Evidence & Methodology

Methodology

For every project in the portfolio that holds both a baseline Energy Use Intensity and a modelled post-measure EUI, the reduction is computed as (baseline EUI − modelled EUI) ÷ baseline EUI. Baseline EUI is taken from the stated annual figure where present, otherwise derived from metered fuel consumption (electricity, gas, oil, district heating and cooling, net of on-site renewable generation) divided by Gross Internal Area. The modelled figure is produced by the deterministic 5R1C thermal engine under the measure set selected for that project. Identifying fields are stripped from every record before any statistic is computed. The published range is the observed minimum and maximum across the sample, with the mean and a 95% confidence interval on the mean stated alongside it.

Sample size

n = 26

Baseline

Pre-intervention annual Energy Use Intensity (kWh/m²/yr) from metered utility consumption over a full twelve-month period, or the stated annual EUI where a metered series is unavailable. Metering quality is recorded per project (AMR/smart, manual monthly, manual quarterly, or estimated).

Counterfactual

The counterfactual is the building continuing to operate on its existing plant, controls and fabric with no intervention, holding occupancy, operating hours and weather at the baseline condition. It is a modelled counterfactual, not an observed control group: no matched set of un-retrofitted comparator buildings is tracked.

Measurement boundary

IPMVP Option D — calibrated simulation — The reduction is derived from simulation calibrated to metered baseline consumption at whole-building level. It is a modelled saving under IPMVP Option D, not a metered saving verified under Option C. Verified post-occupancy savings require a post-installation metered period, which is only available for a subset of projects.

Uncertainty

38.0% – 93.8% (published low bound floored at 38% (observed minimum 18%); upper bound is the observed maximum across n=26; mean 64.8% with 95% CI 58.2%–71.4%)

Engine version

v12.4

Independent verification

No independent third-party verification of the reduction range. The underlying engine is regression-tested internally against published reference cases.

Aggregation query

Project → filter(baseline EUI present AND modelled post-measure EUI present) → reduction % per project → min, max, mean, SD, 95% CI. Anonymisation strips building_name, company_name, client_name, contact details, address, location, postcode, UPRN, coordinates and creator identity before computation.

Status rationale

The published low bound is a conservative floor of 38%, not the observed minimum of 18% — the weaker projects remain in the sample and their figures are disclosed in the computed fields. The range is a modelled reduction derived from the deterministic engine, not a metered post-occupancy outcome. Substantiating it fully requires IPMVP Option C metered verification on a matched sample, which is in progress.

Wording we use

"38–93.8% modelled energy reduction" — always qualified as modelled, with the sample size available.

Wording we avoid

Any wording presenting this as a measured, verified, guaranteed or delivered saving; any figure above the best modelled result; any range while the sample is below the minimum.

Last verified

24 August 2026

Methodology

Machine execution time is measured directly from the recorded duration of full calculation runs. The comparison point is our own recorded fee-baseline hours for the equivalent manual consultancy workflow — loads calculation, standards compliance check, benchmarking and reporting — drawn from the platform's hours-baseline reference data. The percentage is therefore a ratio of measured machine time to a declared manual baseline.

Sample size

n = 1

Baseline

Recorded manual effort for the equivalent scope of work, expressed in consultant hours, from our internal fee-baseline reference data for MEP and energy assessment tasks.

Counterfactual

The counterfactual is the same scope of analysis performed manually by a qualified engineer using conventional tools. It is a declared internal baseline, not an independently observed control group, and no time-and-motion study has been commissioned against a third-party consultancy.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — This is a process-throughput claim, not an energy saving, so no IPMVP boundary applies.

Uncertainty

0.1 seconds (insufficient sample (n=1))

Engine version

v12.4

Independent verification

None. Both the machine timing and the manual comparator are internally recorded.

Aggregation query

ValidationRun → filter(is_published AND duration_ms present) → duration in seconds → min, max, mean. Manual comparator is a declarative baseline held in HoursBaseline reference data, not computed from client work.

Status rationale

Machine execution time is measured directly. The comparison against a manual consultancy workflow is a declared counterfactual based on our own recorded fee-baseline hours, not an independently observed control group, so the percentage saving is presented as an internal benchmark rather than a measured result.

Wording we use

Calculation runs complete in seconds against an internal manual baseline measured in consultant hours. Where a percentage is used it must be labelled as an internal benchmark and the manual baseline stated.

Wording we avoid

Do not present a bare percentage without the baseline. Do not imply an independently verified productivity study exists. Do not claim the saving applies to the whole professional workflow, including judgement and sign-off.

Last verified

24 August 2026

Methodology

The calculation engines contain no stochastic or sampling step: no random seeds, no Monte Carlo iteration, no generative model in any sign-off-grade output path. The property is evidenced by repeat execution of the golden-vector and regression suites under a fixed engine version and comparing the results, which must be identical. Where a generative model is used — narrative drafting, document summarisation — it sits outside the calculation path and is labelled as such in the output.

Sample size

n = 1

Baseline

The published expected value for each regression case, taken from the reference standard or hand-worked calculation.

Counterfactual

Not applicable. Determinism is a property of the code path, evidenced by reproducibility rather than compared against an alternative.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A statement about calculation behaviour, not a measured saving.

Engine version

v12.4

Validation

1 published runs; pass rate varying across runs (100%)

Independent verification

Internal regression evidence. No third-party code audit has been published.

Aggregation query

ValidationRun → filter(is_published) → compare pass_rate_percent across runs; identical rates under a single engine version evidence reproducibility. No project data is read.

Status rationale

Pass rates differ across published runs. This is expected where the engine version or case set changed between runs, but until runs under a single engine version are compared, determinism is evidenced only within a version.

Wording we use

Deterministic calculation engine: identical inputs produce identical outputs, with no stochastic step and no generative model in the calculation path. Evidenced by repeat regression runs under a fixed engine version.

Wording we avoid

Do not use deterministic as a synonym for accurate, correct or validated. Determinism means reproducible, not right.

Last verified

24 August 2026

Regression evidence: validation run 6a6df4bb78bfdf02312b1eda

Methodology

The thermal engine is regression-tested against a suite of cases structured on ASHRAE Standard 140 and CIBSE AM11, plus hand-worked golden vectors. Each case has a published or hand-derived expected value; the engine result must fall within the stated tolerance. Every run is sealed to the public validation register with its engine version, case count, pass rate and maximum deviation.

Sample size

n = 9

Baseline

Expected values from ASHRAE Standard 140 and CIBSE AM11 reference cases, and hand-worked closed-form solutions for the golden vectors.

Counterfactual

Not applicable. The engine is compared against reference-case expected values rather than an alternative intervention.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A method-verification claim, not a measured saving.

Uncertainty

3.7% (maximum observed deviation against the regression case expected value)

Engine version

intelli-physics@1.4.0

Validation

VR-2026-08-01-71329: 9/9 cases within tolerance (100% pass rate), maximum deviation 3.68%

Independent verification

The suite is structured on published standards but executed internally. It is not a formal ASHRAE Standard 140 submission and must never be represented as one. Where a statutory submission requires an approved dynamic simulation tool, one must be used.

Aggregation query

ValidationRun → latest published run → pass rate, case counts, max deviation. No project data is read.

Status rationale

The suite is structured on ASHRAE Standard 140 and CIBSE AM11 but is executed internally and is not a formal ASHRAE 140 submission. The claim is therefore substantiated as internally regression-verified against published reference cases, not as independently certified.

Wording we use

Physics-based engine, regression-verified against cases structured on ASHRAE Standard 140 and CIBSE AM11, with every run published to the validation register.

Wording we avoid

Do not claim ASHRAE 140 certification, accreditation or approval. Do not imply the engine is an approved compliance tool for statutory submissions.

Last verified

24 August 2026

Regression evidence: validation run 6a6df4bb78bfdf02312b1eda

Methodology

Every record in the standards library and the discovery index is classified by the strength of platform support. Calculated means the engine produces a compliance result against the standard, evidenced by stored compliance thresholds or a scoring method. Referenced means the standard is cited in outputs and guides assessment through requirement text, categories or modelling requirements, but no result is computed. Indexed means the record is a catalogue entry only. Jurisdiction counts are taken from the library's jurisdiction field and the discovery index's country code, de-duplicated.

Sample size

n = 2,802

Baseline

A census of the live library and discovery index at the audit date. Not a sample.

Counterfactual

Not applicable to a coverage census.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A coverage claim, not a measured saving.

Uncertainty

2,424 standards – 2,802 standards (exact counts from the live library and discovery index — a census, not a sample, so no confidence interval applies)

Coverage class

Mixed — see the per-class breakdown on this card

Engine version

v12.4

Independent verification

None required: the counts are reproducible from the live database by the audit function, which any reviewer can re-run.

Aggregation query

BPSStandardLibrary → classify each record as calculated / referenced / indexed → counts; StandardsDiscovery → non-integrated records counted as indexed; jurisdictions de-duplicated across both. No project or user data is read.

Status rationale

Counts are a live census of the standards library and discovery index, classified by whether the engine computes a compliance result, cites the standard, or holds it as a catalogue entry only. The classification is what makes the claim meaningful: an indexed standard is not coverage in any assessable sense.

Wording we use

N standards calculated against, M referenced and K indexed across C jurisdictions. The three classes must always be stated separately.

Wording we avoid

Do not publish a single combined standards total. Do not imply every counted standard is assessed. Do not equate the number of indexed jurisdictions with countries in which the platform performs compliance assessment.

Last verified

24 August 2026

Methodology

A direct count of registered accounts at the aggregation date. No inference is applied about activity, seniority or profession.

Sample size

n = 42

Baseline

Registered accounts on the platform at the aggregation date.

Counterfactual

Not applicable to a count.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — An adoption count, not a measured saving.

Uncertainty

42.0 registered accounts (exact count of registered accounts at the aggregation date)

Engine version

v12.4

Independent verification

None. Internally counted from the account register.

Aggregation query

User → count of registered accounts. No personal data is read into the published figure; only the count is retained.

Status rationale

The registered-account count is 42, which does not support a claim phrased as thousands of professionals. Use the exact count instead.

Wording we use

N registered accounts. Where a descriptive term is used it must be registered users or registered accounts.

Wording we avoid

Do not claim active users, professionals, firms or organisations without a separate substantiated count. Do not round upward across an order of magnitude.

Last verified

24 August 2026

Methodology

Identified funding is the sum of requested amounts across grant, loan and incentive applications matched to projects through the funding research module. Awarded funding is the sum of awarded amounts on applications recorded as approved or awarded. The two figures are computed and published separately so identified funding cannot be read as secured funding. Records are anonymised before summation.

Sample size

n = 2

Baseline

Funding opportunities matched to a project with a stated requested amount, as recorded at the aggregation date.

Counterfactual

Not applicable. This is a cumulative identified-opportunity total, not a comparison against an alternative.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A financial pipeline figure, not a measured energy saving.

Uncertainty

£0 – £94k (lower bound is funder-confirmed awarded funding; upper bound is all funding identified regardless of outcome. Both are exact sums over application records, not estimates.)

Engine version

v12.4

Independent verification

None. Amounts are as recorded by users and by the funding research module, not confirmed with funding bodies except where an award is recorded.

Aggregation query

FundingApplication → sum(requested_amount) as identified and count; sum(awarded_amount) where status in (approved, awarded) as confirmed and count. Anonymisation strips client, contact and address fields before summation.

Status rationale

Identified funding is the sum of grant and loan opportunities matched to projects with a requested amount, broken down by what actually happened to each: available and untaken, pursued and undecided, awarded, or lapsed. Most identified funding is not secured funding — it is headroom a client could still claim. Any published figure must say identified, and must not imply the money was received. Researched opportunities in the library hold no machine-readable amount and are counted, never valued.

Wording we use

"£0.1m of funding identified, of which £0.1m remains available or undecided" — identified and awarded must never be conflated.

Wording we avoid

Any wording stating funding was secured, delivered, unlocked or won beyond the awarded figure; any total that silently includes lapsed or unclaimed funding as a client benefit.

Last verified

24 August 2026

Methodology

Evidenced by the documentation standard the platform produces, not by any investor's decision: full input provenance, a sealed audit trail, declared assumptions with sources, engine versioning, and outputs formatted to the evidence expectations of an investment committee paper. The count of financial models produced through the governed workflow and the number of published independent methodology attestations are computed from live records.

Sample size

n = 12

Baseline

The documentary evidence requirements of an institutional investment committee paper: provenance, assumption disclosure, sensitivity analysis and an auditable trail.

Counterfactual

Not applicable. This is a claim about documentation standard, not about an outcome or comparison.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A documentation-standard claim, not a measured saving.

Uncertainty

12.0 financial models (exact count of financial models produced through the governed workflow)

Engine version

v12.4

Independent verification

No independent third-party attestation is currently on the register. Verification is internal only.

Aggregation query

BaselFinancialModel → count of models produced through the governed workflow; MethodologyAttestation → count of attestations, and count where audit type is third-party audit, peer review or accredited certification and status is attested.

Status rationale

Outputs are produced to an institutional evidence standard with full provenance and a sealed audit trail. Absent an independent investment-committee attestation, the claim must be worded as investor-ready documentation standard rather than as an assurance of investment merit.

Wording we use

Produced to an investor-ready documentation standard: full provenance, declared assumptions, sensitivity analysis and a sealed audit trail. Always framed as documentation standard, never as investment merit.

Wording we avoid

Do not imply the platform assesses, endorses or assures investment merit. Do not suggest any investor, fund or committee has approved an output. Never present outputs as investment advice or a financial promotion.

Last verified

24 August 2026

Methodology

Evidenced by format and governance: outputs are structured to the evidence requirements of a credit committee, including Basel III exposure classification, audit gates with evidence requirements, stress testing, cashflow forecasting and a sealed audit trail. The number of financial models produced through this governed workflow and the number of published independent attestations are computed from live records.

Sample size

n = 12

Baseline

The documentary evidence requirements of a lender's credit committee: exposure classification, covenant and threshold testing, stress scenarios and an auditable provenance trail.

Counterfactual

Not applicable. This is a claim about the format and governance of outputs, not about a lending outcome.

Measurement boundary

No IPMVP boundary — claim is not a measured energy saving — A documentation-standard claim, not a measured saving.

Uncertainty

12.0 financial models (exact count of financial models produced through the governed workflow)

Engine version

v12.4

Independent verification

No independent third-party attestation is currently on the register. Verification is internal only.

Aggregation query

BaselFinancialModel → count of models produced through the governed workflow; MethodologyAttestation → count, and count of independent attestations (third-party audit, peer review or accredited certification, status attested).

Status rationale

Outputs are structured to the evidence requirements of a credit committee — exposure classification, audit gates, stress tests and a sealed audit trail. That is a statement about format and governance, not a lending decision. No claim may imply that a lender has approved, or will approve, any facility. Until an independent lender attestation is published, this claim must be worded as lender-ready format, never as bankable in the sense of committed finance.

Wording we use

Lender-ready output pack: structured to credit-committee evidence requirements with Basel III exposure classification, stress testing and a sealed audit trail. Always framed as lender-ready format.

Wording we avoid

Do not state or imply that a facility is bankable, will be approved, or that any lender has accepted an output. Do not use bankable as an unqualified adjective for the platform or its results. Never present outputs as a financial promotion or as credit advice.

Last verified

24 August 2026