Hello,
I have been working on a DMR-concept for people who develop and sell decks, and I would like to head about your feedback here before taking it further.
Main motivation is awareness creation rather than enforcement: a licensed deck should be visibly licensed, a copy that circulates should be attributable to the delivery it came from, and passing one on should be a deliberate act rather than a side effect of a file being trivially copyable.
But let’s start with what it does not do, because that is the part I expect people
here to care and argue most about:
What it does not do
It does not stop anyone from extracting and sharing deck content. That is not reachable for any scheme of this kind and is not attempted. Once a person can study a deck, they can copy it.
The add-on makes no network requests at all: no account, no license server, no device registration, no telemetry. It verifies a signature offline against a key compiled into the build.
It does not modify Anki, fork it, or need a separate sync server. It does not touch the scheduler, does not block exporting, and only ever rewrites notes it delivered itself, tracked by GUID.
What it does
Encryption. A deck can be delivered locked and is opened once with a passphrase that comes with the purchase. After that it is an ordinary deck: it syncs to AnkiWeb, AnkiDroid and iOS, it is searchable, editable and exportable, and the add-on is no longer involved and can be uninstalled. Thecards table is never rewritten, so intervals, due dates and review history survive unchanged.
Marking. Every delivered copy differs slightly: note GUIDs, tags, zero-width characters, and watermarks in images and audio. A copy that turns up where it should not can be traced back to the delivery it came from. The identifier uses a collusion-resistant fingerprinting code (symmetric Tardos), so several buyers comparing their copies and removing the differences still
leave one of themselves identifiable.
The two are independent. Marking works on an unencrypted deck too, so an author who wants no barrier at all can still ship something attributable.
The honest limits
A finding gives attribution, not proof of authorship. The publisher holds every key and the master deck, so they could produce a file mcarrying any buyer’s dentifier. A match says which delivered copy the material corresponds to, not who passed it on, and every report says so.
A fully locked package carries no readable marking, because the marking sits inside the ciphertext. Authors who want locked files traceable can leave media unencrypted; the tool states that trade-off rather than hiding it.
Passing on the passphrase defeats the encryption entirely. That is by design:
the barrier is meant to make handing a deck on deliberate, not impossible. The marking is what remains afterwards.
Capacity depends on the deck. Text-only decks below roughly 2,000 notes are limited to channels that a CSV round trip destroys. The measurements, and the attacks they were run against, are in the repository.
Feedback I am after
- Whether the add-on behaves as an add-on should: hooks used, what it writes,
what it leaves alone. - Whether the limits above are clear enough for someone deciding whether to buy
a deck delivered this way. - Anything in the marking that would break in normal use.
It has been run against real Anki rather than only against test fixtures, but it has not been through AnkiWeb review, and I would rather hear objections here first.