Evidence should be a by-product of commerce.
FromEvidence after the fact→ToEvidence as a by-product
By the time a dispute is raised, the most important moment has passed. If the record is built quietly, step by step, while the transaction happens, the facts are already there if anyone needs them, and each one says who recorded it and how it can be checked.
By the time a dispute is raised, the moment has passed
“This isn’t what I ordered.” “That’s not what we agreed.” “It arrived late.” “The delay wasn’t our fault.”
Now someone has to rebuild what happened from order records, messages, invoices, photos, delivery logs and payment records scattered across different systems. Each holds part of the truth, and none necessarily holds the whole story.
Imagine the record already existed. When Kavita confirms her order, the network writes down what Imran offered, what she accepted, at what price, by when, and what counts as delivered. Then every step adds an entry: order accepted, payment protected, production started, dispatched, delivered, inspected, accepted, money released. Nobody has to reconstruct anything, because nobody has to remember.
Start with commitments, not orders
Before money moves, the record holds an explicit set of commitments: what was offered and accepted, at what price, under what conditions and by when, what counts as fulfilment, and when payment is released. An order says that something was bought. A commitment says what each side promised.
The record is proportionate by design. A ₹500 purchase doesn’t need the evidence trail of a ₹5 crore contract, so the detail scales with the stakes.
A claim is not evidence
If a seller says “I delivered it,” that is a claim. If a logistics provider records “delivered at 14:32 at the agreed address,” that is evidence. So every event carries its provenance: what happened, who recorded it, when, in what context, and how it can be verified.
Even a third party’s record can be wrong, so each piece of evidence also carries a confidence level, and either side can challenge it.
illustrative Constructed to explain the idea, not data.
| Event | Who can record it | What kind of record |
|---|---|---|
| Order accepted | Buyer app and seller app | Both sides, on the same terms |
| Payment protected | Payment provider | Third-party record |
| Production started | Seller | A claim, supported by a timestamped photo |
| Dispatched | Logistics provider | Third-party record |
| Delivered | Logistics provider, confirmed by the buyer | Third-party record, open to challenge |
| Inspected and accepted | The buyer, or an inspection service | A claim by the buyer, or a third-party record |
Go deeper
Many sources, no single owner
Buyers, sellers, payment and logistics providers, inspection services, devices, certification bodies and timestamping services can all contribute to the record, without any one of them owning the whole of it. That fits the shape of an open network, where each participant already holds a piece of the truth.
The seed of it exists. ONDC’s 2022 consultation paper describes a transaction-level contract that creates binding obligations between apps with no prior relationship, and asks sellers to declare return, refund and cancellation terms upfront so that they are coded into that contract[1]. Those are commitments in embryo. This shift asks what happens when the rest of the transaction is recorded the same way.
Existing isn’t the same as visible
A record can exist without being public. It is shared only with the parties to the transaction, authorised resolution providers, the relevant payment providers, and regulators where the law requires. Reputation (Shift 7) uses summaries of verified behaviour, never private transaction histories.
The record is never surveillance. It captures what the transaction needs and nothing more, and both sides can see their own record as it builds.
How long to keep it
Three clocks pull in different directions. India’s data-protection rules say personal data must be deleted once its purpose is fulfilled unless the law requires it to be kept, with notice to the person before deletion, and they require certain logs to be kept for at least a year[2]. A consumer can file a complaint with a consumer commission within two years of the cause of action[3]. A useful record has to outlast the claims it might be needed for, and no longer than that.
The Rules’ main operative provisions take effect 18 months after their publication on 14 November 2025, around May 2027[2]. We have not set a retention period. Settling one, with counsel, is part of the pilot.
What could go wrong
- Surveillance creep.A record that captures more than the transaction needs turns into monitoring.
- Bad evidence treated as truth.A wrong third-party record can decide a case unless it can be challenged.
- Effort for small sellers.Taking photos and confirming steps adds work for a six-person workshop.
- Format adoption.Independent apps would have to record events in a common form.
How we’d test it
Structured commitments for one transaction type, plus four or five key fulfilment events.
| The test | What we would measure | What would end the idea |
|---|---|---|
| Events recorded | Share of key events recorded by the party best placed to record them | Parties record fewer than half of the events they could |
| Resolution from the record | How often a dispute can be resolved from the record alone, time to resolution, and how both sides rate the process | Most disputes still depend on facts the record doesn’t hold |
Sources
- [1]ONDC, Building Trust in the ONDC Network: Consultation Paper (29 September 2022)Checked; proposals for comment, not adopted policy
- [2]SCC Online, Digital Personal Data Protection Rules, 2025: key highlights (26 December 2025)Checked via a legal summary; the Rules’ own text not read
- [3]Consumer Protection Act, 2019, section 69 (limitation)Checked
Spotted a source we missed or a rule we misread? Tell us what we got wrong.
A record that is ready is also a way to see trouble coming.