← Showcase homePage status: Research · nothing operational

ERC · Evidence, readiness and custody

Election oversight that shows its evidence

ERC keeps a claim-by-claim record of what has actually been tested in BB2G's election work — and says plainly what has not. It is an evidence programme. It never counts votes.

Where things stand · 24 September 2026

Nothing here is approved for real elections

Status: no election component is operational or approved for use with real people or real data. Every repository reviewed describes itself as synthetic-only or scaffolding, or has open approval gates.

Decisions already on record

  • Permanently non-tabulating. ERC Evidence never counts, stores or displays ballot choices.
  • Posture shown as “vulnerable”. Every election component displays this until an independent, named cryptographer signs off.
  • Biometrics are research only and are never a precondition to vote.
  • Identity stays out of ballots. Identity or biometric data never enters ballots, cast-vote records, receipts or audit exports.

Open blockers

  • Several codebases each claim to be the canonical election build; one must be chosen before anything else proceeds.
  • An independent cryptographer and election-law counsel are required. These are people, and cannot be engineered around.
  • Test results claimed in earlier notes have not all been reproduced; the register records the differences rather than hiding them.

Demonstration

Follow a synthetic custody record

The demo seals a synthetic ballot container, records each hand-off, chains every event with a SHA-256 fingerprint, and then lets you alter a record to see the check catch it. It also shows what a fingerprint cannot do: it detects changed bytes, not physical tampering or theft.

Simulated Synthetic containers and counts only. No ballot choices, voter records or identities.

Silent screen recording of the synthetic demo, about a minute long, with captions.

Read the transcript
  1. 00:00 This is the ERC chain-of-custody demo. The container, people and counts are all synthetic.
  2. 00:03 First the container is sealed with 250 sheets recorded.
  3. 00:06 Each hand-off is recorded and fingerprinted with SHA-256, linked to the record before it.
  4. 00:11 The scanner counts 249 sheets. The difference becomes an exception for a person to review — not an accusation.
  5. 00:15 A reviewer finds the missing sheet, rescans it, and records why the counts now reconcile.
  6. 00:19 Verifying the chain: every fingerprint matches and every link is intact.
  7. 00:22 Now record three is altered to say 245 sheets, without updating its fingerprint.
  8. 00:25 Verification catches the change and flags the record. A fingerprint detects altered bytes — it cannot detect physical theft.

Interfaces under review

Platform screens, validated with synthetic data

Screenshot of Human verification research
ResearchCivic oversight

Human verification research

A research plan for voluntary, consented, one-to-one check-in that keeps presence, identity, session integrity and eligibility as four separate decisions — organised as 21 checks and controls.

For: Researchers, accessibility reviewers and security reviewers

Next step: try the custody demo yourself, or read the human-verification research.