More farms and sensors on a carbon platform can tempt a company to present every observed methane decline as its own climate result. Yet measured concentration, calculated emissions, reductions against a baseline, and AI Safety Korea's contribution are different accounting objects.

A credible figure must disclose what was measured, which counterfactual was used, what caused the change, and who may claim the result. The objective is not the largest number, but a system that an independent reviewer can reproduce.

Keep four ledgers

Separate: (1) AI Safety Korea's own Scope 1, 2 and 3 inventory; (2) the platform's monitored scope, such as farms, animals, sensors, animal-days, valid hours and coverage; (3) project reductions resulting from interventions; and (4) AI Safety Korea's role in measurement design, data collection, quality control, analytics, alerts, reporting and evidence preparation. Monitoring volume is not a reduction, and a project's entire result is not automatically the platform provider's result.

GHG Protocol recommends disclosing avoided emissions enabled by products or services separately from the corporate inventory. Farm-project reductions therefore cannot be deducted from AI Safety Korea's Scope 1, 2 or 3 emissions.

Start with a counterfactual baseline

A baseline is not merely a pre-installation average. It is the emissions reasonably expected in the same period without the intervention. Because this counterfactual cannot be directly observed, weak baseline design cannot be rescued by precise downstream calculations.

Options include randomized treatment and control groups, crossover trials, quasi-experimental comparison farms or herds, sufficiently long before-and-after series, and models normalized for animal numbers, body weight, dry-matter intake, milk yield, growth, season and ventilation. Confounders, exclusions and missing-data rules should be fixed before results are known. FAO LEAP similarly compares scenarios with and without feed additives and considers impacts across the livestock supply chain.

Quantify methane mass, not concentration alone

A 10% fall in ppm does not prove a 10% fall in emissions. Concentration is a mixing ratio at a time and location; ventilation, wind, temperature, pressure, sensor placement and barn openings can change it even when emitted mass is unchanged.

Conceptually, methane mass is obtained by integrating exhaust airflow × the outlet–inlet concentration difference × temperature and pressure conversion over time. The evidence package therefore needs background and indoor concentrations, spatial representativeness, airflow methodology, environmental conditions, animal and production data, feed composition and adherence, sensor calibration and bump-test records, and transparent treatment of gaps and outliers.

IPCC livestock guidance links the appropriate tier to activity data and emission factors such as animal category, production conditions, intake and digestibility. On-farm sensor data should not be assumed to replace inventory factors automatically; it needs a project-specific measurement design and quality assessment.

Include project emissions and uncertainty

For period p, net reduction can be represented as ERp = BEp − PEp − SEp − CAp, where BE is counterfactual baseline emissions, PE is project emissions inside the boundary, SE is induced emissions outside the principal boundary, and CA is a conservative adjustment required by the methodology.

For a low-carbon feed intervention, additive production and transport, productivity changes and manure effects may matter. Do not apply an arbitrary uncertainty discount. Evaluate instrument accuracy, airflow estimates, sampling representativeness, gap filling and baseline-model uncertainty separately, then apply the methodology's conservativeness rules. ISO 14064-2 covers project and baseline sources, monitoring, quantification, data quality and reporting, so boundaries and data lineage must accompany the result.

Record the rule used to convert tCH₄ into tCO₂e

Calculate tCO₂e by multiplying net tCH₄ reduction by the applicable methane GWP. IPCC AR6 gives a GWP100 of 27.0 for non-fossil methane, but a programme or regulation may prescribe another value. Store the IPCC report and table, time horizon, fossil/non-fossil classification, applicable programme, algorithm version and effective date. GWP100 is an aggregation metric, not an automatic answer for every question about short-term climate effects.

Assess AI Safety Korea's causal contribution separately

A decrease on a farm does not prove that the measurement platform caused it. The farm implements the action, a feed supplier may provide the intervention, and a research partner may design the trial. A documented pathway could be: better observability → data quality → detection of anomalies or change → operating decision → correction or continuation of the intervention → reduction against baseline.

Useful indicators include connected farms, animals and sensors; valid hours, coverage, missingness and calibration compliance; reports and follow-up after alerts; feeding adherence and completed actions; estimated and verified tCH₄ or tCO₂e; and verification-ready evidence packages. If the platform only measured, say that it contributed to measurement and quantification. If analysis demonstrably changed operations, say that it supported implementation and persistence. Do not invent a percentage attribution and multiply the whole project result by it.

Prevent double counting and label maturity

Several partners may describe the same project while stating their roles, but their figures must not be added as separate global reductions. Maintain a project ledger with a unique ID, period, boundary, baseline, raw-data and calibration identifiers, formula and algorithm versions, emission factors and GWP, participants, claim language, and credit issuance, transfer and retirement status.

Label figures as: observed data; calculated emissions; estimated project reduction; independently verified reduction; or issued and retired carbon units. Under ISO 14064-3, verification evaluates whether a historical GHG statement meets criteria without material misstatement, while validation assesses the reasonableness of assumptions and methods for future outcomes. Neither is merely a sensor accuracy test.

Example

If an adjusted baseline is 10.00 tCH₄, project emissions are 9.20 tCH₄, induced emissions are 0.10 tCH₄ and the conservative adjustment is 0.05 tCH₄, the estimated net reduction is 0.65 tCH₄. Using AR6 non-fossil methane GWP100 of 27.0 gives 17.55 tCO₂e.

A defensible statement is: “A project supported by AI Safety Korea's measurement and analytics system produced an estimated reduction of 17.55 tCO₂e. This is an unverified estimate under illustrative assumptions and includes the farm's action and other partners' contributions.” Saying “AI Safety Korea reduced 17.55 tCO₂e” omits causality, shared contribution and assurance status.

Implementation checklist

  • Separate the corporate inventory from customer-project effects.

  • Predefine the boundary, period, intervention and counterfactual.

  • Distinguish concentration from mass emissions and document airflow.

  • Control animal, feed, production, season and ventilation variables.

  • Preserve calibration, missing-data, outlier and revision histories.

  • Include project emissions, secondary effects and conservative rules.

  • Link every result to its GWP source, version and methane type.

  • Maintain traceable raw data, algorithms and calculation versions.

  • Distinguish observed, calculated, estimated, verified and issued units.

  • Agree each participant's role and permitted claim.

  • Check for double counting and independently review reproducibility.

Conclusion

AI Safety Korea's contribution cannot be calculated from sensor counts or data volume. It requires a counterfactual baseline, mass-emission accounting, uncertainty assessment, secondary effects and a clear separation between project outcomes and the company's role. Transparent boundaries, coverage, baseline quality, calculation versions and assurance status are more valuable than an early headline number. With a durable evidence chain, contribution can be communicated as reproducible carbon data rather than promotional language.

Sources