Where the data came from, and who agreed to it.

Our lawful basis is consent, the primary route, and the one contributors and sites give. The Act's other grounds are a closed list of consent-free exceptions for things like State functions and employment purposes, and there is no legitimate-interests ground for a private company to balance-test against. Consent works for people we record on purpose. Nothing reaches a stranger walking through a shop.

So the answer is procedural: capture zones agreed with the site, signage, an audible recording indicator, a workflow that prompts the contributor to stop. If someone is recorded anyway the episode is redacted or held back. Since automated detection misses around one in ten, the choice of environment is the control that does the work.

The provision with the most commercial consequence, and the one vendors tend not to raise. Consent is withdrawable at any time. Processing already done stays lawful, but §8(7) requires the fiduciary to erase the data on withdrawal, and to cause its processors to erase it too. Cessation isn't the end of it; erasure is.

Which makes one term load-bearing: whether you hold delivered episodes as an independent fiduciary or as our processor. If the latter, erase-through reaches your copies. Our per-episode chain means we can identify every episode from a withdrawing contributor and tell you which batches contain them.

Rules notified November 2025, phased. The Board exists on paper: constituting rules commenced, though no Chairperson or Members were appointed as of mid-2026. Consent Manager registration opens November 2026. Substantive obligations bite May 2027, and until then the older SPDI regime runs alongside, which, unlike DPDP, treats biometric information as sensitive.

One detail easy to get backwards: notice must give the data principal the option of English or any of the twenty-two scheduled languages. It isn't a menu we pick from on their behalf.

Today, yes: §16 is a blacklist model, transfers permitted unless the Government notifies a restricted destination, and nothing has been notified. Permissive by default rather than by approval.

The risk to name is Rule 13, in force from the same May 2027 date, letting the Government require specified data to stay in India for entities designated Significant Data Fiduciaries. A high-volume video operation is a plausible candidate, and the single lever that could change how this business delivers.

Yes to both, written down rather than a research licence renegotiated at deployment. Exclusivity is available, priced per schema cell rather than across the whole batch. Redistribution and sublicensing are settled in the agreement and survive an acquisition.

Consent comes from the contributor and from the site. An arrangement with someone's employer is not consent from the person wearing the camera.

Paid per episode that passes, at a rate set so someone working the protocol properly earns more than the going local rate. On its own that pushes all the risk onto the person least able to carry it, so there are two floors: a session minimum for booking a kit and working the protocol, and a fault clause paying in full when a rejection traces to our equipment, calibration or coaching.

A flawless recording with a broken chain is rejected under RTS-01. It doesn't enter the batch and it doesn't get paid for.

If our reason for collecting here were labour cost, it would stop being a reason the moment somebody bid lower. The reason is that reaching real workplaces means sending the kit out, and sending the kit out means checking what comes back.