Evidence of what you’ve done, including how you’ve grown.
FromStar rating→ToReputation as proof
Star ratings gave digital commerce a simple, shared signal of quality. Open networks can build on them with verified evidence that travels with the participant: what they have demonstrably done, in what context, and how they have grown.
Same number, different stories
Two sellers are both rated 4.3 stars.
The first has delivered 2,000 orders. Local deliveries are reliable and remote ones less so, and every dispute is resolved within days. The second is Imran. He had a difficult first year, took up capability support (Shift 6), and has delivered 250 orders since with almost no delays illustrative.
The number is the same. The stories are very different. A buyer deserves to see the story.
Where ratings help, and where they run out
Star ratings gave digital commerce something it badly needed: a simple, shared signal of quality, and they have helped millions of people choose with confidence. They also have a known limit. When almost everyone sits near the top, the number cannot tell sellers apart.
A study of eBay data found that the average seller’s share of positive feedback was 99.3%, with a median of 100%, so that a score of 98% put a seller below the tenth percentile. In October 2011 only 0.39% of transactions drew negative feedback, yet more than 1% had a dispute ticket opened. A different measure the authors constructed averaged 64% and varied far more between sellers[1]. One platform and one year do not describe every network, but the pattern shows how a rating can compress and how much a bad experience can go unrecorded.
Reputation as a set of proofs
| Trust dimension | What it tells a buyer |
|---|---|
| Fulfilment reliability | Do they deliver what they commit? |
| On-time performance | Do they meet agreed timelines? |
| Disputes | How often do transactions reach formal dispute? |
| Verified outcomes | How often was responsibility actually established? |
| Resolution record | Do they honour resolutions? |
| Protected payment | Will they transact under protection? |
| Capability | Have they shown they can do this kind of work? |
| Recent performance | Is performance improving or declining? |
Go deeper
Four qualities of fair reputation
Contextual. A logistics provider may be excellent for local deliveries and small parcels but new to temperature-sensitive goods. Reputation answers: reliable for what, under what conditions?
Portable. Your reputation belongs to your participation in the network, not to the app you used to reach it. Move to another app, and your verified history comes with you.
Fair. A complaint alone never lowers reputation. Only verified outcomes do. That stops reputation from becoming a weapon and keeps people willing to use the dispute process.
Recoverable. History isn’t erased, but it is read in context: how recent, how severe, how frequent and what happened after. It tells a trajectory: difficulties, then support, then new processes; then improving and verified; then consistently reliable. A trust system that remembers failure but can’t recognise improvement eventually becomes a system of exclusion.
A new incentive
The old loop: poor performance, then a bad rating, then lower visibility.
The new loop: poor performance, then diagnosis, support, verified improvement, recovered reputation and better discoverability.
The system doesn’t only punish failure. It gives everyone an economic reason to improve.
What has already been proposed
ONDC’s 2022 consultation paper proposed that buyer and seller apps share ratings with independent agencies, which would compute scores for sellers, logistics providers and apps from past scores and dispute history. The scores would be portable, so sellers keep their reputation if they switch apps, and the agencies would work “in a manner similar to credit rating agencies.” The paper left open how to avoid or remove fake reviews and scores[2].
This shift keeps that direction and changes what is carried: not a single score but a set of verified proofs, each in its context.
Getting it right
Privacy first. Reputation shows summaries of verified behaviour, never private transaction details.
Old records fade. There are clear rules on how long past failures are shown, and older events carry less weight. Those rules have to sit inside India’s data-protection rules, which are being phased in from 2027[3].
Correctable. Participants can see their own record and challenge anything inaccurate.
Governed openly. How reputation is calculated is transparent and decided with participants.
What could go wrong
- Too much to read.Eight proofs may be more than a buyer will take in, so the display has to summarise without flattening.
- Gaming.Manufactured transactions and fake evidence can raise a record, and portability raises the payoff.
- Who computes it.Whoever calculates the profile holds power, so the method has to be open.
- The stigma of support.A seller who took help and recovered must not look weaker for having done so.
How we’d test it
Proof-based profiles for sellers in one category, shown alongside existing ratings.
- The test
- Proof beside ratings
- What we would measure
- Whether buyers find the profiles more useful, how they affect choices, and whether recovered sellers regain business.
- What would end the idea
- Buyers ignore the profiles, or recovered sellers don’t regain business.
Buyers ignore the proof, or recovered sellers don’t regain business.
Sources
- [1]Nosko and Tadelis, The Limits of Reputation in Platform Markets: An Empirical Analysis and Field Experiment, NBER Working Paper 20830 (2015), pp. 2 and 9Checked; eBay data from 2011
- [2]ONDC, Building Trust in the ONDC Network: Consultation Paper (29 September 2022), section 4.7.4Checked; proposals for comment, not adopted policy
- [3]SCC Online, Digital Personal Data Protection Rules, 2025: key highlights (26 December 2025)Checked via a legal summary; the Rules’ own text not read
Spotted a source we missed or a rule we misread? Tell us what we got wrong.
Each of these records is also a lesson. The last shift is about learning from all of them together.