LIFE AFTER CODE

AI writes the code. Should it ship?

VeriGreen checks software speed, security and estimated environmental impact before release.

NEW HERE? START WITH THE WORKING EXAMPLE

Same business result. Less wasted compute. Proof you can inspect.

We compare two versions of an included sample service: an inefficient version with deliberate security flaws, and an optimized version. Every run executes real code on generated sample records.

1 · Compare the codeRun both versions on the same sample input. See real time and memory measurements.

Start the demo to see a measured result. You do not need to upload code or connect an account.

What is real, what is estimated, and what do these terms mean?

Real measurementsElapsed execution time, Python heap memory, structural source-code checks, test execution and saved evidence.

Sample inputGenerated telemetry records for the included example. This demo does not currently audit an arbitrary uploaded repository.

Estimated carbon and costFormulas combine measured runtime/file bytes with supplied energy, region, image-size and pricing assumptions. They are not meter readings or invoices.

SCI / CCPAC / CWE / MCPCarbon per unit of software work / estimated cost per accepted change / a class of software weakness / the protocol through which agents invoke tools.

The region selector simulates grid-carbon assumptions; it does not move the running service. VERIFIED means a local check passed. It is not proof of an autonomous GitLab agent run.

Loading persisted evidence from your backend…Measured runtime · modeled carbon & cost

SCI CARBON SCORE↘

—gCO₂eq / 1k req

Before —Awaiting measurement

COST PER ACCEPTED CHANGE$

—CCPAC / change

Manual baseline —Signed audit estimate

AI —CI —Canary —

EMPIRICAL PERFORMANCEϟ

—latency speedup

Awaiting heap measurement

SECURITY & RELEASE EVIDENCE◇

—detected production CWEs

Awaiting structural security audit

SHA-256 signed auditAwaiting ledger

Carbon Estimated from measurements + reference inputsCost Estimated tokens/time × reference ratesSpeed & memory Measured on this runSecurity Structural code checks; no CVE scan

01 / EMPIRICAL VERIFICATION

Run the workload. See the difference.

Same business output. Real elapsed time and Python heap measurements.

PATH B · BEFORE → AFTER

Live before and after workload comparison
Measurement	Legacy / Before	Remediated / After
Waiting for measured workload evidence.

Benchmark evidence is generated live; no canned scores.

02 / GOVERNED DELIVERY

Nine stages. One evidence trail.

Inspect a stage to open its actual, integrity-checked JSON artifact.

Awaiting pipeline state

Stage artifact inspector

Select any stage to inspect its actual artifact.
Missing evidence is generated by a real local audit.

Readiness verification does not deploy Cloud Run.

03 / AGENT EXTENSIBILITY

Give the MCP tools a real task.

The same four tools used by the GitLab Duo flow, exposed through JSON-RPC 2.0.

HTTP ↔ MCP ↔ SQLITE

JSON-RPC 2.0 response

Invoke a tool to inspect the live protocol response.

04 / DURABLE PROOF

Evidence that survives the request.

Signed governance snapshots and empirical runs, persisted in SQLite.

AUDITABLE BY DESIGN

Persisted SQLite benchmark history
Run ID	Timestamp	Records	Region	Legacy ms	After ms	Speedup	Before SCI	After SCI	Gate
Loading SQLite history…

Awaiting historical runs

ACT2_GOVERNANCE_SUMMARY.md · raw evidence

Execute the full audit to generate the executive Markdown ledger.

Active SHA-256 Evidence SignatureAwaiting persisted ledger