How ERW Crediting Works

Introduction

We recently published Everest Designer, a tool that allows users to design and simulate enhanced rock weathering projects by adjusting 67 parameters, resulting in a distribution of potential economic and CDR outcomes as well as an expected outcome. In order to piece the tool together, we reviewed 445 pages of documentation from the 3 leading methodologies to derive crisp formulas that outline in which exact scenarios project proponents can issue how many credits.

This article walks “expert” users through the math and key assumptions. If you know what differentiates a 2-plot from a 3-plot approach and what liquid phase sampling densities are reasonable on a control plot, you will likely find the summary below highly insightful. For everyone else, ask your favorite LLM about any terms below that are foreign to you, or jump to our blog post that outlines what this tool taught us about the future of ERW.

Overview: 3 ERW Protocols, 3 Philosophies of Risk

The current generation of leading ERW crediting methodologies—Isometric, Puro, and Rainbow—all attempt to answer the same core question: how much CO₂ removal can be credited with high confidence, given noisy field data and incomplete observability?

They differ not in whether they subtract losses, counterfactuals, emissions, and leakage—all three do—but in how uncertainty is handled, when projects are allowed to earn non-zero credits, and where conservatism is applied in the accounting chain. The standards also differ in the vocabulary they use: treatment plot = evaluation plot = low-density plot. For easy comparison across the equations, we stuck with Isometric language as they were first to issue credits. Leakage is excluded in our analysis due to its highly project-specific nature and typically low impact, resulting in the common crediting equation below:

\[
\mathrm{CO_{2}e}_{\mathrm{Stored}}
= {CO_{2}e}_{{\mathrm{Drawdown}}}
- {CO_{2}e}_{{\mathrm{Losses}}}
- {CO_{2}e}_{{\mathrm{Counter\;factual}}}
- {CO_{2}e}_{{\mathrm{Emissions}}} \,
\]

Quantification of carbon drawdown is primarily achieved via in-field measurements. Some standards further require secondary validation measurements using a different approach to corroborate these measurements. At a high level, the differences between how the standards use these measurements can be summarized as follows:

Isometric spreads conservatism across multiple statistical tests and percentile choices, allowing partial credit even when tests fail.
Puro is binary and gate-driven: strata (homogenous project areas) either pass all tests and earn conservative credits, or earn nothing.
Rainbow is structurally simpler, embedding conservatism directly into drawdown estimates via fixed discounts and does not require validation measurements.

ERW Methodology 1: Isometric — Layered Tests, Continuous Outcomes

Isometric treats ERW crediting as a probabilistic estimation problem with multiple safeguards, rather than a pass/fail exercise. For each spatial stratum, it computes CO₂ stored as measured drawdown minus losses, counterfactuals, and emissions—but applies different statistical treatments depending on whether the data clear specific tests.

Two tests govern how conservative the accounting becomes:

Representativeness test: checks whether large deployment area measurements are statistically consistent with smaller treatment plot measurements. If they are, the protocol trusts the treatment distribution; if not, it caps drawdown using the more conservative of treatment and deployment distributions.
Validation test: checks whether the independent validation measurements align with quantification results. If it fails, downstream and other losses are inflated from a point estimate to a conservative upper bound.

Conservatism is applied via percentiles of empirical distributions rather than fixed discounts. Importantly, failure does not zero out credit; it simply pushes the accounting toward more conservative bounds.

‍

Simplified Isometric Crediting Formula

\[
\mathrm{CO_{2}e}_{\mathrm{Stored}}
= \sum_{i=1}^{S}max\!\bigl(0,\quad {CO_{2}e}_{{\mathrm{Drawdown},i}}
- {CO_{2}e}_{{\mathrm{Losses},i}}
- {CO_{2}e}_{{\mathrm{Counter\;factual},i}}\bigr)
- {CO_{2}e}_{{\mathrm{Emissions}}} \,
\]

\[
{CO_{2}e}_{\mathrm{Drawdown},i} =
\begin{cases}
P_{16}\!\bigl( {M}_{Q,T,i} \bigr),
& \text{if } R_{\mathrm{Test},i}=1, \\[10pt]
\min\!\bigl( P_{16}\!\bigl( {M}_{Q,T,i} \bigr),\, P_{16}\!\bigl( {M}_{Q,D,i} \bigr) \bigr),
& \text{otherwise}.
\end{cases}
\]

\[
R_{\mathrm{Test},i} =
\begin{cases}
1,
& \text{if } \overline{M}_{Q,D,i} \in
\mathrm{CI}_{95}\!\bigl( \overline{M}_{Q,T,i} \bigr), \\[10pt]
0,
& \text{otherwise}.
\end{cases}
\]

\[
{CO_{2}e}_{\mathrm{Losses},i} =
\begin{cases}
L_i,
& \text{if } V_{\mathrm{Test},i}=1, \\[10pt]
P_{84}\!\bigl( L_i \bigr),
& \text{otherwise}.
\end{cases}
\]

\[
V_{\mathrm{Test},i} =
\begin{cases}
1,
& \text{if } \overline{M}_{V,T,i} - \overline{M}_{V,C,i} \in
\mathrm{CI}_{99.7}\!\bigl( \overline{M}_{Q,T,i} - \overline{M}_{Q,C,i} \bigr), \\[10pt]
0,
& \text{otherwise}.
\end{cases}
\]

\[
{CO_{2}e}_{{\mathrm{Counter\;factual},i}} = \overline{M}_{Q,C,i}
\]

\[
{CO_{2}e}_{{\mathrm{Emissions}}} = E \,
\]

$S$: Spatial strata over which the project is partitioned across indexes $i$

$R_{\mathrm{Test},i}$: Representativeness test of treatment against deployment quantification mean for stratum $i$

$V_{\mathrm{Test},i}$: Validation test of treatment quantification against validation mean for stratum $i$

$P_{k}(\cdot)$: Takes the $k$th percentile value from a given distribution

$M_{t,a,i}$: Individual ${CO_{2}e}_{\mathrm{Drawdown}}$ measurement (sample) of type $p$, in area $a$, and stratum $i$.

$\overline{M}_{t,a,i}$: Sample mean of measurement type $t$, project area $a$, and stratum $i$.

$t \in \{Q, V\}$: measurement types quantification $Q$ and validation $V$.

$a \in \{C, T, D\}$: project areas control $C$, treatment $T$, and deployment $D$.

$L_i$: Estimate of ${CO_{2}e}$ losses in stratum $i$.

$E$: Estimate of project-related ${CO_{2}e}$ emissions.

‍

ERW Methodology 2: Puro — Binary Gates and Deep Conservatism

Puro takes a fundamentally different stance: either a stratum is proven beyond doubt, or it earns no credits at all.

As before, each stratum must pass both a representativeness gate and a validation gate. The representativeness test, however, requires that even the conservative lower confidence bound of net drawdown is strictly positive. The validation test requires broad agreement between quantification and independent validation measurements. If either test fails, the credited amount for that stratum is exactly zero.

Only if both gates pass does the protocol compute potential storage—and even then, it applies a highly conservative lower percentile to the entire net balance of drawdown minus losses, counterfactuals, and emissions. The standard describes this as a Factor of Conservativeness multiplication, but the math is equivalent.

‍

Simplified Puro Crediting Formula

\[
\mathrm{CO_{2}e}_{\mathrm{Stored}}
=
\sum_{i=1}^{S}max\!\bigl(0, \quad \mathrm{CO_{2}e}_{\mathrm{Stored, i}}\bigr)
\]

\[
\mathrm{CO_{2}e}_{\mathrm{Stored, i}}
=
\begin{cases}
\mathrm{CO_{2}e}_{\mathrm{Stored, potential, i}},
& \text{if } R_{\mathrm{Test},i}=1 \quad and \quad V_{\mathrm{Test},i}=1\\[10pt]
0,
& \text{otherwise}.
\end{cases}
\]

\[
\mathrm{CO_{2}e}_{\mathrm{Stored, potential, i}}
=
\quad P_{10}\!\bigl( {CO_{2}e}_{{\mathrm{Drawdown},i}}
- {CO_{2}e}_{{\mathrm{Losses},i}}
- {CO_{2}e}_{{\mathrm{Counter\;factual},i}}
- {CO_{2}e}_{{\mathrm{Emissions}}} \, \bigr)
\]

\[
R_{\mathrm{Test},i} =
\begin{cases}
1,
& \text{if } \ \mathrm{LCB}_{95}\!\bigl(\overline{M}_{Q,D,i}
- \overline{M}_{Q,C,i}\bigr) > 0, \\[10pt]
0,
& \text{otherwise},
\end{cases}
\]

\[
V_{\mathrm{Test},i} =
\begin{cases}
1,
& \text{if } \overline{M}_{V,T,i} - \overline{M}_{V,C,i} \in
\mathrm{CI}_{99}\!\bigl( \overline{M}_{Q,T,i} - \overline{M}_{Q,C,i} \bigr), \\[10pt]
0,
& \text{otherwise}.
\end{cases}
\]

\[
{CO_{2}e}_{\mathrm{Losses},i} =
L_i,
\]

\[
{CO_{2}e}_{{\mathrm{Counter\;factual},i}} = \overline{M}_{Q,C,i}
\]

\[
{CO_{2}e}_{{\mathrm{Emissions}}} = E \,
\]

‍

$S$: Spatial strata over which the project is partitioned across indexes $i$

$R_{\mathrm{Test},i}$: Representativeness test of treatment against deployment quantification mean for stratum $i$

$V_{\mathrm{Test},i}$: Validation test of treatment quantification against validation mean for stratum $i$

$P_{k}(\cdot)$: Takes the $k$th percentile value from a given distribution

$M_{t,a,i}$: Individual ${CO_{2}e}_{\mathrm{Drawdown}}$ measurement (sample) of type $p$,
in area $a$, and stratum $i$.

$\overline{M}_{t,a,i}$: Sample mean of measurement type $t$, project area $a$, and stratum $i$.

$t \in \{Q, V\}$: measurement types quantification $Q$ and verification $V$.

$a \in \{C, T, D\}$: project areas control $C$, evaluation $T$, and application $D$.

$L_i$: Estimate of ${CO_{2}e}$ losses in stratum $i$.

$E$: Estimate of project-related ${CO_{2}e}$ emissions.

‍

ERW Methodology 3: Rainbow - Embedded Discounts, Minimal Testing

Rainbow simplifies the structure by embedding conservatism directly into drawdown estimates and eliminating validation measurements.

Each stratum’s drawdown is taken as a conservative percentile of treatment measurements. A representativeness test still exists, but instead of switching distributions or zeroing credits, failure triggers a fixed multiplicative discount (the details of the discount can depend on project specifics).

Losses, counterfactuals, and emissions are then subtracted deterministically without further binary gates.

‍

Simplified Rainbow Crediting Formula

\[
\mathrm{CO_{2}e}_{\mathrm{Stored}}
=\sum_{i=1}^{S}max\!\bigl(0,\quad {CO_{2}e}_{{\mathrm{Drawdown},i}}
- {CO_{2}e}_{{\mathrm{Losses},i}}
- {CO_{2}e}_{{\mathrm{Counter\;factual},i}}\bigr)
- {CO_{2}e}_{{\mathrm{Emissions}}} \,
\]

\[
{CO_{2}e}_{\mathrm{Drawdown},i} =
\begin{cases}
P_{10}\!\bigl( {M}_{Q,T,i} \bigr),
& \text{if } R_{\mathrm{Test},i} =1, \\[10pt]
P_{10}\!\bigl( {M}_{Q,T,i} \bigr)*0.8,
& \text{otherwise}.
\end{cases}
\]

\[
R_{\mathrm{Test},i} =
\begin{cases}
1,
& \text{if } \overline{M}_{Q,D,i} \in
\mathrm{CI}_{95}\!\bigl( \overline{M}_{Q,T,i} \bigr), \\[10pt]
0,
& \text{otherwise}.
\end{cases}
\]

\[
{CO_{2}e}_{\mathrm{Losses},i} =
L_i,
\]

\[
{CO_{2}e}_{{\mathrm{Counter\;factual},i}} = \overline{M}_{Q,C,i}
\]

\[
{CO_{2}e}_{{\mathrm{Emissions}}} = E \,
\]

$S$: Spatial strata over which the project is partitioned across indexes $i$

$R_{\mathrm{Test},i}$: Representativeness test of treatment against deployment quantification mean for stratum $i$

$P_{k}(\cdot)$: Takes the $k$th percentile value from a given distribution

$M_{t,a,i}$: Individual ${CO_{2}e}_{\mathrm{Drawdown}}$ measurement (sample) of type $p$,
in area $a$, and stratum $i$.

$\overline{M}_{t,a,i}$: Sample mean of measurement type $t$, project area $a$, and stratum $i$.

$t \in \{Q \}$: measurement type quantification $Q$

$a \in \{C, T, D\}$: project areas control $C$, high-definition $T$, and low-definition $D$.

$L_i$: Estimate of ${CO_{2}e}$ losses in stratum $i$.

$E$: Estimate of project-related ${CO_{2}e}$ emissions.

‍

Summary: Same Equation, Different Risk Postures

All three methodologies share the same structural backbone: credited CO₂ equals drawdown minus losses, counterfactuals, and emissions.

What differs is where uncertainty lives:

Isometric distributes uncertainty across multiple tests and percentiles.
Puro concentrates uncertainty into binary gates and deep tail conservatism.
Rainbow internalizes uncertainty through fixed discounts and simplified structure.

These choices directly affect sampling strategies, credit volatility, downside risk, and incentives for measurement and validation. Understanding the math ultimately means understanding these risk philosophies—and choosing which one aligns with a project’s data quality and tolerance for uncertainty. Use our tool to better understand how your project might fare under each methodology: design.everestcarbon.com

‍

Other Modeling Assumptions

For those who want to understand other key assumptions or design choices in the model, we have provided a brief summary below. If you have further questions, feature requests, or would like to explore custom use cases, please reach out to our commercial lead, Jonte Boysen (jonte@everestcarbon.com).

Weathering rate
Mineral weathering is modeled to follow an exponential decay curve. The rate of decay is set by the user by specifying the number of years it takes for 80% of the mineral to weather.

Strata
Strata are currently assumed to be of equal size with equal means and identical measurement distributions. Expected results are calculated at the stratum level and then multiplied by the number of strata. In future versions, each stratum will be simulated independently. The Isometric requirement of a maximum 2,000 ha stratum size is applied to all standards by default.

Minerals
Mineral characteristics are not intended to be representative of an entire mineral category. For example, basalt exhibits wide variance in weathering rates, carbon capture potential, and application practices. Please let us know if there are specific minerals or mineral characteristics that would be useful to include.

Spatial heterogeneity
We use a proprietary model to estimate how spatial heterogeneity scales with project size. This non-linear model allows users to better estimate how increasing project scale affects the optimal number of measurement points per hectare required to maximize project outcomes. The “Baseline spatial heterogeneity” factor allows users to scale expected heterogeneity up or down (e.g., lower values for very sandy or highly homogeneous fields).

Spreading heterogeneity
Spreading heterogeneity is an often under-appreciated contributor to measurement variance. We model this using a lognormal distribution, reflecting the fact that standard deviation can be a multiple of the median while negative values are physically impossible.

Temporal heterogeneity
Temporal heterogeneity introduces significant noise for temporally sampled measurements. Because standard deviations can easily exceed the mean while the lower bound is capped at zero, temporal variability is also modeled using a lognormal distribution.

Measurement performance
Measurements for each approach are assumed to include all inputs required to calculate carbon captured. Soil-phase measurement uncertainty, for example, also includes uncertainty from loss-term measurements such as strong-acid weathering. Measurement uncertainty is also modeled using lognormal distributions, with inputs defined as the 95th-percentile measurement divided by the median when repeatedly measuring the same value - essentially precision. There is a separate field for entering the expected measurement bias. We were unable to find robust, standardized quantifications of measurement error for different ERW measurement approaches; as a result, default values should not be interpreted as empirically validated. Users are encouraged to adjust these assumptions to their best estimates.

Sampling distributions
Samples drawn for each plot and measurement type are assumed to be lognormally distributed, consistent with the underlying variance assumptions used elsewhere in the model.

Measurement cost
Measurement costs are simplified to a single per-sample value. Actual costs can vary substantially by region and logistics. To adjust for project-specific conditions, users can modify the per-sample cost accordingly.

Other costs‍
Cost currently do not scale with project area. This means to obtain a more representative view of project financials asprojects scale, costs need to be adjusted manually.

Sampling density
Sampling density assumptions determine the number of measurements taken per hectare and directly affect statistical power, uncertainty, and total measurement cost. You typically need to adjust these values from their default to find the sampling densities for the different plots that maximize project outcomes.

Allowed measurements for quantification
All measurement types are currently allowed for quantification across all standards. This does not fully reflect current real-world restrictions under all registries and may be refined in future versions.

Puro validation test failure
When the Puro validation test is failed, the project proponent is asked to provide supplementary information to explain the difference. To calculate an expected value in our model, we assumed that 50% of the time a satisfactory explanation is found, resulting in full credit issuance for the stratum in question.

Leakage
Leakage is not explicitly modeled. If desired, leakage can be incorporated by adding it to the losses term provided in the model.

Standard issuance costs
Issuance costs generally decrease with project scale, though the scaling mechanism differs by standard. Some fees scale with aggregate credited volume. By default, we use the fee level applicable to the smallest project sizes. Users should verify current pricing directly with each standard: