Understanding Mapping Agent Quality Labels

What are quality labels?

When Mapping Agent (formerly known as Autopilot) matches a business activity description to an emission factor, not every match is equally reliable. Some descriptions have a clear, unambiguous best match. Others are vague or could cover multiple activities.

Quality labels make this uncertainty visible. Each match Mapping Agent returns is tagged with a label that tells you whether a match is strong and ready to use, or plausible but worth a closer look before you rely on it.

This matters because emissions calculations feed into financial disclosures, regulatory reports, and sustainability commitments. A mismatch between an activity and its emission factor, even an unintentional one, can become a compliance issue. Quality labels give your team a clear signal for where to focus human review, rather than reviewing everything or nothing.

How labels are determined

Mapping Agent assigns quality labels through a two-stage process:

1. Filtering

If you provide filters in your request (such as region, year, sector, or scope), the full pool of candidate emission factors is narrowed by those filters first. Only the remaining factors are considered for matching and labeling.

2. LLM evaluation

From the filtered candidates, a large language model (LLM) evaluates each one by comparing the name of your input description against the emission factor name. Based on that reasoning, the LLM assigns one of three outcomes:

• accept - the match is strong
• review - the match is plausible but uncertain
• Internally rejected - removed in post-processing and not returned in the response

A note on calibration

Labels are calibrated against evaluation data, so the meaning of accept and review stays consistent even as the underlying model improves. The specific inputs used by the LLM during evaluation may also evolve as we continue to experiment and refine the pipeline.

Available labels

accept

The match is strong. The emission factor should be a good fit for the activity you described based on: the material, the stage in the supply chain, the level of specificity inputted. You can typically use this result in your calculations, however a human validation step is always encouraged.

review

The match is plausible but not certain. The emission factor may cover the right broad category but at too general a level, or it may be one of several valid interpretations of an ambiguous description. A person should verify this match before it is used in formal reporting.

How to use quality labels in your workflow

A straightforward approach:

1. Prioritize accept matches. Treat them as strong recommendations rather than guaranteed correct answers. A sense-check is always good practice, and is required before including them in formal reporting.
2. Review review matches before use. For each one, check whether the emission factor genuinely represents your activity. If it does, keep it. If not, search for a better match or flag it for your sustainability team.
3. Look at patterns in your review results. If a large proportion of your results come back as review, this usually points to an input data quality issue: vague descriptions, internal codes, or missing context. Improving the descriptions you send to Mapping Agent is the most effective way to improve your overall match quality. See Optimizing Inputs for Mapping Agent. If match quality remains low after improving your descriptions, it may be that stronger matches exist in premium datasets you don’t currently have access to. You can check this by setting availability: "include_unavailable" in your Suggest request to see all matches regardless of license — Core Datasets  are available on all plans.

Adjusting for your reporting context

How rigorously you apply the labels depends on what the results are used for.

For formal disclosures (CDP, CSRD, SBTi, audit-facing reports): All matches should be reviewed to confirm they reflect your actual business activities. For review matches, a more thorough check is required: each one should be justified and documented as accurately representing your business reality before inclusion.

For internal estimates or directional analysis: You may choose to accept review matches for lower-materiality activities, focusing manual effort on the categories that contribute most to your total footprint.

The role of human judgment

Quality labels are a guide, not a substitute for domain knowledge. A sustainability analyst who knows their business will sometimes override a label: accepting a review match they recognize as correct for their specific context, or questioning an accept match that seems off. The labels are designed to reduce the volume of manual review, not to eliminate it.