Case 28 · Graduation Day
WikiGen
The paradox at the center — the most valuable biological data is the data least likely to move — is a fundable problem, and the custody-plus-TEE mechanism is complete on paper; but the pitch buries both under tangled syntax, a seven-item feature dump, zero traction, zero market math, and a closing ask that is a slogan, not a number.
01 Scorecard · 12 dimensions
02 Klaff 5-lens audit · 1.5/5
THE ONE REORDER — Open on the paradox, put the Q&A line about the person with no test and no cure right after it, and end on a numbered ask instead of "Help us make verifiable rewards a real reality."
03 Delivery
11 fillers ("um" ×4, "like" ×4) — genuinely low. The delivery problem is syntax, not filler: false starts like "home genome-- whole genome" and "I'm building it" do the damage fillers usually do.
04 Beat-by-beat
| Time | Beat | What works | The fix |
|---|---|---|---|
| 0:00 | Name open | "No one speaks that alphabet… so I go by Wiki" — warm, human; the company name gets a face | One sentence, then straight to the paradox — the next 45 seconds spend the goodwill |
| 0:20 | Biotech evals tangle | The instinct to establish scientific credibility is right | "Evals for your own evals… a laser is damp" is not parseable live. One concrete lab sentence: what you measured, what went wrong, what it cost |
| 1:15 | The paradox | "The most valuable biological data in the world is also the data least likely to move" — the best line in the pitch | Open with it |
| 1:50 | Not a bug | "The data is private for a good reason… It belongs to someone's life" — the emotional core of a custody product | Keep verbatim |
| 2:30 | Mechanism | Funded capped evaluation → custody kept → TEE → "only the output that was agreed in advance" — complete | Three short sentences instead of one nested one; cut "the verifier can be swappable" — it un-sells the core |
| 3:20 | Feature dump | "Attested diligence rooms… governed data vaults… policy safeguards" — evidence a lot exists | Seven features in one breath means the room retains zero. Pick one — the signed evaluation receipt — and walk it through |
| 4:15 | Arena-session line | — | "The view is filtered only by the address recovered from your signed arena session" is documentation, not pitch. Cut, or show the screen it describes |
| 4:40 | Close | — | "Help us make verifiable rewards a real reality" → replace with the ask: amount, first paid evaluation, first data owner |
| 4:55 | Q&A: data value | "As valuable as someone is affected by a condition that there is no test… or no cure" — the hottest cognition of the night | Promote into the pitch; source or drop the 0.03–0.10-cent EHR figure [confirm source] |
05 The gaps investors block on
- No ask — "Help us make verifiable rewards a real reality" gives the room nothing to act on.
- Zero traction — no biobank, hospital, sponsor, or pilot named; a two-sided marketplace pitch with neither side present is a thesis, not a company.
- The chicken-and-egg is unaddressed — which side comes first? Every marketplace investor will ask.
- No business model — "funds a capped evaluation" gestures at a take rate that is never stated.
- No moat, and "the verifier can be swappable" actively argues against one.
- The only number in the session — the 0.03–0.10-cent EHR figure from Q&A — is unsourced and confusing as spoken. [confirm source]
Go demand-first: one sponsor, one question type. The pitch names five sponsor types and four data-owner types and commits to none. Land one biopharma or AI-science sponsor with a single funded, capped evaluation — a funded question is the only thing that gives a data owner a concrete reason to approve a use. The pitch's own mechanism starts with the sponsor; the company should too.
“WikiGen lets a biopharma company test its model against patient data a hospital will never share — the data never leaves the hospital's custody, the computation runs in an attested enclave, and the only thing that comes out is the score both sides agreed to in advance.”
06 The rewritten pitch
The most valuable biological data in the world is also the data least likely to move. Patient records, biobanks, proprietary sequences — private for good reason: it's regulated, it's sensitive, it belongs to someone's life. But somewhere there's a person with a condition that has no test and no cure, and the data that could change that is sitting in a vault nobody is allowed to open. When data can't move, science can't either.
I go by Wiki — I spent [N years / role] in biotech [confirm: lab/company], where I ran into that wall myself. So I'm building WikiGen: bring the model to the data, never the data to the model. A sponsor — a biopharma company, an AI scientist — defines a question and funds a capped evaluation. The hospital or biobank keeps full custody and approves that one use. The computation runs inside an attested confidential environment, and the only thing that ever leaves is the output both sides agreed to in advance: a score, a match, a signed evaluation receipt. Nobody ever sees the raw data. Not the sponsor. Not us.
This isn't another centralized data lake — it's the layer that lets institutions do science on data they will never share.
We're starting demand-first: [sponsor name / type] is [confirm: status] our first funded evaluation against [data owner]'s data. We're raising [$X] to run the first [N] paid evaluations and sign the first [M] governed vaults. If you know a biobank, a hospital data office, or a pharma eval team — that intro is the ask. Thank you.
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: demand-first, one sponsor type, one question type. Everything else reorganizes around this.
- Write the ask — amount, number of paid evaluations it funds, first-vault milestone, and the intro you want from this room.
- Get one real name on each side of the marketplace — until then, every marketplace claim is bracketed.
- Promote the Q&A answer ("no test and no cure") into the pitch; it is the emotional engine and it was spent off-stage.
- Cut the arena-session line and the seven-item feature list; replace with a 60-second signed-receipt walkthrough.
- Source or drop the EHR pricing figure — the pitch's only number and its least defensible sentence.
- Rehearse for syntax, not filler — filler is already excellent (1.3/100); the false starts are the tell of unmemorized transitions.