Voundry is now live and proving itself in public, but to a serious investor it still has one gap to close: no graded track record yet. The first verdicts land 13 Sep 2026; calibration reads insufficient_data until then. Since this memo was written, Tier A has largely shipped: the sensing layer now covers 7 of 8 evidence types, the out-of-sample backtest engine and the verifiable seal ledger are live, and the first-verdict clock is public. The near-term game is unchanged: manufacture credible, honest proof of skill before you raise.
"Is there any measured evidence this works, or just a polished dashboard of pending calls? Show me a number I can check, not a promise about September."
Produce a real, honest skill score today from resolved history + short-horizon live calls, made independently verifiable. Turn "trust us" into "audit us."
The promise→proof converters. None of these require waiting until November. This is where the next investor conversation is won or lost.
Replay Voundry's prediction method against hundreds of already-resolved historical events, awarded tenders, closed funding rounds, shipped launches, enacted regulations, with strict as-of data cutoffs, no leakage, and a pre-registered question set.
It yields a measured Brier score, skill-vs-baseline, and a real reliability curve, today. This is the single biggest credibility jump available: it fills the empty calibration section with numbers instead of "first grade lands Nov 6."
Wire real feeds for the published blind spots, regulation (gov.uk / EUR-Lex), hiring (job boards), funding (registers/press), pricing and competitor moves, beyond today's procurement + academic research signals.
Coverage 0.25 reads as "thin sensing." Closing the blind spots you just published proves the sensing layer is real, not academic, and it makes the A1 backtest far richer.
Seal a batch of 30-45 day predictions on imminent, publicly-checkable outcomes, so a dozen-plus actually grade before or during the raise, not only the flagship 90-day call resolving in November.
They watch the grading machinery fire live, seal → hash-chain → resolve → score, instead of being promised it will. Proof of process beats proof of intent.
A "verify any prediction" tool over the hash-chained ledger, anchored to an external timestamp (e.g. OpenTimestamps) so backdating is provably impossible, anyone can check a seal without trusting Voundry.
The moat's whole claim is "you can't fake an honest history." This makes that claim independently checkable, the difference between a marketing line and a defensible asset.
Once Tier A gives you a number, these make it withstand a sophisticated investor's technical diligence.
A written account of how calls are generated, how confidence is set, the scoring rules, and the known limitations. The document a technical DD team asks for on day one.
Transparency about method (and its limits) reads as maturity. Its absence reads as a black box, and black boxes don't get funded at serious valuations.
Score Voundry against the naive base rate and a public benchmark (prediction markets / Metaculus / an expert panel) on identical questions.
"Beats a coin-flip" is table stakes. "Beats the crowd" is a fundable claim, it's the evidence of genuine, ownable edge.
The reliability curve and skill score fill automatically from A1 + A3 and track over time on the public page, no manual staging.
A live, self-updating scorecard is a compounding proof asset they can revisit, and it demonstrates the system genuinely runs itself.
Proof gets the meeting; a business closes the round. Investors fund a company, not a demo.
Decide who pays for what first, e.g. a "venture-radar" data subscription for VCs/corp-dev, or a diligence-signal API, well before the eventual fund.
Right now it's a proof engine with no revenue. A concrete, near-term wedge turns "interesting science project" into "company with a first dollar."
A written, gated path from today's £0 shadow mode to small, ring-fenced real capital, with explicit calibration/skill thresholds that must be hit to unlock each stage.
This is the investment thesis: "fund the proving; when the bar is cleared, real allocation switches on." A credible ladder is what they're buying an option on.
Replace the illustrative pretend-fund margins with a defensible model derived from the backtest, clearly labelled projected, with its assumptions shown.
Sophisticated money will pull on any number. Economics grounded in measured results survive that pull; illustrative ones become a credibility liability.
The items that de-risk the round and set up the next one. Lower urgency, non-negotiable before real capital moves.
A short compliance memo documenting that every source is lawful and within terms, the claim already made on the page, made defensible.
Data provenance is a standard diligence line. A ready answer removes a deal-slowing question.
Documented governance: the invariants that keep capital locked, who can unlock, and the automatic halts, hardened and written down before any real money is in scope.
Before real allocation, they need to see the brakes work. Strong controls are what make the eventual "switch to real capital" investable rather than reckless.
The tight deck: the proof (from Tier A), the wedge (C1), the ladder (C2), the moat, and why this team. Built on top of the evidence, not in place of it.
It's the vehicle. But it only lands once the numbers behind it exist, which is exactly why it comes after Tier A, not before.
All four Tier-A converters are done and public: the out-of-sample backtest engine, a sensing layer across 7 of 8 evidence types, a first-verdict clock counting to 13 Sep, and a verifiable seal ledger anyone can check. The single highest-leverage move from here is the one that arrives on its own: the first graded results on 13 Sep 2026, which turn the backtest and the sealed calls into a measured, public track record. In parallel, the two items that most move a raise are B2 (benchmark Voundry against the crowd) and C2 (the £0-shadow to real-capital unlock ladder). Say the word and I'll take either next.
