One workbench takes an engagement from plan to evidence to a signed submission package, for CCTS and ADEETIE. Every value is traced to its source, every number is computed by code, and every decision has a name on it.
For ACVAs · Energy auditors · Verification teams
18,247 t
“Quantity dispatched 18,247 MT”
How it works
One path, whatever the scheme. The methodology pack decides what each step requires.
Rate the risk factors, set the sampling policy and get the plan approved. Every save is a new version.
PDFs, scans, spreadsheets, SCADA exports and evidence packs, plus site observations, photos and meter readings.
Values arrive with their source. Declared figures are compared with the evidence and a person decides.
A versioned, hashed engine computes from accepted facts only, using the method the scheme prescribes.
Findings become issues. The entity answers through a link, and the verifier checks the response.
An independent technical review, two named sign-offs and a hashed submission package.
Capabilities
Not a document viewer with a model bolted on. Each part exists because a verifier has to answer for it.
A fact is committed only with a source: a document page and region, a spreadsheet cell, a time range of meter data, or a site observation. Nothing else is accepted.
Page
p.3 · region · verbatim text
Cell
Sheet2!D14 · formula kept
Time range
kiln_kw · 1–30 Apr · 720 rows
Observation
visit 2 · 14:32 · observer
The workbook says one thing, the invoice another. The gap is shown, measured against a tolerance, and never accepted quietly.
Rate scope, data and control risk. The approved plan sets the sampling rate, and a changed plan is a new version, never an edit.
Low-risk values can be accepted in one audited batch, but only after a deterministic sample has been reviewed clean. The share is disclosed in the report.
Observations, photos and meter readings captured on site. Text capture queues offline and syncs. The observer cannot be the one who accepts it.
Works through patchy connectivityPhotos online
Suggested, open, responded, verified, closed. The entity answers through a single-use link with no account, and blocking issues cannot be closed by decree.
A second verifier who did not draft or accept the findings reviews a fixed snapshot. If the numbers change, the approval goes stale and sign-off waits.
A deterministic ZIP with the report, findings, accepted facts and an evidence manifest of SHA-256 hashes. It states its own limits instead of hiding them.
Bring an agency's registry in from CSV or XLSX with a dry run first, then track each ADEETIE engagement from EOI to measurement and verification.
For ADEETIE, a baseline specific energy consumption, savings per measure and simple payback, computed from accepted values and nothing else. Financial ratios wait for the confirmed DPR format.
Schemes
Sector methodology lives in packs, as data, and the engine never branches on sector. A scheme without checked content stays marked as such.
Carbon Credit Trading Scheme
BEE scheme for energy efficiency in MSMEs
Embedded-emissions verification for installations. The flow and data model are drawn; the calculation rules, default values and verification requirements wait for a primary source and a practitioner.
Emission intensity per kilogram of hydrogen against a threshold. The boundary, renewable-matching rules and certificate format are still to be confirmed before anything is built on them.
Principles
Every result can be traced back to a source and a person. These are the rules the product will not bend.
The calculation engine is versioned and hashed. Language models never perform arithmetic.
A page, a cell, a time range or a site observation. Without one of those, the value is refused.
Material values always need a named reviewer. Nothing approves them automatically.
Reconciliation rules raise findings for a verifier to judge. Drafts cannot invent numbers.
The person who observed does not accept it, and a drafter does not review. Two people sign.
Every run records the engine version, pack version and a hash of its inputs, chained to the last.
Trust and security
Competing firms share this product, so isolation and an unalterable record are designed in, not added later. Controls we have not built yet are tracked, not hidden.
Row-level security on every tenant table. Signed-in users can read their firm's rows and write nothing directly.
Audit events, facts, runs and sign-offs cannot be edited or deleted, even by the server.
Files are hashed on arrival and stored privately under that hash. Their type is read from the bytes.
Four roles, one capability map, checked by the menu, the page and the route.
FAQ
Verification firms and energy auditors working on India's BEE schemes. It is built for the verifier and is never sold to the audited entity.
No. Models classify documents and propose field values with their source location. Only values a person accepts become facts, and only deterministic code turns facts into results.
Only low-risk ones, and only after a sample has been reviewed clean. The sample size follows the approved verification plan. Anything material, below the confidence threshold or contradicting a declared figure always goes to a person. Each batch is audit-logged, and the share accepted in bulk is disclosed in the submission package.
Only the one issue addressed to them, through a single-use link. They can write a response and attach files. They get no account, and they cannot see other issues, calculations or engagements.
Text observations and meter readings are queued on the device and sync when the connection returns, without duplicates. Photos need a connection at the moment of capture.
That depends on the scheme and the accreditation rules, so it is a setting on each methodology pack. When it is on, the review must come from someone who did not draft or accept the findings, and a changed calculation makes an earlier approval stale.
Not yet. The submission package is a generic, hashed bundle of the report, findings, accepted facts and evidence manifest, and it says so. Scheme-specific layouts will be added once the portal formats are confirmed.
Files are stored in your firm's workspace. When a Gemini API key is configured for your deployment, evidence may be sent to Google Gemini for classification and extraction. Extracted values are not used to train models in this application.
Request a demo
Tell us which scheme you work on. We will walk through one engagement from the plan to the signed package, with your questions in mind.
Prefer to start straight away?
Create your firm's account