inside the plot — evidence, and an input to Whisp’s forest indicators | Annual release | European Commission Joint Research Centre — reached through the Whisp API |
| Copernicus Sentinel-1 / Sentinel-2 | Not queried directly | The source imagery behind RADD (Sentinel-1) and GLAD-S2 (Sentinel-2) alerts. No imagery is displayed. | Per sensor revisit | Contains modified Copernicus Sentinel data (ESA) |
| Logging concessions (World Resources Institute) | Not exposed by the Whisp API | Hectares of the plot inside a mapped logging or managed-forest concession (GFW_logging_before_2020) — carried as evidence (the logging-concession flag), never used in the verdict | Per WRI update | WRI — CC BY 4.0 (Indonesia and Gabon data excepted; see the attribution page) |
| National layers (Brazil, Colombia, Côte d’Ivoire, Cameroon) | Returned only for plots inside those countries | Brazil PRODES/DETER, MapBiomas, INPE; Colombia IDEAM; Côte d’Ivoire BNETD cocoa map; Cameroon — carried as evidence; Whisp folds them into its risk rating | Per national release | Per source — see the attribution page |
Every screening records the Whisp engine version and the UTC time the analysis ran, and keeps the raw response verbatim, so any figure can be traced back to the response that produced it. The Whisp API exposes no per-dataset release identifier, so a screening cannot pin a Hansen or GLAD release, and a re-screen later may return different figures. Full attribution and licence notes →
KNOWN ERROR RATES
Satellite data carries error. We say so.
These datasets carry published false-positive and false-negative rates. We surface that honestly rather than rounding it away — it is why our vocabulary says “no signal detected”, never “clean” or “compliant”.
The per-dataset accuracy figures will be added from the cited validation publications (UMD for Hansen GFC and GLAD; Wageningen University & Research for RADD). Until then, treat dataset accuracy as unquantified.
HANSEN GFC · TREE-COVER LOSS
Published validation studies report both false positives (loss flagged where none occurred) and false negatives (real loss missed).
GLAD / RADD · POST-2020 DISTURBANCE
Alert-based disturbance layers trade precision for speed: a share of alerts are false alarms, and some loss events are missed between passes.
WHISP · INDICATORS & RISK RATING
Whisp’s indicators and rating inherit the error characteristics of the underlying datasets they summarize.
HOW A VERDICT IS COMPOSED
Deterministic steps, fixed vocabulary.
- 01
Geometry validated before anything is sent — latitude and longitude must be finite and in range, and a boundary must have at least four positions (an unclosed ring is closed for you). A plot that fails here is reported to you, not guessed at.
- 02
FAO Open Foris Whisp is queried for the plot. It returns tree-cover-loss hectares by year, forest-type indicators at the 31 December 2020 cut-off, and its own risk rating for the commodity — for cattle, which Whisp does not yet rate, the most severe of its published crop and timber ratings is used.
- 03
Forest type at the cut-off decides the branch: primary forest is always assessed strictly; a mapped planted plantation with no primary forest treats post-2020 loss as timber harvest; land that was not forest at the cut-off follows Whisp’s rating, with the loss shown as evidence.
- 04
Post-2020 tree-cover loss (2021 onward) is set against the hectare thresholds and combined with Whisp’s risk rating — the more severe of the two stands, so a verdict is never more reassuring than Whisp itself. On primary forest, and on other land that was forest at the cut-off, Whisp flagging post-2020 disturbance lifts a no-signal result to review recommended. On a mapped planted plantation with no primary forest that floor does not apply: the verdict follows Whisp’s rating and the loss is reported as evidence.
- 05
Every field the verdict depends on must be present and well-formed — the hectare figure, the unit, the forest-type indicators and the risk rating. An absent figure is a contract violation and the screening fails; it is never read as 0 ha.
- 06
Result composed → verdict chip (decision-support vocabulary), the plain-language reason for it, every figure, the named source layers and the engine version — then hashed into the evidence vault.
THRESHOLDS · CURRENT ENGINE
- 0 ha post-2020 tree-cover loss inside the plotNO SIGNAL DETECTED
- ≤ 2 ha post-2020 lossREVIEW RECOMMENDED
- > 2 ha post-2020 lossELEVATED SIGNAL
The commodity you declare changes the reading. Whisp scores a plot against the commodity claimed on it — perennial crops, annual crops and timber each have their own rating — so the same land can read differently depending on what you say it produces. Intact forest with no disturbance is an ordinary timber origin but an unexplained one for cocoa or soy, and the verdict follows. Declare the commodity you are actually placing on the market.
Thresholds apply to tree-cover loss inside the plot from 2021 onward — the EUDR cut-off is 31 December 2020, and earlier loss is shown as evidence but does not drive the verdict. The threshold result is then combined with Whisp’s own risk rating for the commodity, and the more severe of the two stands. Outside the planted-plantation branch, Whisp flagging post-2020 disturbance on land that was forest at the cut-off can never read as no signal. Every figure, the layer it came from and the engine version are shown alongside the verdict.
DEMO MODE
While demo mode is switched on, the engine runs a documented deterministic demo kernel so the workflow can be evaluated end-to-end. Demo results are always watermarked DEMO. When live mode is on, a screening the live engine cannot complete is reported as failed — it is never replaced by a demo result.
CUSTODY HASH · PUBLIC VERIFY ENDPOINT
Every result carries a tamper-evident fingerprint.
Each screening is logged with a SHA-256 custody hash computed over the engine version, the threshold rule, the geometry as submitted, the Whisp version and analysis time, every figure returned and the verdict. Paste any hash at the public verify page to confirm exactly what was rendered and when — without exposing customer identity.
SAMPLE DDS EVIDENCE PACK · DEMO
What your evidence pack looks like.
COVER · PLOT AS SCREENED
- PLOT 04 · COCOA · POLYGON · 12.4 HA
- DECLARED COMMODITY: COCOA
- GEOMETRY AS SUBMITTED · WGS84
SCREENING FIGURES
- POST-2020 LOSS 1.8 ha · HANSEN GFC
- DISTURBANCE · EVIDENCE
- VERDICT REVIEW RECOMMENDED + REASON
VERSIONS & VERIFY
- ENGINE 2.2.0 · WHISP VERSION RECORDED
- ANALYSIS TIME · UTC
- HASH 8f3a…c21e · /VERIFY REFERENCE
PACK CONTENTS
- ·PLOT AS SCREENED — geometry as submitted, WGS84
- ·VERDICT + REASON — fixed vocabulary
- ·POST-2020 LOSS BY YEAR — every figure named by its dataset
- ·ENGINE + WHISP VERSIONS — with the UTC analysis time
- ·CUSTODY HASH + /VERIFY REFERENCE — and the attribution block
Engine and Whisp version recorded, exportable, and yours to review before anything is signed.
Create your own — start freeIndicative screening for decision-support only — not a compliance determination, certification, or legal advice. Due-diligence obligations remain with the operator (EUDR Art. 10).
INDEPENDENCE & ATTRIBUTION
EUDR Clear is an independent commercial product. It is not affiliated with, sponsored, or endorsed by the European Commission, the European Union, the FAO, or any dataset provider. Regulation (EU) 2023/1115 and its amendments are the only authoritative sources; where this site simplifies, the regulation prevails.
READ FULL ATTRIBUTION →See it on your own plot.
Free scope-checker — one plot, instant demo-engine screening, no card required.
Run the scope-checkerIndicative screening for decision-support only — not a compliance determination.
ENGINE CHANGELOG
When the engine changes, it changes here first.
| VERSION | DATE | THRESHOLD RULE | DATASETS | NOTES |
|---|---|---|---|---|
| v1.0.0 | 2026-09-21 | no signal ≤0 ha · review ≤2 ha · elevated >2 ha | Whisp live + demo kernel | Initial pin — verdict composition as published on /methodology |
| v1.1.0 | 2026-09-21 | no signal ≤0 ha · review ≤2 ha · elevated >2 ha | demo convergence: Hansen GFC + GLAD alerts + JRC TMF (demo-correlated) | Layer agreement matrix added; verdict composition unchanged |
| v2.0.0 | 2026-09-22 | post-2020 loss only — no signal ≤0 ha · review ≤2 ha · elevated >2 ha, combined with the Whisp risk column for the commodity | live: FAO Open Foris Whisp (engine 3.0.0a17) — GFC_loss_after_2020, Ind_01/03/04/05/06/07, risk_pcrop/acrop/timber | Live kernel rebuilt against the real Whisp contract and VERDICT COMPOSITION CHANGED, hence the major bump. Verdict is the more severe of the post-2020 hectare thresholds and Whisp's own risk rating, so a result is never more reassuring than Whisp. Land not forest at the 31 Dec 2020 cut-off follows Whisp's rating and the loss is reported as evidence only. Land that was forest at the cut-off and holds primary forest is always assessed strictly; a mapped planted plantation with no primary forest is treated as timber harvest. Only loss from 2021 onward feeds the verdict (the cut-off is 31 December 2020); the headline figure is Whisp's GFC_loss_after_2020 and the per-year series from 2021 is stored as evidence. An absent field is a contract violation, never 0 ha. Rows written under 1.x are unaffected and still verify against the engine version recorded on them. |
| v2.1.0 | 2026-09-22 | unchanged from 2.0.0 — post-2020 loss thresholds combined with the Whisp risk column for the commodity | adds the producer country's own official layers where Whisp covers them: Brazil (PRODES and DETER deforestation, MapBiomas, INPE), Colombia (IDEAM), Côte d'Ivoire (BNETD cocoa map), Cameroon | National layers are now requested on every screening. Whisp returns them only for plots inside those countries, and it also feeds them into the risk column it returns — so a plot in Brazil, Colombia, Côte d'Ivoire or Cameroon can be rated differently than under 2.0.0, using its own country's official record rather than global datasets alone. Our verdict composition is unchanged. The national figures are carried in the evidence beside the global ones and never summed with them. Rows written under 2.0.0 and earlier are unaffected and still verify against the engine version recorded on them. |
| v2.2.0 | 2026-09-23 | unchanged — post-2020 loss thresholds combined with the Whisp risk column for the declared commodity | unchanged from 2.1.0 — FAO Open Foris Whisp with the national layers | Oil palm is now scored on Whisp's perennial-crop rating (risk_pcrop), as documented. Under 2.0.0 and 2.1.0 a key mismatch sent every oil palm screening to the worst-of-three rule meant for cattle; on our 500-plot test corpus this moves 4 of 46 oil palm verdicts, each toward Whisp's own rating. 'Other' now takes the worst of the three ratings explicitly, as cattle does. The declared commodity is recorded with each screening and sealed into its custody hash, and it is published on /verify, because it changes the verdict on identical land. The rating the verdict was combined with is stored by name beside the verdict reason. The country obligation carried in the evidence now cites Implementing Regulation (EU) 2025/1093 as written — one Annex with a low-risk and a high-risk list, simplified due diligence under Article 13 and its Article 13(2) exception, and checks on operators (and non-SME downstream operators and traders) plus, for high risk, 9% of the quantity. Rows written under 2.1.0 and earlier are unaffected and still verify against the engine version recorded on them. |
Every screening’s custody hash pins the engine version it was rendered under — past results never silently change.