Engineering write-ups

How this platform is actually built

Long-form notes on the decisions behind the platform: why delivery is compiled rather than bought, where the boundary between rented and owned belongs, and what each of those choices costs. Written for the engineer who was sent a link, not for a brochure.

Reference architectures Talk to us

Delivery18 August 20264 min read

Why we compiled our own delivery edge

A general-purpose cache treats a four-gigabyte film like a large web page. Video is not that — viewers seek, abandon and resume, and every one of those is a range request. This is what changed when we stopped renting the tier that answers them.

edge
caching
http3
egress

Read the study

Architecture11 August 20262 min read

Where the boundary between rented and owned belongs

Most streaming services rent everything because it all arrived in the same bundle. The line worth drawing is not between cloud and metal — it is between the tiers that spike and the tiers that bill by the gigabyte.

hybrid
aws
cost
architecture

Read the study

Delivery29 July 20262 min read

The mid-film seek is the whole cache design

Everything interesting about a video cache follows from one interaction: a viewer dragging the scrubber into the middle of an object nobody has requested before.

caching
range-requests
origin

Read the study

Security14 July 20262 min read

Authorise the segment, not the session

Most streaming security checks entitlement once, when playback begins, and then serves everything that follows to whoever asks. That makes a copied manifest URL a working copy of the film.

drm
tokens
aes-128
edge

Read the study

Practice30 June 20262 min read

The console change nobody wrote down

Estates do not become unreproducible through a decision. They get there through a sequence of individually reasonable exceptions, each made at two in the morning by someone doing the right thing.

terraform
gitops
operations

Read the study

Encoding12 June 20262 min read

The transcoding bill that multiplies

Per-minute transcoding pricing looks reasonable until it is multiplied by ladder depth, and then again by every re-encode. That is how catalogues end up stuck on ladders nobody would choose today.

encoding
gpu
cost
hls

Read the study

Which of these is your problem?

If one of these write-ups describes the thing you are currently paying for, that is the conversation worth having. Send your current setup and the bill that goes with it.

Talk to us →Reference architectures →