Podcast, Production Plan

01  How the show is made

The show is recorded, not broadcast. Almost nothing is decided in the room.

Every camera rolls continuously. Ryan records himself. All audio records separately. Every choice about what the viewer sees, which angle, when to show Ryan full screen, where B-roll goes, happens in Premiere afterward.

• No switcher. A switcher makes one finished feed for a live audience. There isn't one.
• Full-screen Ryan is an edit. We have his own recording as a separate file. When he talks, we cut to it in Premiere.
• AI segments are prepared, not live. Generate between episodes, drop into the timeline. Real-time generation is hard. Pre-generation is just editing.

The one exception is section 02.

02  The in-room screensthe only live decision, and the only one with no undo

What we want them to do

• Default: Ryan's talking head, so the host has a face to talk to.
• On demand: swap to other media. B-roll, a chart, a product shot, a generated clip, whatever the segment calls for.
• Independently: the projector and the CRT should be able to show different things.

What that needs

A dedicated machine driving those displays, with someone choosing what is on them. Not a broadcast switcher. Closer to running a presentation.

03  Media flowshoot day to timeline

The Rodecaster lane is empty for ep1. From ep2 onward it carries the real audio, and camera audio goes back to being sync glue only.

Full step by step in 10-premiere-episode-workflow.md.

04  Pre-shoot checklistrun in order · every item exists because it cost us something

The principle: every knob set to Auto is a decision the camera makes without you, and you pay for it in the edit. Items marked in red are ones ep1 proved the hard way.

• Wipe every lens. Free, 30 seconds, ruins an episode if skipped.
• Check storage on every card. An episode is 100 GB. Ep1 measured 101 GB across five files.
• Check record limits. Some bodies stop at 30 minutes. The show runs 45+.
• Batteries and power on everything, including the laptop driving the screens.
• Format the cards in-camera, not on a computer.

• One key light per face. Minimum. Ep1: Cam 3 had none. No key light, so the lens opened wide, so depth of field collapsed, so autofocus hunted. Permanently soft. Unfixable in post.
• Same colour temperature on every key light. Faces must match across cuts.
• Background lighting is separate from face lighting. LEDs, lamps, and the projection wall are set dressing. Colour them freely. Ep1: the projector wash was doing the job a key light should do.
• Check the projector isn't washing the couch. Bright content on the screens changes the room's exposure.

• Frame rate: 23.98. All three. No exceptions. Ep1 shot 23.98, 59.94, and 29.97. Every mismatched angle conforms, so frames duplicate or drop.
• Resolution: 4K on all three. Ep1: Cam 2 wrote 1080p letterboxed inside a 4K container. The black bars are baked into the file. Scaling scales the bars too. The angle was unusable.
• Picture profile: the same on all three. All log or none. Never one log and two standard. Ep1: the FX30 shot log alone. Jordan hand-matched it every week.
• White balance: manual, same Kelvin number on all three. Never auto. Three cameras on auto find three different compromises, and you cut between different-looking faces.
• Exposure: manual. Never auto.
• Focus: manual, locked on the eyes. Autofocus off. It breathes every time someone leans.
• Audio enabled on every camera. Even a bad onboard mic. It's sync glue, not sound. A camera with no audio can't audio-sync and has to be aligned by hand, every week, forever.
• 48 kHz on every camera.

• Cam 1 — host: face lit, framing right, focus locked on the eyes, confirmed sharp on a real monitor.
• Cam 2 — wide: lit, framed, focused. Check the screens and CRT read correctly in frame.
• Cam 3 — guest: face lit, framing right, focus locked on the eyes, confirmed sharp.
• Check focus on a monitor, not the flip-out screen. Ep1's soft camera looked fine on a 3-inch screen.
• Nobody is cut off at the edge of frame when they lean or gesture.

• Every mic into the Rodecaster, one track each. Host, guest, producer.
• Dynamic mics, six inches from the mouth. Not condensers. A condenser hears the projector fan, the chairs, and the street.
• Everyone in earbuds. Never speakers. If Ryan's voice comes out of a speaker it enters all three camera mics. That's echo, and it cannot be removed afterward. Cheapest fix available, biggest effect.
• Mix-minus confirmed: Ryan hears everyone except himself. Send his own voice back and he hears himself delayed. He'll stumble and stop talking.
• Level check every voice before rolling, not during.
• Multitrack recording armed on the Rodecaster.

• Ryan records locally on his end, not through the meeting. Ep1: his capture was variable frame rate. The file declared 24 but averaged 23.76. It drifts on its own and has to be transcoded before import.
• Ryan is lit, framed, and mic'd before the show starts.
• Ryan does not pause. Ep1: he paused mid-show. Two files, two independent syncs, and the gap is real time only the cameras recorded.
• Ryan is on camera for the clap, and claps with the room. Ep1: he clapped alone and early while the room's hands were occupied. A clap only works if it's shared. His was worthless as a sync point.

• Roll everything a minute early. All three cameras, the Rodecaster, Ryan. Heads are free.
• Everyone claps at the same instant, on camera, all in frame. This is the one universal sync point, and it's free. It took thirty seconds on ep1 and it's the only reason the multicam exists.
• Nobody pauses. For 45 minutes. Storage is cheaper than sync. A pause breaks continuous time and nothing in the file records the gap.
• If anything restarts, everyone claps again. Every restart needs its own sync point.
• Stop late. Let it run a minute past.

• Live clip log. Someone notes the time when something lands or breaks. "12:14, Ryan's pricing answer." Replaces a full review pass, and it's the clip-package input: the moments worth scoring get written down as they happen.
• Someone watches the screens. Whatever is on the projector or CRT is baked into the wide shot permanently. It's the only decision in the room with no undo.

• Offload to one folder per episode.
• Rename every camera file to say which camera it is. Do it now, on the dump. Ep1 files were C4177, C3780, C3777. Nobody could tell which was which. And renaming after Premiere imports breaks the link.
• Keep untouched originals in their own folder, sidecar XML included. Never separate a clip from its sidecar.
• Verify every file plays. Before anyone leaves the room.
• Get Ryan's file. Don't let it wait a week.
• Copy to the external drive. Media never lives only on a laptop.

05  Camera plan

The reasoning: the most frequent cut in the show is host close-up to guest close-up. Those two shots must match exactly or the seam shows. So the two matched bodies go there, and the odd body takes the wide.

That is why FX30 on wide is right. Not because it is the weakest camera, it isn't, but because it is the one without a twin.

Note: the wide is also the shot carrying the projector and CRT. It is the establishing shot and the screens shot at once.

Current state

• FX30. Jordan owns it. Lens currently rented.
• Sony ZV-E10 x2. Rented. Vlogging bodies: no viewfinder, no fan, weak in low light. Fine as a stopgap, not a foundation.
• Tripods. Rented.

06  Rentalsfree, and still a problem

We have a subscription with a local media company. Rentals are free. Cost is not driving this decision.

07  Equipment roadmapbuy in this order

08  Audiothe one place worth spending first

09  Shoot daythe eight-step short version

Before anyone sits down

• Wipe every lens. Free, 30 seconds, ruins an episode if skipped.
• Light every face. One key light per person, minimum. Not the projector, not the LEDs. A light pointed at a face.
• Check storage on every card. Ep1 was 101 GB. Know you have room before you roll, not at minute 30.
• Check record limits. Some cameras stop at 30 minutes. The show runs 45+.
• Batteries or power on everything. Cameras, Rodecaster, the laptop driving the screens.
• Set the screens. Whatever is on the projector and CRT is baked into the wide shot forever. Decide before you roll.

Camera settings, all three identical

• Frame rate: 23.98 on all three. No exceptions. Ep1 shot 23.98, 59.94, and 29.97. Two conformed, frames duplicated.
• Resolution: matched on all three. Ep1's cam2 wrote 1080p letterboxed inside a 4K container. The black bars are baked in. The angle was unusable.
• Picture profile: the same on all three. All log or none. Never one log and two standard. Ep1's FX30 shot log alone. Jordan hand-matched it every week.
• White balance: manual, same Kelvin number on all three. Never auto. Three cameras on auto find three different compromises and you cut between different-looking faces.
• Exposure: manual. Never auto.
• Focus: manual, locked on the eyes. Autofocus off. It hunts in low light and breathes every time someone leans.
• Audio enabled on every camera. Even a bad onboard mic. It is sync glue, not sound. A camera with no audio gets aligned by hand, every week, forever.

Check every angle, individually

• Cam 1: face lit, sharp on the eyes, framed, audio present?
• Cam 2: face lit, sharp on the eyes, framed, audio present?
• Cam 3: face lit, sharp on the eyes, framed, audio present?
• Zoom in on the monitor to confirm focus. A small screen makes everything look sharp. Ep1's soft camera looked fine in the room.
• Check focus on a face, not the background. If the bookshelf is sharp and the eyes are not, the focus plane is wrong.
• Sit someone in each seat while you check. An empty chair does not tell you whether the shot works.

Ryan, or any remote guest

• Recording locally on his end, not through the meeting. Kills variable frame rate, kills compression, probably kills the pause. The single cheapest fix available.
• His camera framed and lit before you start.
• He does not pause. If he must, he claps again on resume.
• He claps with the room, not before it. Ep1: he clapped alone and early while the room's hands were busy. A clap only works if it is shared.

Audio

• Every mic into the Rodecaster, each on its own track.
• Everyone wearing earbuds. Never speakers. Ryan out of a speaker lands in all three camera mics. Unfixable echo.
• Mix-minus confirmed: Ryan hears everyone but himself.
• Levels checked with people actually talking, at the volume they will actually talk.
• Mics close, six inches, dynamic. The room is untreated.

Roll

1. Roll earlyStart all three cameras, the Rodecaster, and Ryan a minute before the top.
2. Everyone claps, at the same instant, on camera, all in frameOne hard clap. This is the entire reason multicam sync works. It took thirty seconds on ep1 and it is the reason there is an edit at all.
3. Confirm every camera is actually recordingRed light, timecode running. Not armed. Recording.
4. Nobody pauses for 45 minutesStorage is cheaper than sync. A pause breaks continuous time and nothing in the file records the gap.
5. If anything restarts, everyone claps againEvery restart needs its own sync point.
6. Stop lateLet it run a minute past.

During

• Live clip log. Someone notes the time when something lands or breaks. "12:14, Ryan's pricing answer." Costs nothing, replaces a full review pass, and it is the input to clip selection later.
• Watch the multiview if it is wired. A camera drifting out of focus is fixable at minute two and permanent at minute forty.

After, before anyone leaves

• Verify every file plays. All three cameras, the Rodecaster, Ryan's file.
• One folder per episode. Camera name in every filename. Ep1 files were C4177, C3780, C3777. Nobody could tell which was which.
• Rename now, on the card dump. Never after Premiere has imported. Renaming after breaks the link and you get media offline.
• Keep untouched originals in an originals/ folder, sidecar XML included. Never separate a clip from its sidecar.
• Copy to the external drive. Not the laptop, not a cloud-synced folder.
• Get Ryan's file before he closes his laptop and forgets.

10  Episode 1, what was shot

All three cameras: stereo, 48 kHz, audio present and audible.

Episode weight: 101 GB. That is the number that sizes storage.

Three problems, three causes

• Cam 3 soft. No key light, so the lens opened wide, so there was almost no depth of field, so autofocus hunted. A lighting failure wearing a focus costume.
• Cam 2 washed. FX30 shooting log while the others baked their look in. Not a color problem, a profile mismatch. Jordan hand-matches it every week.
• Ryan hard to sync. His recording was paused mid-show, and part 1 was variable frame rate. Two separate problems that compound.

11  Ryansettled, he's a fourth angle

His recording is a clean, framed, mic'd shot of his face. When he talks, we cut to it full screen in Premiere. Separately, his live feed is what the projector and CRT show during the shoot, which is section 02.

Rules for his footage

1. Part 1 and Part 2 are two independent sourcesSync each against the cameras separately. Never join them, never nest them into one clip. The gap is not ours to reconstruct. The cameras hold the truth of what happened during it.
2. Transcode to constant frame rate before import23.76 is not a frame rate, it is the average of a variable rate capture. It drifts on its own, pause or no pause.
3. Clap on every resumePart 2 needs its own sync point, not a guessed offset from Part 1.

Known and accepted: his image is webcam-soft against cinema cameras, and his color is warm where the room is blue. Every remote guest show has this. It is a grade, not a setting.

12  Open