Investor one-pager/ pre-seed → seed/ the model proposes · the code disposes
The code-enforced-correctness civic-video pipeline for local newsrooms.
Built for, and running live in, a real hyperlocal Brooklyn newsroom — Bushwick Daily. verified · reference customer
spoken word · grounded
A wrapper passes a prompt to a model and ships the model's output. vidtranscript inverts that exactly where it matters — the model proposes editorial judgment; deterministic code decides what is true, what is fair, and what may ship. Three gates carry the load.
Moat 1 · correctness
Exact normalized-token subsequence match against the Deepgram word array; quotes stored as word indices, display text regenerated from the array; unmatched quotes land in rejected_quotes[] with a reason. The renderer never sees the model's string.
6-mo copycat? · No — needs the indices-not-strings data model + reject pipeline
Moat 2 · fairness / liability
Airtime recomputed on the final trimmed cut, moderators excluded; render returns HTTP 409 until a justification is recorded. Server-enforced — it holds even if the front end is bypassed. The debate profile is forced to full diarization.
6-mo copycat? · No — a render-blocking instrumented gate encoding a liability model
Moat 3 · operational capture
Global on-disk circuit breaker (6h cooldown), failure-classified client rotation, 45s pacing, transcript-first tiering, cookie rotation — earned the expensive way. Lead with YouTube (proven); HLS partial.
6-mo copycat? · Partial — relearnable only by paying the same IP-ban / over-bill tuition
Why this is the page's spine
Every section below answers the moat from a different angle: economics proves it is also cheap to run ($0 render); competition proves an incumbent would have to invert its product to copy it; traction proves it runs live; the ask proves the gaps are all around the moat, never in it.The board meetings, debates, and press conferences still happen — now with no one in the room. The raw material is abundant and public; the labor to turn it into coverage has collapsed. This is accountability loss, not nostalgia.
On-page copy — verified, investor-only
The collapse, in three verified figures
All three are VERIFIED macro figures (Medill State of Local News 2025), cited here as the structural wound — not as adoption.
The audience that left newspapers now gets civic news from short video and "news influencers." The audience is on the platform; the trustworthy supply is not. And the enabling tech only crossed the grounded-extraction threshold roughly 18 months ago.
On-page copy
Local-news attention · 2016 → 2025Pew · verified
Red declining line = local-news attention 37%→21% (Pew, VERIFIED). Green dashed marker = the ~18-month-ago point where ASR + LLM crossed usable-for-grounding. The format demand and the enabling tech arrived at the same moment.
One operator — no engineer, no command line — turns a pasted civic-video URL or uploaded footage into three reviewable outputs. Minutes, not hours — without the fabrication risk that keeps newsrooms off AI.
The four-move operator flow
The three outputs, named
Four stages, three deterministic gates drawn as distinct non-LLM nodes. The model proposes editorial judgment; deterministic code decides what is true, what is fair, and what may ship. This diagram is the anti-wrapper argument — the gate-as-node motif that recurs across the page.
The pipeline — the gate is a non-LLM code node
The three gates — mechanics verbatim with the customer deck
Crosstalk caveat — diligence-grade candor
Airtime is recomputed on the final cut with moderators excluded; one honest limit — where speakers talk over each other, overlapping words are single-attributed, so airtime can undercount crosstalk.
The honest boundary — never claim perfect ASR
Grounding proves quote-matches-transcript, not transcript-matches-audio. The human ear at Gate 2 (play-through before Approve) is the mis-transcription safeguard. We do not claim perfect ASR; the human-in-the-loop is the design, and it is a selling point.The defensibility is the deterministic code around the model — plus operational know-how a prompt can't supply. The model is never trusted with the load-bearing claim.
The moat matrix — verbatim with the pitch-outline
| Moat | Wrapper failure mode feared | 6-mo copycat? |
|---|---|---|
| 1 · Correctnessquote grounding | An LLM fabricates or paraphrases a quote; the newsroom publishes words nobody said; trust is gone. | No |
| 2 · Fairnessrender-blocking | An AI edit skews a candidate's airtime in election season → legal + reputational exposure. | No |
| 3 · Captureoperational | A naive scraper gets IP-banned or over-bills within days of daily pulls from an anti-bot platform. | Partial |
"None of these three fall out of an afternoon of prompting. They are deterministic code encoding editorial and operational knowledge — and the secondary asset is a compounding corpus of human-approved, grounded civic clips a copycat can't assemble."
The moat, shown working — investor-only states
Honesty box — stating the edges strengthens credibility
Grounding proves text-matches-transcript, not transcript-matches-audio. Fairness can undercount crosstalk. The capture moat is a lead, not a patent. HLS is partial; YouTube is proven.The cost killer of AI-video startups — paid generation / render APIs — is absent by design. Render is ffmpeg-only. Only ASR + LLM are metered, and both are held to cents per asset.
On-page copy
cost_summary()).What we do NOT claim
No invented MRR / ARR. No Deepgram $/min ($0.0052) stated as billed truth (appendix only, pending a billed-test confirmation). No "fully autonomous" claim — the human-in-the-loop is a selling point.Per-asset cost · one shared horizontal scalebars 1–2 illustrative
Bars 1–2 = ILLUSTRATIVE (in-code estimate, not invoiced). Bar 3 = VERIFIED ($0.00, an architectural seal). All three share one horizontal scale, so the hybrid saving and the free render read instantly.
Land where pain is highest and incumbency lowest. Expand coverage by configuration — a new civic format is a JSON file. Then carry the same correctness moat into better-funded "provably-spoken quote" buyers.
The 136-closures/year cohort — highest pain, lowest incumbency. Live reference customer: Bushwick Daily.
verified · 4 profiles community_board, press_conference, fireside_chat, debate ship today.
Depositions, hearings, press shops — all buyers of a provably-spoken quote. A sequenced path, not a claim of being there.
The retention hook
Each footage archive becomes a searchable, attributed quote database — every clip the operator approves compounds into a corpus a copycat can't assemble. Switching cost grows with use.
The full core path is implemented and runs live. The edges are named out loud — and they are exactly what the round funds. We do not present artifact counts as growth.
On-page copy
Honest-edges matrix — authored once, identical in both decks, all five edges
| roadmap | No fairness-ack UI form yet — server-enforced via HTTP 409; the typed-justification UI is roadmap. |
| emerging | Auto-recap / debate-reel is implemented but CLI-only, not over HTTP — never demoed as a web feature. |
| scoping | Single-operator — no auth, localhost-CORS-locked, in-memory non-durable jobs. |
| ops | No scheduled capture-health probe — the check is implemented and cheap, but no scheduler runs it; "daily" is documented intent, not an automated job. |
| partial | HLS capture is partial — lead with YouTube civic capture, which is proven. |
Read this as reality, not traction
DB row counts are real use at single-operator scale — never dressed as user or usage growth. Bushwick Daily is a reference customer (proof), never an adoption metric.SOM is built bottom-up from newsroom seats — not a top-down "1% of a $40B market." The AI-video tooling curve is the tailwind the wedge rides, cited as momentum, not as TAM math.
On-page copy
TAM / SAM / SOM — subordinating TAM to a bottom-up SOM
Each tier narrows. TAM = ILLUSTRATIVE. The SOM bar carries the vermilion left-edge as the live wedge — computed bottom-up from seats, deliberately subordinating TAM (the opposite of the usual top-down emphasis).
Competitors optimize for virality. We are the only one that makes the quote provably real and the edit provably fair — the two things a newsroom is legally and reputationally unable to compromise.
On-page copy
X = correctness / fairness enforcement · Y = civic-video fit
The empty top-right quadrant IS the argument. Competitors are paper-toned; vidtranscript is the elevated civic-green dot, visually isolated. The real substitute is labor, not a feature-peer.
The moats exist because the builder lived the failure modes. A correctness-obsessed pipeline shipped solo, embedded with the actual buyer — earned editorial-domain insight, not assumed it.
Use-of-funds maps 1:1 to the audit's honest gap list. None of these gaps touch the safety core; they are the periphery around it. This is the deck's only call to action.
Use-of-funds — maps 1:1 to the audit gap list
| Audit gap — verified, honest edge | bucket |
|---|---|
| No auth / multi-operator; localhost-CORS-locked; in-memory non-durable jobs | productization |
| Fairness-ack server-enforced (409) but no UI form | UX gap |
| Auto-recap / debate-reel CLI-only, not over HTTP | feature reach |
| No scheduler for the capture-health probe ("daily" is intent) | ops |
| Cost rates are in-code estimates, not billed-confirmed | validation |
| Single operator; no GTM motion yet | commercial |
One CTA — the round & ask
The round funds the productization layer — multi-tenant, auth, and the UI for the gates we already enforce server-side — plus the first design-partner newsrooms.
Raising: $[X] pre-seed for [N] months · Milestone bought: [N] paying design-partner newsrooms + multi-tenant hosted GA · Numbers are the founder's to set; the structure is fixed here.
Lead this round → Request a demo →Alec Meeker, founder · alec.meeker@gmail.com
Paste a civic URL. Get a trustworthy clip. The quote is provably real — and the code, not the model, guarantees it.
the verbatim tagline clause · closes the page