Recording the keynote: backup strategies that have saved us

Keynotes happen once. The client is paying for a recording that will be used in marketing, in internal communications, in next year's pitch deck. The pressure to nail the recording is large, and the failure modes are unforgiving — a corrupted SD card, a stuck record button, a sound problem on the recorded mix.

We use a layered approach to keynote recording. The principle is that no single failure should be able to destroy the recording. If one layer fails, another picks up. If two layers fail, the third is still running.

Here is the actual stack we use.

Layer one: the in-camera recording

Each camera records to its own internal media. SD cards, CFexpress cards, or internal SSDs depending on the camera. This is the highest-quality version of the footage, and it captures each angle in full.

The risks here are media failure, file corruption, accidentally formatting the card before offload, and the operator forgetting to hit record. We mitigate the operator risk with a record check during the talker check, a visible recording indicator in the viewfinder, and a checklist that the camera operator runs at the start of each program segment.

Layer two: the camera ISO recording

In addition to the in-camera recording, we route each camera's output to an ISO recording on the video switcher or a separate recorder. This gives us a second copy of each angle, on a different physical device, that we can fall back to if a camera card fails.

The ISO recording is usually at slightly lower quality than the in-camera version — the switcher's recording bitrate is typically lower than the camera's. But "slightly lower quality footage we can use" is dramatically better than "no footage of that angle."

Layer three: the program recording

The switched program output goes to its own recorder. This is the version that is mixed and ready to share — multi-cam cuts, lower-thirds, transitions, the works. It is what the client wants for distribution. It is also useful if a single camera failed completely, because the switcher would have cut away from that camera anyway.

Layer four: the audio recording

The audio recording is a separate concern. The program audio is recorded to a multi-track recorder, with each input on its own track. The mixed program audio is also recorded as a stereo or 5.1 mix. Both go to the same hardware recorder, with the multi-track being the source of truth and the mix being a reference.

The risk is that a microphone problem in the room — a noise on the lavalier, a wireless dropout — does not get noticed during the show and ends up in every layer of recording. We mitigate this with real-time monitoring at the mixer, and with the multi-track giving us the ability to re-mix in post if a channel needs to be replaced with audience-captured audio.

Layer five: the audience-camera footage

A keynote will generate dozens or hundreds of phone videos from the audience. Most of them are unusable. A few of them are surprisingly good — especially the ones from the front rows. We capture this informally by asking the client's social media team to forward us anything that looks shootable, and we treat it as a last-resort layer if something catastrophic happened with the production cameras.

This has rescued exactly one event in our history. Once is enough to justify the practice.

The offload protocol

Recording is only half the battle. Getting the recordings safely off the cards is the other half. Our protocol:

• Cards do not leave the stage area without being checked into a labeled bag
• A dedicated assistant offloads cards to two separate hard drives during teardown
• The offload computer verifies file integrity before the card is reformatted
• A backup of both drives goes to cloud storage before the team leaves the venue
• The original cards are kept untouched for a week, in case the offload was corrupted

A surprising number of clients have lost recordings because someone reformatted a card without verifying the offload first. The protocol exists to make that impossible.

What we promise the client

We tell clients we will deliver clean recordings of the program from every camera angle plus a finished program cut, plus multi-track and mixed audio. We do not promise zero failures — that is impossible. We promise that no single failure can destroy the deliverable.

That promise is what the redundant stack is for. The cost of the additional recorders, cards, and offload labor is small relative to the value of an unrepeatable keynote. Clients who have had a recording disaster understand this without further explanation. Clients who have not, we tell the stories to.