01 and 02
Domain
Video, broadcast and production; networking and transport.
The engineering corpus
A cited corpus, loaded a module at a time.
The research runs on a modular corpus covering the overlap between media over IP and applied intelligence, the part most learning resources treat as two separate fields with two separate vocabularies. It is written to be loaded one module at a time, by a person or by a model, and everything non-trivial is cited to a publisher.
01
Fourteen modules, one router
Module 00 is the always-on core: identity, principles, conventions, citation policy, and a router that decides which of the rest to pull for a given task. The other thirteen load on demand, so the working context stays lean. Read top to bottom and the shape of the chain appears: signal, network, time, standards, transport, code, intelligence, agency, security, measurement, and what happens to all of it once it leaves the building.
- 00 Master brain
- Identity, principles, modes, conventions, citation policy, pedagogy, and the router.
- 01 Video and camera
- Colour and HDR, chroma, bit depth, codecs, frame rates and timecode, sensors, ancillary metadata.
- 02 IP networking
- Multicast and IGMP, QoS and DSCP, RTP, RTCP and SDP, FEC, ST 2022-7, fabric design.
- 03 Timing
- PTP (IEEE 1588), ST 2059, grandmasters, boundary and transparent clocks, BMCA.
- 04 ST 2110 and NMOS
- The 2110 suite, ST 2022-7 and 2022-6, AES67, NMOS IS-04 through IS-12, TR-1001-1, IPMX.
- 05 Contribution transports
- NDI in depth, plus SRT, RIST, JPEG XS, WebRTC, HLS and DASH, and how to choose between them.
- 06 Software engineering
- Secure code and services around ffmpeg, NDI and control planes; golden-vector testing.
- 07 AI and ML for media
- Vision and audio models, QoE and VMAF, encoding, learned coding, provenance, inference and MLOps.
- 08 Agentic media ops
- Model assistance and gated automation for operating media infrastructure.
- 09 Security
- Plane segmentation, control and timing plane security, the ML attack surface, incident response.
- 10 Calculations and labs
- Bitrate, storage, latency and inference maths; lab template; acceptance and observability.
- 11 Artifact templates
- Design doc, lab, runbook, threat model, ADR, tool spec, model card, checklists.
- 12 Source authority index
- Which authoritative document backs which module, where the coverage gaps are, and version flags.
- 13 OTT delivery, ad-tech and field QoE
- What happens to media and to the cue after the packager: the manifest as a control surface, the ad-signalling chain, the verification gap that chain leaves open, field quality of experience, experiment discipline, content protection.
Why the split at 05 and 13
Module 05 covers the left half of the chain, how media reaches the packager. Module 13 covers the right half, what happens to it afterwards, all the way to a player. Almost every interesting failure lives at the boundary between the two.
02
Six vocabularies, kept apart on purpose
Broadcast and software use the same words for different things: frame, stream, buffer, reference. Each glossary owns one sense and cedes the others to its counterpart. It sounds pedantic until the first time it saves you an hour.
03 and 04
Engineering
AI and ML engineering; software engineering and operations.
05 and 06
CueSeal
The operating vocabulary, including the three planes, and the cross-track engineering glossary.
03
Where the method meets a real build
Each appendix takes one concrete system through the full rigour: threat model, evaluations, acceptance criteria and artefacts, all from one build. These are the parts most worth reading if you only have twenty minutes.
Appendix A
Naming enforcer
A worked spec: deterministic rules, golden vectors, and the failure modes a naming layer has to survive.
Appendix B
Civic engine
A human-directed studio and civic video pipeline: a shared threat model, safety evaluations kept distinct from accuracy, and a two-stage binder against fabrication.
Appendix C
Verification desk
A provenance and verification workflow for a working newsroom, with the human gates drawn explicitly.
Appendix D
SCTE cue provenance
One ad break carried end to end and re-attested at every hop, the worked example behind the verification gap in module 13.
04
Three lenses over one spine
Each track charter scopes what that plane owns and where it defers to the shared record. The strategy note spans all three.
Charter v0.5
Genlock
Timing and control. PTP, ST 2110 and NMOS, and understanding a fault while there is still time to act on it.
Charter v0.4
MuxNode
Encode and deliver. Ladder economics measured against a fixed quality reference, then checked in the field.
Charter v0.6
Cuebus
Meaning and provenance. Extraction, verification, and attestation that survives the asset moving.
The lockstep note
Why the operability layer is where the durable engineering work sits, and what that implies for the model layer.
05
Recently added
The corpus changes most weeks, and is not finished. Here is what moved recently, partly so returning readers can skip what they have already seen, and partly because watching a knowledge base grow is a reasonable way to judge whether the work behind it is real.
- Module 13, OTT delivery, ad-tech and field QoE. The largest addition so far. It extends the corpus past the packager and introduces two asymmetries the rest of the work now runs on: a cue that is correctly signalled is not necessarily honoured, and lab quality is not field quality.
- A third evaluation axis. Capability and safety were already distinct. Module 13 adds field measurement, which turns a savings or accuracy claim into something falsifiable in the environment that actually pays for it.
- Cue standards corrected. The single-document SCTE 35 citation was replaced with the real current split across parts, and the vague SCTE 214 reference replaced with the specific carriage parts. Recorded here because a corpus whose job is preventing stale citations should say when it made one.
- Software-defined facility refresh. The media exchange layer moved past release-candidate status, with a named vendor implementation shipping; the time-addressable store is now referenced as prior art for the provenance ledger rather than reinvented.
- Standards tables re-verified against publisher pages, with the verification date treated as the most important line in each file.
Every version and edition here is re-checked at runtime. If you are reading this months later, assume at least one line has moved.
06
Available on request
The corpus is licensed CC BY 4.0 and the intention is to publish it openly. It is not a public snapshot yet, for a simple reason: it still moves fast enough that a snapshot would be out of date on arrival, and the source authority index in module 12 is mid-pass. So for now, ask and it will be sent.
Researchers, video and broadcast engineers, streaming and ad-tech people, and anyone working on trustworthy media infrastructure are all welcome to ask. Say roughly what you are working on and you will be pointed at the right modules first: three good modules beat fourteen unread ones.
Corrections welcome
Happy to hear from you even if you just want to argue with something. Corrections are useful, and a corpus that only circulates among people who already agree with it is not being tested.