Case 24 · Graduation Day

DealProof

The proof stack is unusually real — benchmark numbers, 256 passing tests, a live enclave deployment, a paper author's endorsement — but the pitch proves the cryptography and never the company: no user, no buyer, no market, no ask, and one word in every eight is a filler.

Christian2:48 pitchreviewed 2026-08-17

01 Scorecard · 12 dimensions

Problem

6"Salary negotiations run on unverifiable claims" + faked competing-offer screenshots + NDA-locked experience is concrete — but stated as hypotheticals ("candidates might fake"), with no evidence of frequency or cost.

Why now

2Never addressed. TDX enclaves being newly practical is the obvious why-now sitting unspoken inside the pitch.

Insight / secret

6"Not just private, but also provable" is a genuine reframe — most privacy pitches stop at private; this one names verifiability as the real product.

Solution clarity

6Two agents, one per party, inside a TDX enclave, "sealed input, attested outputs," human approval gateway, signed execution — the mechanism is complete in ~45 seconds despite the halting delivery.

Product / demo

4"It is live, deployed" and 256 tests — but the room never sees or hears what a user experiences. Not one screen, flow, or negotiation described.

Moat

2Nothing on defensibility. "Two closed agents… it might be open source in the future" actively muddies whether the closed agents are the asset or a temporary state.

Market

2Zero sizing. "Anywhere two parties negotiate with sensitive data, that's the pattern" is a pattern, not a market.

Business model

1Absent. Candidate pays? Employer pays? Per negotiation? Never touched.

Competition

2Never mentioned. The status quo (screenshots, reference calls) is implied in the problem but no alternative is named and displaced.

Traction

370% acceptance and 48.4% over a naive baseline are real evidence — but they are benchmark results, not buyers. No real negotiation has run.

Team

2"I'm Christian, founder and developer" — solo, no background given. The Andrew Miller line does more for his credibility than anything he says about himself.

The ask

1None. The pitch ends on the vision sentence and "Thank you." The room cannot act.

02 Klaff 5-lens audit · 2/5

Croc brainFAILThe open is name + title. The pattern interrupt the pitch owns — candidates faking competing offers with screenshots — is buried in sentence three.

FramePARTIAL"Not just private, but also provable" claims a category, but the whole pitch is spoken in the hedged analyst frame ("might fake," "might be open source").

STRONG arcPARTIALProblem → mechanism → validation → future → vision. A real proof beat — rare in this cohort — but no stakes and no ask beat.

Hot cognitionPARTIALFaked-offer screenshots and NDA exposure are hot — but delivered as hypotheticals, not as one story that happened to one person.

StatusPARTIALAndrew Miller, an author of the paper, calling it "the most complete work implementation" is a legitimate status borrow — spent mid-list, surrounded by 39 fillers, with the paper never named.

THE ONE REORDER — Open on the faked screenshot ("Right now, someone is negotiating a salary with a Photoshopped competing offer"), close on the Miller endorsement plus an ask.

03 Delivery

Length

2:48

Pace

113 wpm

Fillers / 100 words

12.3

39 fillers ("uh" ×22, "um" ×17) — one word in eight. Strip them and the content rate is ~99 wpm. The material underneath is dense and true; the fillers make it sound uncertain.

04 Beat-by-beat

Time	Beat	What works	The fix
0:00	Intro	Name + role stated plainly	Don't open here. Open on the fake screenshot; introduce yourself at the close, next to the Miller quote.
0:10	Problem	Three-sided pain: faked offers, NDA-locked experience, employers "over-trusting or stonewalling"	Make it one story in past tense, not three hypotheticals in "might."
0:40	Solution claim	"Cryptographically provable, not just private, but also provable" — the best line in the pitch	Say it once, clean, without the false starts.
0:55	Mechanism	Two agents, one per party, "entirely inside a TDX hardware enclave. Sealed input, attested outputs."	Cut "might be open source in the future" — it raises a licensing question mid-mechanism and answers nothing.
1:20	Human gateway	"Accept, decline, or ask for more rounds" pre-empts the agents-run-wild objection	Keep. This is the trust beat; give it one unhurried sentence.
1:40	Validation	70% acceptance, 48.4% over naive baseline, 256 tests, live deployment — the densest proof paragraph of the day	Name the benchmark. "Real academic benchmarks" is unfalsifiable as spoken; without the paper, the numbers float.
2:10	Endorsement	"The most complete work implementation of this paper" — Andrew Miller, an author — genuine borrowed status	Name the paper. An endorsement of a paper the room never heard named is a credibility loop that doesn't close.
2:20	Future plans	Provenance beyond GitHub: NDAs, recommendation letters, email exchanges	Roadmap creep before the first market exists. One sentence or drop.
2:40	Close	"Anywhere two parties negotiate with sensitive data, that's the pattern where Dealproof fits" — clean vision sentence	A pattern is not a close. Follow it with the ask, then "Thank you."

05 The gaps investors block on

No ask. The room hears benchmarks and a vision and is given nothing to do.
Zero market traction — every number is a lab number. One real hire negotiated through DealProof is worth more than 256 tests.
The unnamed paper: the pitch's biggest credibility line (Andrew Miller's endorsement) hangs on a paper the room never hears named.
No buyer identified — candidate-pays and employer-pays are different companies; the pitch commits to neither.
Two-sided cold start unaddressed: nothing on how the first employer is convinced to negotiate inside someone else's enclave.
"Closed agents… might be open source in the future" — for a verifiability product, what exactly can I verify if I can't read the negotiator?

The one strategic decision

One negotiation, one buyer. The close gives it away — "anywhere two parties negotiate with sensitive data, that's the pattern" — and the future-plans beat is already drifting to NDAs and recommendation letters before a single real salary negotiation has run. Pick the salary wedge (the pitch is already "GitHub developer-centric" by its own description), sold to the employer side — the party with budget and the acute verification pain. Everything else waits.

“DealProof is where engineering offers get negotiated on proof — candidates prove their experience and competing offers without exposing a single private repo or NDA, employers verify everything without seeing any of it, and the whole negotiation comes back cryptographically signed.”

06 The rewritten pitch

Somewhere right now, a candidate is negotiating a salary with a Photoshopped screenshot of a competing offer — and the employer across the table either swallows it or stonewalls, because neither side can prove anything. Engineers have the opposite problem: their best proof of experience lives in private repos and behind NDAs they can't open.

DealProof makes the negotiation provable — not just private, provable. Two agents, one for each party, negotiate entirely inside a TDX hardware enclave: sealed inputs, attested outputs, neither side ever sees the other's raw data. Nothing binds without a human — each party accepts, declines, or asks for another round — and the whole execution comes back cryptographically signed, verifiable by anyone.

This isn't a slideware protocol. Against [name the benchmark/paper], agentic negotiation reached a 70% acceptance rate — 48.4% better than a naive baseline — with 256 automated tests passing, live on [confirm: CVM platform] today. Andrew Miller — an author of that paper — called it [confirm exact quote] the most complete implementation of the work.

I'm Christian, founder and developer. The next milestone is the first [N] real negotiations: [hiring teams/companies] piloting verified offers. I'm asking for [the ask: $X / intros to N hiring teams] to get there. If you've ever been lied to in a negotiation — come talk to me.

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 buyer: employer-side salary negotiation for engineers. Every other surface follows from this.
Write the ask — amount or concrete non-monetary ask (N pilot hiring teams), use of funds, one milestone: first real negotiation completed.
Name the paper and get the Andrew Miller quote in writing — it's the pitch's best asset and currently unverifiable as spoken.
Run one real negotiation — even a friendly one — so the next pitch has a story in past tense instead of three "mights."
Stage the verification demo: 40 seconds, one finished negotiation's signature checked live from a stranger's machine.
Filler drill: 12.3 per 100 words ("uh" ×22, "um" ×17). Record, count, repeat until under 3/100.