Evidence reviewed 27 July 2026 · Assessment principles v1.0

Evidence behind every label.

Aurxo makes each conclusion understandable by publishing its scope, metrics, evidence quality, sources, confidence and known limitations.

Aurxo publishes enough evidence and reasoning to understand and challenge each assessment. Exact thresholds, weights, ranking coefficients and implementation details remain proprietary to protect the integrity of the product.

Four signals, kept deliberately distinct.

Energy
Operational demand
Energy used for the stated product, task and functional unit.
Carbon
Climate impact
Greenhouse-gas impact for the stated system boundary and electricity context.
Water
Water demand
Water consumption associated with the stated functional unit and boundary.
Evidence integrity
Confidence and transparency
What is measured, reported, modelled or unknown. This never disguises environmental impact.
A consistent internal model. Aurxo evaluates the complete evidence set for the defined scope. Strong performance in one area cannot erase material weaknesses or missing evidence elsewhere.

No evidence, no invented conclusion. When evidence is incomplete, stale or not comparable for the stated scope, Aurxo can withhold the overall grade and explain why.

How Aurxo reaches a public result

01

Define the exact scope

Product or reference model, task, functional unit, date, deployment assumptions and system boundary.

02

Collect traceable evidence

Every published metric links to a named source, publication date and evidence type.

03

Classify evidence integrity

Provider measurements, externally reviewed evidence, independent estimates, claims and unknowns remain visibly different.

04

Apply the Aurxo assessment model

The same internal rules are applied consistently to comparable scopes. The detailed mechanics are proprietary.

05

Publish the record

The result is released with its scope, values, confidence, sources, limitations, date and version.

What is public. What is protected.

Public

Everything needed to assess the claim

Scope, measurements, evidence type, source links, confidence, limitations, assessment date, version and correction process.

Proprietary

The machinery that creates the product

Exact thresholds, weights, coefficients, tie-break rules, internal benchmark library, structured datasets and implementation code.

Product grades and reference labels are different.

A current product grade applies only when Aurxo has sufficiently current product-level evidence. A reference label belongs to a named model, prompt size, deployment and study date. A historical model reference cannot automatically become a universal grade for a routed product.

Aurxo Engine v1.6.4: capable first, greener among capable options.

The public interface sends the request securely to the private Aurxo Engine, which identifies the intended task, input format, desired output and critical requirements. It then removes unsuitable tools, compares documented product capabilities and asks one focused clarification when the request is materially underspecified.

Privacy by design. The Aurxo Engine processes task text transiently in AWS memory. The raw wording is not stored in the Aurxo analytics database or included in custom application logs. Detailed capability weights, ranking coefficients and tie-break rules remain proprietary.
01

Interpret the need

Identify the job, input, output and critical requirements in English or Spanish.

02

Remove unsuitable options

A tool must be capable of completing the actual task before environmental evidence can influence the choice.

03

Compare documented capabilities

Aurxo uses dated public product documentation and may show fewer than three options instead of filling the list with weak matches.

04

Clarify when necessary

One useful question is better than a confident generic recommendation.

05

Prioritize the greener capable option

After weak matches are removed, Aurxo prioritizes the strongest scope-compatible environmental signal among tools that can do the job well. If comparable evidence is unavailable, Aurxo says so and keeps the task-fit order.

Capability profile date: 28 July 2026. Profiles are editorial comparisons of publicly documented functions, not laboratory benchmarks or guarantees of output quality. Availability can vary by plan, region and workspace.

Selected official capability sources: OpenAI data analysis · Anthropic Claude professional workflows · Gemini file analysis · Copilot in Excel · Perplexity file uploads · Consensus Research Agent · DeepL document translation · Runway Gen-4.5 · Descript speaker detection · Zapier Agents · Replit build and publish.

View current assessments →