Where the records live and who can touch them.
This page is for the person who has to approve VachanSetu™ before it holds consent records. It describes the mechanisms, and says where they stop.
Consent records stay in India.
The primary database is hosted in India. So are its backups, and so is the archive. Disaster recovery stays in India as well.
The decision log, the webhook delivery log and the evidence packs are produced from the same database and sit in the same place.
The data processing addendum names the processing location and the sub-processors.
The ledger accepts new records and refuses edits.
The ledger is protected in layers. Each one below catches what the one before it could miss, and the last two sit outside the application.
| Layer | Mechanism | What it stops |
|---|---|---|
| Application | The service has no update or delete path to ledger tables. It can only append. | A bug or a misused admin screen rewriting a consent. |
| Database | Ledger tables belong to a role that cannot log in. The service role may insert and read. A trigger refuses updates and deletes on every partition. | A compromised service account altering stored records. |
| Signatures | Each consent is signed with the organisation's Ed25519 key before it is stored. Each block is signed too. | An edited record that still looks plausible. The signature no longer matches. |
| Sealing | Consents are sealed into Merkle blocks. Each block carries the hash of the one before. | Removing or reordering records. The chain breaks at that point. |
| Independent timestamp | Hourly, a timestamp authority signs a root that covers every block sealed in the hour (RFC 3161). | Back-dating, or rebuilding the whole chain after the fact, by us or by anyone with server access. |
| Archive | Database logs are archived continuously to object storage with a compliance-mode lock. The application holds no credential that can delete from it. | Losing history to a destroyed server, or to an attacker who has the application's credentials. |
| Scheduled verification | A job recomputes signatures, chains and timestamps and raises an alert when any of them disagree. | Tampering that nobody looks for. |
Records are canonical JSON, hashed with SHA-256. A block is sealed within a minute of the consent. Anyone can re-run the checks with the public keys and a published verifier. How verification works
Erasing a person leaves the chain intact.
The ledger never holds a name, email address or phone number. It holds a random pseudonym for each person. Your own subject id is turned into that pseudonym with a keyed hash, so the raw id is not stored either.
Contact details live in a separate vault. Each person's details are encrypted with their own key, and that key is wrapped by an organisation key. Decryption happens only in the application service.
To erase a person, we delete their key. The details can no longer be read by anyone. The signed records and blocks are untouched, so every receipt and every block signature still verifies. How long proof of consent is kept after erasure is a question for your counsel, and it is a setting.
- In the ledger
- Pseudonym, purposes, notice version, language, channel, signature
- In the vault
- Name, email address, phone number, encrypted per person
- On erasure
- The person's key is deleted. The vault entry becomes unreadable.
- After erasure
- Receipts and block signatures still verify
Keys sit with the application service, away from the data.
Each organisation has its own signing key and its own key for wrapping vault keys. They are held by the application service and reached through a key-provider interface. They are never stored in the database that holds the records.
A database dump therefore contains signed records and encrypted details, and no key to forge or decrypt either.
Rotation adds a new key with a validity window. History is never re-signed. Old receipts verify against the key that was current when they were made, and public keys are published with their windows.
The key ring is backed up offline by a named person. A provider backed by an HSM or a managed key service can replace the current one without a change to stored data. Bring your own key is planned.
Access is by role, with two steps for staff.
- Sign-in
- Staff sign in through the organisation's identity provider. Every staff role requires two-step verification.
- Roles
- Data Protection Officer, Compliance, Product owner, Developer, Grievance officer and Auditor each hold only the privileges their work needs. The Auditor role reads and exports.
- Tenant isolation
- Every table carries the organisation and has a row-level security rule. A query for one organisation cannot return another's rows, whatever the application code asks.
- Public clients
- Browsers receive publishable keys bound to listed origins. Those keys can render a notice and record a grant or withdrawal.
- Server clients
- Server systems use OAuth client credentials. Access tokens last fifteen minutes. Writes take an idempotency key and are rate limited.
- Audit trail
- Privileged actions by staff are logged.
Operations staff have shell access to the servers. They hold no key material beyond what the application service needs to run. All traffic uses TLS.
What the design does not prove.
A valid receipt shows that the organisation's key signed exactly these bytes, that they sit in a signed block, and that the block belongs to one unbroken history. It shows no more than that.
- That the person who clicked is who they say they are. The ledger holds a pseudonym. Identity checks are your process, and the signed record shows only what that process sent.
- That a consent was given if it was never submitted. The ledger proves what it received.
- That your other systems honour a withdrawal. The decision log and the webhook delivery log show what they were told and whether they acknowledged it.
- A record from before an organisation started using the ledger.
What you can check today
ISO 27001 certification and a SOC 2 report are targets. We have completed neither. An external penetration test is also a target and has not been done.
What you can examine now is the ledger itself. The specification is public. The verifier runs in your browser on a worked example, and a command line verifier checks your own evidence pack offline, with none of our servers involved.
If your policy needs a certificate before onboarding, you should wait for one. If it accepts evidence you can test, start in the sandbox.
Send this page to your security team.
Ask them to test the sample receipt, read the specification and put their questions to us. The sandbox is free and runs in test mode.