C6 Carbon Model

C6 Carbon Model — Canonical Metric Catalog

Date: 2026-05-07 Purpose: Define the C6-native carbon model. Every methodology (Verra legacy, VM0048, VM0047, ART TREES, Plan Vivo, Gold Standard, CDM, IPCC) maps onto this catalog as a projection — a subset of metrics with methodology-specific transforms. C6 owns the data model; methodologies are configurations on top.

Pairs with: methodology-inputs-comparison.md (the upstream input survey) and app/units/registry.py (the canonical-unit enforcement layer).

Design principle: if a metric appears in ≥2 methodologies it is core and lives directly in farm_metrics. If it appears in only one methodology and is high-cardinality (per-tree, per-plot, per-pixel), it lives in a side table (carbon_measurements, plot_measurements, etc.) and is aggregated into a farm_metric row at compute time.

1. Metric layers

Three layers, top to bottom:

This catalog is the L1 list — what every farm should carry to feed any methodology. L0 sources and L2 transforms are referenced but not enumerated here.

2. Canonical metric catalog

Columns:

• code — farm_metrics.metric_type.code (matches app/units/registry.py where applicable)
• canonical unit — what value_canonical carries (matches REGISTRY[code].canonical)
• accepts — input units the converter accepts (raw side)
• pool / class — IPCC pool, LULC class, or driver category
• methodologies — which standards consume it (V7=VM0007, V15=VM0015, V48=VM0048, V47=VM0047, V9=VM0009, ART=ART TREES 2.0, PV=Plan Vivo, GS=Gold Standard, CDM=CDM A/R)
• C6 status — live (in DB today), partial (live but missing fields), gap (need to add)
• L0 source — where raw data comes from

Pool codes: AGB = above-ground biomass; BGB = below-ground; DW = dead wood; L = litter; SOC = soil organic carbon; HWP = harvested wood products. Carbon fraction (CF) and 44/12 stoichiometry applied at engine layer, not storage.

2.1 Geometry & area

2.2 Forest cover & stratification

2.3 Carbon pools — per-stratum density

2.4 Allometric inputs (per-plot, aggregated)

These live at L0 (plot_measurements) and aggregate up. Listed here for completeness — C6 must store these to defend its own AGB numbers.

2.5 Activity data — historical & projected deforestation

2.6 LULC class areas (current state)

Already live as lulc_*_ha. These feed stratification + leakage analysis.

2.7 Driver / risk layers

2.8 ARR-specific (V47, GS, CDM, PV)

2.9 Site-prep & non-CO₂ emissions

2.10 Fire & disturbance history

2.11 Leakage

2.12 Uncertainty & QA per pool

2.13 Buffer / permanence / risk

2.14 ART TREES jurisdictional (mostly project-level, but some farm contributes)

2.15 Plan Vivo non-carbon indicators

Required for issuance under PV. Not “carbon” but blocks credits if missing.

2.16 Project-level financial / output (already L2)

These belong in project_computation / vintage_ledger, listed for completeness — they should not be stored as farm_metrics (output, not input).

3. Coverage matrix — methodologies vs C6 metrics

How complete is C6 today, by methodology?

REDD coverage is the strongest area (40–47%). ARR and Plan Vivo are weakest (18–29%) — C6 has barely the framework for ARR-only projects today.

4. Gaps prioritized

The big rocks, in order of impact:

P0 — blocks current methodology accuracy

1. Carbon pools beyond AGB/BGB/SOC: add dw_density, litter_density, hwp_carbon. VM0007/V47/ART/GS all need at least DW + litter or a documented de-minimis exclusion. Without these the engine silently under-estimates and mis-applies the conservativeness deduction.
2. Allometric provenance: wood_density_t_m3, allometric_equation_id, carbon_fraction (today buried in app/methodologies/params). Audit defensibility requires per-row traceability to the equation + ρ used.
3. Per-pool uncertainty (95% CI): *_uncertainty_pct_ci95 for AGB, BGB, DW, L, SOC, AD. VM0048 and ART TREES apply quantitative uncertainty deductions at issuance — not having these means engine cannot legitimately deduct.
4. baseline_source discriminator: enum on the methodology-protocol level. Without this the engine cannot tell “VM0007 historical projection” from “VM0048 jurisdictionally allocated” and mixing logic produces silent errors.
5. Stratification table: stratum_id + forest_type_class + disturbance_class. Today agb is one number per farm; methodologies require per-stratum EFs.

P1 — unlocks ARR / V47

6. arr_eligible_area_ha, pre_project_land_use_class, pre_project_land_use_years — V47 eligibility.
7. stocking_index, matched_control_plot_id, matched_control_si_delta — V47 dynamic baseline. New paradigm; not currently representable.
8. forest_gain_ha per vintage — ART TREES enhancement category + V47 monitoring.
9. species_planted, survival_rate_pct, tree_count, forest_age_years — ARR project monitoring.

P2 — VM0048 jurisdictional ingestion

10. jurisdictional_jad_tco2e, allocated_baseline_tco2e_yr, fcbm_forest_ha_t1/t2/t3 — ingest VMD0055 jurisdictional files. Today engine fakes baseline from project rate; under V48 it must be jurisdictional.
11. Driver layer ingest: dist_to_road_m, dist_to_settlement_m, slope_pct, elevation_m — needed even under V48 because EF stratification depends on them.

P3 — Leakage + non-CO₂

12. leakage_belt_area_ha + leakage_belt_defor_ha_yr + market_leakage_factor_pct — currently project-level compliance_leakage_pct is a hand-set scalar; methodology requires per-belt monitoring.
13. fertilizer_n_kg_ha, fossil_fuel_co2_t, combustion_completeness, ch4_emission_factor — non-CO₂ project emissions required by V47/GS/CDM.

P4 — Plan Vivo path (only if pursuing PV registry)

14. livelihood_indicator_score, ecosystem_indicator_score, pra_baseline_id — non-carbon indicators required for PV issuance.

5. Recommended schema changes

5.1 New farm_metrics codes (P0–P3)

Codes to add to lookup_metric_type + app/units/registry.py:

# Pools
dw_density                       tC/ha
litter_density                   tC/ha
hwp_carbon                       tC
non_woody_biomass                tC/ha

# Allometric provenance (per-plot side table; only summary on farm)
agb_method                       enum: field_plot / gedi / vendor / ipcc_default
soc_method                       enum: tier1 / tier2 / tier3_measured

# Stratification
forest_type_class                enum
disturbance_class                enum
forest_age_years                 yr

# ARR
arr_eligible_area_ha             ha
degraded_forest_area_ha          ha
pre_project_land_use_class       enum
pre_project_land_use_years       yr
stocking_index                   dimensionless
species_planted                  text (multi)
survival_rate_pct                %
tree_count                       count
mean_height_m                    m

# Activity data extensions
degradation_ha                   ha
forest_gain_ha                   ha
jurisdictional_jad_tco2e         tCO₂e
allocated_baseline_tco2e_yr      tCO₂e/yr
fcbm_forest_ha_t1, _t2, _t3      ha

# Drivers
dist_to_road_m                   m
dist_to_settlement_m             m
dist_to_river_m                  m
dist_to_prior_clearing_m         m
slope_pct                        %
elevation_m                      m
tenure_class                     enum

# Leakage
leakage_belt_area_ha             ha
leakage_belt_defor_ha_yr         ha/yr
market_leakage_factor_pct        %

# Non-CO₂
fertilizer_n_kg_ha               kg N/ha
biomass_burning_ha               ha
fossil_fuel_co2_t                tCO₂e

# Uncertainty (per-pool)
agb_uncertainty_pct_ci95         %
bgb_uncertainty_pct_ci95         %
dw_uncertainty_pct_ci95          %
litter_uncertainty_pct_ci95      %
soc_uncertainty_pct_ci95         %
activity_data_uncertainty_pct_ci95   %
combined_uncertainty_pct_ci95    %

# ART
hfld_score                       dimensionless
nfms_emission_factor_tco2e_ha    tCO₂e/ha

# Plan Vivo
livelihood_indicator_score       score 0–100
ecosystem_indicator_score        score 0–100

5.2 New side tables (L0 raw observations)

• plot_measurements — per-tree DBH, height, species, ρ, plot_id, geom. Used for AGB/BGB derivation. Today AGB lands directly in farm_metrics with no traceability to plots.
• plot_definitions — plot polygon, area, sampling protocol, allometric equation used.
• stratum — forest type × disturbance × age class. FK from farm_metrics for per-stratum EFs.
• leakage_belt — belt geometry per project (separate from project geometry).
• control_plot_match — V47 k-NN matched control plot reference + multivariate distance + match date.
• jurisdictional_data_file — VMD0055 / ART NFMS files: jurisdiction, file hash, vintage, JAD, FCBM, risk map blob.

5.3 New project-level fields

• baseline_source enum on project_methodology_history (or project): historical_projected | jurisdictionally_allocated | dynamic_control_matched | community_PRA | national_NFMS.
• accounting_area_jurisdiction text — required when baseline_source = jurisdictional_allocated or national_NFMS.

6. Engine-side principles

1. Engine never reads value_raw. Always reads value_canonical via read_canonical_value(). Already enforced.
2. Pool list is methodology-config, not engine-hardcoded. MethodologyDef.required_inputs already does this; extend it with optional_with_de_minimis_test to model V48’s stricter rules.
3. Carbon fraction (CF) and 44/12 stoichiometry happen in the engine when projecting tC → tCO₂e. Storage stays in tC/ha. Already true in the registry — preserve this invariant.
4. Uncertainty deductions are explicit. Add a uncertainty_deduction_tco2e row to vintage_ledger so the receipt shows it. Today the ledger has gross → leakage → buffer → issuable; insert uncertainty between buffer and issuable.
5. Leakage and buffer are distinct — never collapse. vintage_ledger already has both rows; preserve.
6. baseline_source selects which calculation path the engine takes. Don’t try to unify legacy + jurisdictional in one function.

7. Open questions for the team

1. Is C6 doing field plots itself, or always consuming vendor PDFs (Canopy, ClearBlue, C2050)? If always vendor, plot_measurements becomes ingest of vendor’s plot data; if both, schema needs to support mixed.
2. Will C6 ever pursue VM0047 (ARR) or only stay on REDD? P1 gaps are big.
3. Is Plan Vivo on the roadmap? P4 work is otherwise wasted.
4. For VM0048: does C6 have the relationship with Verra to ingest jurisdictional files directly, or will it consume them via a partner (e.g. Everland, Wildlife Works)?
5. Carbon fraction default: today app/methodologies/params likely uses 0.47 — confirm vs 0.5 (older IPCC default still common in vendor reports).

8. Sources cross-reference

This catalog is a synthesis of:

• methodology-inputs-comparison.md (the 112-row methodology survey, sibling file)
• app/units/registry.py (canonical-unit invariants)
• app/methodologies/{vm0007,vm0048,art_trees,verra_old}.py (current REQUIRED_INPUTS)
• migrations/versions/{868f5ac927ad,d8e1c4a3b9f2,d6a3e9b1c4f7}.py (seed data for current 48 metric_types)
• docs/needs/units-design.md (the dual-write design)
• IPCC 2006 GL Vol. 4 + 2019 Refinement
• Verra VM0007 v1.8, VM0009 v3.0, VM0015 v1.2, VM0047 v1.1, VM0048 v1.0
• ART TREES 2.0 standard
• Plan Vivo PV Climate Methodology Requirements 1.0
• Gold Standard LUF A/R v2.0

The existing 48 lookup_metric_type rows already cover the REDD baseline well. The biggest single-decision lever is whether C6 commits to ARR + Plan Vivo paths — that determines whether ~30 of the 60 gaps get worked.