the second film
a film for v0.2.0. it starts where the first one said three relays, shows what the release added, and puts the measured numbers on screen, with what they cost.
correction, 2026-10-16: the series of release v0.3.0 on revision 2078098 does not show relays on their own clocks bringing the attack to chance; they lower its auc to medians between 0.44 and 0.67, and five runs per point cannot show more. the claim of chance below is withdrawn; the current numbers are on the research page.
the film on the front page is new. it was made for v0.2.0 and it opens with the first film, which stops on the words “three relays.” and is corrected: there could be more.
what it shows
- five relays on the testbed today, and a chain of three drawn at random from them.
- relays that sign a descriptor, and a client that takes keys only from bundles it has verified.
- an onion key that changes in epochs, so a recorded setup cell stops opening.
- cells that are counted on every link: a copy, a gap or a swap closes the circuit.
what it measures
the second half is the correlation attack from the research page, drawn as a scene: ten flows through three relays, and an observer on two links.
- no protection: attack auc 1.000.
- cover traffic on top, at x2.99 bandwidth: still 1.000.
- a constant rate at the client, 70 ms: 0.951.
- relays on their own clocks, 66.5 ms: 0.523, close to chance, against an attack that only counts cells. the price is x3.00 bandwidth and a 185 ms round trip.
the series is preliminary: five runs per point. the film says so on every frame that shows a number.
what it does not hide
one card reads “today the last relay reads the message.” there is no recipient client and no end-to-end layer yet. the limits page has the rest.
the film is in english. the music is looming.sorrow.descent by sewerslvt; the opening keeps the music of the first film. the first film is still there, under the player.