What we are building, and what we find

HyperC is opening a community round

We want the people who use HyperC, challenge our research, and believe in the future of autonomous businesses to have an opportunity to become investors in the company behind it.

HyperC is testing the waters for a potential offering under Regulation Crowdfunding. No money or other consideration is being solicited, and if sent in response, it will not be accepted. No offer to buy securities can be accepted, and no part of the purchase price can be received, until an offering statement on Form C is filed and then only through Wefunder’s platform. An indication of interest involves no obligation or commitment of any kind.

Two ways to beat a market

Finance theory allows two sources of excess return in a market without price discovery: be cheaper to operate, or be right where others are wrong. The seven reasons a deal gets rejected sort between them, and 67 newly screened markets show which side each family lives on.

Beyond the catalogue: 67 new computable markets in one page

67 markets outside the catalogue, scored on nine criteria. Ten clear the bar today; one rule predicts fit better than any criterion: anyone can register, few can execute.

Where inefficiency can be captured

The theory behind computable markets: seven mechanisms from Grossman–Stiglitz to the winner's curse, which side of the edge each one sits on, and the statement that can be falsified.

Candidate markets — the waiting list

Every market we have screened but not catalogued, with scores, evidence and a first experiment. Proposed blueprints, not catalogue entries.

The school of fish

One scenario for how economic bot wars develop: once the obvious deals are gone, traders defend a market by generating noise — and filtering noise is itself a barrier to entry. Whether you run out of compute or run out of surviving evidence, it is the same computational efficiency problem, which is why fast models on small compute may decide who can still operate.

Teaching P34 to learn markets, not memorize them

A business only knows what happened to the deals it took. Training on that history produces a model that is confidently wrong exactly where the evidence is missing. How we train against synthetic worlds whose hidden truth we hold, label the model on whether it was right about its own predictions, and make competing interpretations of a market compete.

P34 needs a quantity. Some markets do not have one.

The model requires an explicit volume per deal. On markets where a deal is one indivisible thing, sizing leaves the regularized path and selection falls back to portfolio-level risk calibration — and adversarial action on the individual deal is lost entirely. What that costs, and the three approaches we are weighing.

The Replicator Economy — our vision statement

Why the bottleneck was never production, why we started by scoring a single deal, and why the alignment target has to widen from one business surviving to an ecosystem of collaborating businesses compounding together. To be updated and expanded.

Founding membership opens

The P34 Membership opens to builders: the decision model, a managed agent workspace, market intelligence and the community — at founding terms, with the profit-share rate fixed by registration order for the first cohorts.

Computable Markets: the working paper

The operator-relative theory of computable markets, the Menu-Sales-Description data contract proposed for the category, application templates across ten-plus verticals, and a falsification protocol — with the notebook results, including the selection-bias trap measured end to end.

P34: Learning When Not to Trade

The technical report behind the model class: benchmark construction under partial observability, selection bias and regime change; baseline tuning; results and stated limitations, with notebooks published for inspection.