Case 27 · Graduation Day

Daedalus Anchor

A genuinely earned insight — prediction markets aren't a risk product, so build the underwriting layer — told as a lab notebook instead of a pitch: the best customer number comes out garbled, the real wedge arrives at minute three, and the only ask in five minutes is "please follow us on Twitter."

Ria5:13 pitchreviewed 2026-08-17

4.5

01 Scorecard · 12 dimensions

Problem

6Markets hedge "price volatility rather than events" is a real, big gap — but it spends two minutes as an abstraction before the GPU-cost story makes it concrete.

Why now

4The parts exist (prediction markets, AI companies eyeing them, a 49% GPU-cost surge) but are never assembled into a why-now argument.

Insight / secret

8"We stopped asking what prediction markets can be used for. We started asking what event risks already exist but nobody can hedge against" — earned through three published experiments, not claimed.

Solution clarity

4The product shape-shifts: agent hedging layer → OTC desk → underwriting solution → RFQ system → reality graph → general forecasting model. What can a customer buy today is unclear as spoken.

Product / demo

5Real artifacts exist — agent layer, interactive simulator, historical backtests, two published demos — but the pitch describes the process more than it shows the product.

Moat

2Never argued. Nothing on why an exchange, insurer, or another agent shop can't build the same underwriting layer.

Market

4"A lot of risk in reality that no one has already priced" is a gesture, not a market. Compute is named as the wedge — with zero sizing.

Business model

3An OTC desk + RFQ system implies a fee or spread, but not one word on who pays or how much.

Competition

3"There's a couple of reasons behind that" — and the reasons are never given. No incumbent named, no alternative framed.

Traction

4One real design partner (an AI founder in confidential computing) and a built underwriting solution — but unnamed, and the headline result is unintelligible as spoken.

Team

2"My name is Ria. I'm from Delos Research" is the entire team story. For a company underwriting risk, zero credentials spoken.

The ask

1"Please follow us on Twitter for future updates" is the only ask in the pitch. The room cannot act.

02 Klaff 5-lens audit · 1.5/5

Croc brainFAILOpens with an apology ("I guess I'm too tall… Sorry") and an abstraction ("the general forecasting model for underwriting reality"). Nothing concrete for 40 seconds.

FramePARTIALThe scientist frame — three hypotheses, honest answers — is intellectually high-status and rare. The delivery keeps ceding it back.

STRONG arcPARTIALA real arc exists (idea → experiments → insight → customer → roadmap) but the hook is buried at minute three and there is no ask beat.

Hot cognitionPARTIALThe 49% GPU-cost surge and the tariff-shock replay are felt — but both arrive wrapped in hedge-words and garble.

StatusFAILAn apology in the first breath, 48 "uh"s, and a close that asks for a Twitter follow. The pitch never claims the room.

THE ONE REORDER — Open on the AI founder's 49% GPU-cost surge, deliver the insight as the punchline of the experiments, end on an ask.

03 Delivery

Length

5:13

Pace

152 wpm

Fillers / 100 words

11.2

89 fillers ("uh" ×48, "um" ×18, "actually" ×14). More than one word in ten is filler, densest exactly on the number-bearing sentences.

04 Beat-by-beat

Time	Beat	What works	The fix
0:00	Intro	"The general forecasting model for underwriting reality" is a memorable category claim	Don't open with an abstraction followed by "I guess I'm too tall… Sorry." Cut the apology entirely.
0:35	V1 product	The agent orchestration layer (portfolio → risk → event contracts → execute) is concrete	This product gets a minute and is then abandoned by the pitch's own findings. One sentence.
1:30	Tariff-shock backtest	The best evidence in the pitch: a 3% hedging budget covers two-thirds of the loss, ceiling ~89%	Lead with this number instead of burying it mid-list. Rehearse it until it comes out clean.
2:00	Liquidity wall	"Anything above a hundred k, the liquidity just runs out" — honest, and it justifies the OTC desk	Keep, one sentence.
2:20	The insight	"What event risks already exist but nobody can hedge against?" — the pitch's thesis	Move it to the first 30 seconds.
2:50	The customer	The AI founder whose GPU rental costs surged ~49% with no futures market — the one human in the pitch	This is the hook. Open the pitch here. Name him or bracket him.
3:25	The build	A complete underwriting solution: interactive simulator, historical backtests, custom contract terms	The result line is unintelligible as spoken. The proof point must be one rehearsed, unambiguous sentence.
4:05	Roadmap	Compute market + RFQ → reality graph → general forecasting model — a real staged plan	Attach a date or milestone to stage one.
4:30	Vision + close	"Price any risk anywhere" is a big, legible ambition	"BlackRock has been the pricing engine for options" will make a finance listener stumble, and "follow us on Twitter" is the anti-ask. End on money and milestones.

05 The gaps investors block on

No ask. Five minutes, and the only call to action is a Twitter follow.
The headline customer metric is garbled as spoken — the proof the whole pivot rests on cannot be repeated by anyone who heard it.
Who takes the other side? An OTC desk and custom contracts mean someone underwrites with capital — whose balance sheet, what regulatory posture, never touched.
No business model. RFQ/OTC implies a spread or fee; nobody said who pays what.
No moat, no competition. "There's a couple of reasons behind that" — the reasons are never given.
Team is one first name; zero spoken credentials for a credibility business.

The one strategic decision

Be the compute-cost hedging desk first. The pitch tries to be the general forecasting model (platform) and a hedging desk at once. They have exactly one validated finding chain — agents map exposure to events, hedging covers two-thirds of a shock at a 3% budget, public venues die above $100k — which is precisely the argument for a bespoke desk serving the underserved buyer they already found: AI companies with unhedgeable GPU costs. The platform is where the desk compounds to, not where the pitch starts.

“Daedalus Anchor is the desk where AI companies hedge their GPU bills — custom event contracts, priced by our own forecasting models and proven in backtests, for a risk no futures market will touch.”

06 The rewritten pitch

An AI founder we work with runs confidential computing workloads. Last year his GPU rental costs surged forty-nine percent — and there is no futures market where he can hedge that. That's not unusual; that's the norm. Financial markets let you hedge price volatility. They don't let you hedge events.

I'm Ria, from [company name as it should be spoken]. We spent [N months] testing whether prediction markets could close that gap, and we published every experiment. Three findings. One: agents can map a real portfolio to tradable event contracts — if you treat it as retrieval grounded in live market catalogs. Two: event hedging works — replaying the April 2025 tariff shock, a three percent hedging budget covered two-thirds of the loss, and the ceiling is around eighty-nine percent. Three: public markets can't do this at scale — above a hundred thousand dollars, the liquidity just runs out.

So we stopped asking what prediction markets are for, and started asking what risks already exist that nobody can hedge. Prediction markets aren't a risk product — nobody there is pricing or underwriting anything. We are. For that founder we built a complete underwriting solution — an interactive simulator, historical backtests, and a custom contract — that covers [confirm: the exact coverage result, one clean sentence] of his compute-cost risk.

Next: the compute market and an RFQ desk, then the reality graph — modeling how events propagate through the real economy — toward a general forecasting model that can price any risk anywhere. We're raising [$X] to sign [N] more compute-cost customers and stand up the desk. The demos are published as artifacts — go try to break the backtests.

Brackets are the founder's homework — the copy is done, the proof is theirs to supply. Every claim must match what's actually built.

07 Do these before the next take

Pick the wedge out loud — open with the compute-cost hedging desk; say "general forecasting model" last, not first.
Write the ask — amount, use of funds, milestones. One paragraph of work; it converts a talk into a pitch.
Fix the proof sentence — the customer coverage metric must be one unambiguous, rehearsed line.
Answer the counterparty question — who underwrites, whose capital, what regulatory posture.
Add one team sentence — why Ria and this team get to price risk that markets don't.
Filler drill — 11.2/100 words ("uh" ×48), clustered on the number-bearing sentences.
Cut to 3:00 — the abandoned V1 story and hypothesis mechanics give back two minutes.