Byron, this is the best round of the seven and it still contains one idea that will embarrass you and one spec that hasn't done its arithmetic. In order.
---## 1. What actually ships for the $1M/yr goal
Ask who writes the check. It's not a listener. It's a CTO, a GC, a conference chair, a client's board — someone who arrives once via a link, spends 90 seconds, and needs to leave thinking "this person has operationalized AI governance and I have not." That buyer does not join a 24/7 stream. That buyer does not tune a squelch unprompted. That buyer looks at one screen and either sees an artifact of authority or a hobbyist's art project.
The Faceplate wins, and it isn't close. It's the only one of the three where the governance metadata — per-claim verification bars, attestation tiers, the citation counts in mono — is the content of the first screen. The Carrier hides your governance behind a clock; the Noise Floor hides it behind a gesture the visitor must discover. The Faceplate prints it on the chassis.
But be honest about the dependency: the Faceplate is a UI that writes checks your pipeline must cash every single day. Three-segment claim bars, judge-fire ticks in the scrub rule, live citation tables — if the citation-validation run on Ep 1 comes back flaky, this design broadcasts your pipeline's failures in tabular figures on your own homepage. That's either the whole point or a self-inflicted wound depending on whether the pipeline is actually ready. The Ep 1 test results are the real gate for this round, not the design.
Both, and which one depends entirely on the default position and the framing of the tiers — neither of which the synthesis resolves, which tells me nobody on the panel gamed the hostile reading.
The hostile reading: ATTESTED · 6 OF 41 AUDIBLE. A skeptic reads that as "he stands behind 15% of what he publishes." Your AUTO stations exist precisely so throughput doesn't require your signature — but the squelch collapses that nuance into one axis where AUTO content is definitionally "below floor." You built a two-class content system on purpose and then built a control that makes the lower class look like sediment.
The profound version exists, but it's a performed moment, not a resting state. Sliding to ATTESTED in front of a room and watching your catalog go dark is a genuinely great five seconds of theater — the best RAI demo in the whole proposal. As a default homepage state, it's a store with empty shelves. Resolution: default the floor per-station to that station's declared tier (The Signal rests at ATTESTED and loses nothing; Field Notes rests at JUDGED and is honest about it), and ship the full-catalog slide as an explicit, user-initiated "prove it" gesture. The SCAN mode — playing each survivor's highest-uncertainty claim — is quietly the smartest idea in Round 7. It's adversarial self-audit as a listening mode. Protect it.
Three separate failures:
Listener experience: Radio dead air was powerful because a live audience already existed to experience the absence. On the web, silence is indistinguishable from a bug. Your one witness at 3pm on a Tuesday refreshes the tab, assumes your stream broke, and leaves. A statement nobody witnesses isn't a statement; it's a diary entry with CDN costs. And with three stations, the "broadcast day" is mostly reruns — the clock inverts your density problem rhetorically but not experientially.
Legal/tone: "Which claim failed and why he wouldn't sign it," read aloud over silence. If the claim failed because a cited third party's material was wrong, you are now publishing dated, audible assertions that named sources published falsehoods — in a cloned voice, unskippably. Your lawyer will make you sand every notice down to mush, and mush read over dead air is worse than nothing. If instead the failures are all your own pipeline's hallucinations, you're scheduling a recurring public segment titled "my system generated garbage again." Neither branch lands where you think.
Infra: The entire existing stack is static files and RSS — that's why it's cheap and durable. A live 24/7 carrier means HLS/live-edge infrastructure, a scheduler, and a maintenance liability, all to serve the platform's least-valuable visitor pattern.
The salvage: the Air Strip is a good visualization and needs zero live infrastructure. Render the day-clock statically on the Faceplate, mark killed slots as visible gaps, and make each kill a produced 40-second artifact — the one-line notice as text plus a signed audio stinger — filed in The Cutting Room and in the RSS feed where subscribers will actually encounter it. You keep 95% of the statement and delete 95% of the risk.
Do the math on a 390×844 viewport. Subtract browser chrome (~110px), the 112px cluster, and a six-key preset row (~64px minimum at 44px touch targets plus labels): you have roughly 550px for the entire center stack — station strips with detent paging, position readouts, and three-segment per-claim micro-bars in each cell. Those micro-bars at cell scale are sub-pixel decoration, not legible state. Something gets cut; decide now, not in build.
Worse, the interaction verbs fight the platform: long-press-to-bind collides with iOS text-selection and context menus; hold-500ms for the lid collides with scroll intent; detent paging fights every scroll habit a visitor brings. And you've declared WCAG-AA a launch gate while specifying two load-bearing interactions with no obvious keyboard or screen-reader equivalent in the spec. That's not a detail — under your own rules, the lid and preset-binding as specced cannot ship. The appliance metaphor is a real differentiator, but 1968 solved a driver who couldn't look away from the road; your user is holding the device with full attention. Keep the cluster and tell-tales; be suspicious of every gesture the metaphor smuggles in.
Homepage: the Faceplate. Cluster + tell-tales + station strips, with the Air Strip rendered statically as a row (killed slots marked, kill-notices as produced artifacts). Noise Floor ships as the Faceplate's first control, defaulted per-station to declared tier, with the catalog-wide slide as an explicit demo gesture. The Carrier as a live stream is cut — the Air Strip and kill-artifacts absorb its only defensible ideas. The v6 streaming Lineup is dead as a homepage (the palette correction already killed it) but its rails survive as the DOM-first crawlable layer underneath. STANDBY and Second Needle stay P2. /listen pages ship P1 — they're boring and they're the actual distribution surface.
Let the buyer point the instrument at their own material. The entire apparatus — judge, citation validation, attestation tiers, Noise Floor — currently only audits your content. Add one gated demo: paste a URL or transcript, and the pipeline renders a claim table and a floor reading for someone else's podcast or press release. That's the moment the Faceplate stops being your trophy case and becomes a tool the $1M buyer imagines pointed at their own organization. Authority isn't demonstrated by governing yourself; it's demonstrated when others want your governance applied to them.