LIVING SYSTEMS INTELLIGENCE
Learning intelligence

Learning

Learning Intelligence helps the system track what was expected, what was observed, what was learned, and whether confidence should change. This is not a live impact reporting system — it is a structured intelligence layer for learning from outcomes, using example learning records for the existing proof cases.

How the loop works
Decision
Expected outcome
Observed outcome
Learning record
Confidence update

The loop feeds back: what is learned can update confidence and improve the next decision.

Amazon — learning examples

Protection status alone is not enough

PartlySupported
Expected

Protected areas may reduce conversion and support carbon and habitat.

Observed

Outcomes are context-dependent; governance and enforcement appear decisive.

Learned

Protection status alone is not enough — governance, enforcement and local context shape outcomes.

Decision implication

Treat protected areas as necessary but not sufficient; weight governance alongside designation.

Confidence MediumMedium(Unchanged)Review ReviewedWWF_AMAZONRAISG
Data gap: Quantitative effect sizes vary by region.

Rights and governance are key conditions

Supported
Expected

Indigenous stewardship may support forest protection and integrity.

Observed

Strong association in many regions, varying with recognition and pressure.

Learned

Legal recognition, rights and territorial protection are important contextual conditions for outcomes.

Decision implication

Support rights recognition and governance as part of the pathway, not just designation.

Confidence MediumMedium(Unchanged)Review ReviewedRAISGWWF_AMAZON

Monitoring needs a response attached

PartlySupported
Expected

Monitoring may improve early detection of loss.

Observed

Detection is strong, but does not by itself create enforcement or restoration.

Learned

Monitoring is most useful when connected to response capacity and governance.

Decision implication

Pair monitoring investments with response capacity to realise value.

Confidence MediumMedium(Unchanged)Review ReviewedINPEMAPBIOMAS
Data gap: Response capacity not modelled.

Restoration is long-term, not a substitute for protection

NotYetObserved
Expected

Restoration may rebuild carbon and habitat over time.

Observed

Recovery is slow and depends on method, history and future protection; not yet observable here.

Learned

Restoration is a long-term resilience pathway, not an immediate substitute for protecting intact ecosystems.

Decision implication

Prioritise protecting intact forest first; treat restoration as a slower, complementary pathway.

Confidence MediumLow(Decreased)Review DraftAMAZON_INSTITUTIONAL
Data gap: No observed time-series in the system yet.
Pollination & the food system — learning examples

Pesticide reduction works best with habitat

PartlySupported
Expected

Reducing pesticide pressure may support pollinator health and pollination.

Observed

Beneficial direction, conditioned by pesticide type, exposure, habitat and management.

Learned

Pesticide reduction is stronger when combined with habitat and landscape-level measures.

Decision implication

Combine pesticide reduction with habitat measures rather than treating it as a standalone fix.

Confidence MediumMedium(Unchanged)Review ReviewedIPBES
Data gap: Field vs laboratory effect sizes differ.

Habitat quality matters more than area

Supported
Expected

Pollinator habitat may support pollinator populations and pollination.

Observed

Outcomes appear driven by habitat quality, connectivity and plant diversity.

Learned

Habitat interventions should focus on quality and connectivity, not only area.

Decision implication

Design habitat for quality and connectivity; measure value, not hectares alone.

Confidence MediumMedium(Unchanged)Review ReviewedIPBES
Confidence updates
Restoration outcomes proved slower and more condition-dependent than initially assumed; no observed time-series yet.EO_RESTORATION · MediumLow
Confidence held, but the limiting factor shifted from detection to enforcement and response capacity.DS_MONITORING · MediumMedium
Data gaps & limitations
  • Outcomes vary by enforcement and governance quality.
  • Outcomes vary by legal recognition and governance context.
  • Detection alone does not create enforcement.
  • Recovery is slow and condition-dependent.
  • Effects depend on pesticide type and farm management.
  • Area alone is a weak proxy for habitat value.
  • Quantitative effect sizes vary by region and dataset.
  • Causal attribution is difficult; context varies.
  • Response capacity is not modelled here.
  • No observed time-series in the system yet.
  • Field vs laboratory effect sizes differ.
  • Wild vs managed pollinator dynamics not fully mapped.
  • Quantitative effect sizes vary by region.
  • Response capacity not modelled.