Draft · research

EDDIE UA Lab – synthetic customer journeys and separate processes

Зміст статті / On this page (13)
  1. Why a laboratory matters
  2. Browser-only playground
  3. Local three-process integration testbed
  4. Ten acceptance scenarios
  5. What has been demonstrated
  6. Next technical gate
  7. A polished button is not proof of national infrastructure
  8. Permission revocation is more than a Boolean
  9. Corrections must remain explainable
  10. Transfer without contradictory authority
  11. What a reproducible test proves
  12. Next adversarial scenarios
  13. Explore

Why a laboratory matters

Policy statements about permission revocation, provenance or safe transfer between nodes should be reproducible tests rather than assumptions. The lab makes negative outcomes visible: unauthorised reads must fail, old readings must remain traceable after correction, and a stale operator must not write a new authoritative record after transfer.

Browser-only playground

Open the synthetic playground

An invented commercial customer is linked to a fictional metering point. The user simulates representative verification, grants an aggregator access, reads a fabricated 100.0 kWh value, corrects it to 97.5 kWh, transfers technical responsibility A → B, revokes the grant and simulates directory failure. The entire model runs in the browser tab; it does not legally authenticate an identity or connect to a live server.

Local three-process integration testbed

The Python integration bridge starts an authority directory and two distinct HTTP node processes on 127.0.0.1, each with an independent SQLite file. Operations are handled by a synthetic directory with recorded durable intent. Stopping the directory is a real subprocess event and access fails closed. It is a one-machine laboratory, not a multinational or multi-organisation federation.

Developers with access to the repository can run: python integration/bridge.py --port 18200. The resulting loopback port must never be publicly exposed. It is not an Ukrenergo DataHub endpoint.

Ten acceptance scenarios

  1. Denied access where permission or other lawful basis cannot be demonstrated.
  2. Permission scoped to the correct recipient and data category.
  3. Authorised reading returns value, interval, unit, revision and provenance.
  4. Revoked permission cannot authorise a new read.
  5. Correction creates a new version and preserves history.
  6. Node A → B transfer increments authority epoch.
  7. Stale A activity is rejected after transfer.
  8. Directory-process failure blocks new unauthorised actions.
  9. Restart, retry and invalid-message cases use durable recorded state.
  10. Every test has reproducible input, result and code/evidence reference.

What has been demonstrated

Prior GitHub Actions runs have passed relevant synthetic test suites on known PR commit hashes, including the integrated build and browser checks. That proves specified software behaviour under tested conditions. It does not prove production split-brain safety, independent cross-organisation identity, fully managed PKI, a lawful metering entitlement, throughput at national scale or compliance certification under EU Regulation 2023/1162.

Next technical gate

Real independent nodes and certificates, signed receipts, network partition and crash injection, secure key rotation, legal identity integration, external test implementations, a qualified security review and a lawful pilot with limited real data. There is no market production pilot yet.

A polished button is not proof of national infrastructure

A browser demo is an excellent way to show a customer journey: invented company, synthetic EIC, grant, reading, correction, node A and node B. But a sequence performed in the browser's memory proves only the illustrative state machine. It does not establish lawful identity, production security or access to official metering data.

EDDIE UA also has a localhost experiment with an authority directory and two separate HTTP/SQLite node processes. A process can actually be stopped to explore failure handling. However, all processes still share one physical environment and synthetic credentials. This is not a network of accredited independent organisations.

Permission revocation is more than a Boolean

A simplified model flips granted from true to false. A real access system must determine when the decision became effective, how other nodes learn about it, how in-flight queries are treated, whether old history may be downloaded again and which independent lawful bases still apply.

Tests should verify denial and preserved audit evidence after revocation, including failures of the directory. The exact legal effect and allowed exceptions cannot be invented by laboratory code.

Corrections must remain explainable

The synthetic example begins with 100.0 kWh and later corrects the interval to 97.5 kWh. Simply overwriting the previous value would make two different customer analyses appear inconsistent without explanation. Revision, source, publication time and correction lineage should remain traceable according to the relevant legal and retention rules. Laboratory provenance does not confer statutory settlement authority.

Transfer without contradictory authority

Moving technical representation from node A to B is difficult when messages may be delayed or lost. If both claim active authority, a split-brain condition emerges. The local research tracks an authority epoch and durable operation records to explore stale actions and recovery. Production evidence would require separate machines, independent keys, signed receipts, partition and crash testing, and a resilient trustworthy directory.

The objective is a safety property under a clearly defined fault model, not an impressive transfer animation.

What a reproducible test proves

A useful test records setup, operation, expected and observed outputs, code version, environment and negative alternatives. Synthetic unit tests, multi-process integration tests, independent penetration testing and legal accreditation are distinct levels of evidence and must never be conflated.

Public demonstrations should use fabricated identifiers and should not collect real EIC codes, electricity histories or customer agreements. A simulated representative-verification button is not actual legal authentication.

Next adversarial scenarios

Partition the network, delay messages, replay permissions, duplicate requests, rotate or compromise test keys, crash between snapshot preparation and authority change, create clock drift, exceed audit storage, and test failure during recovery. Only external, reproducible evidence can justify specific resilience claims. Lawful rights to real customer data remain a separate prerequisite.

Explore