Precise · internal · the network brief

The network is separate on purpose, and the window is open.

Why Boombox and Konstant sit outside Precise, what the IP actually is, what years of research built, and the deployment thesis underneath this week's three customer arcs: Mediaocean, Polymarket, and iHeart.

compiled Wed Aug 5, 2026, New York · sources: docs/atlas/PRECISE-TEAM-MEETING-CONTEXT.md · PRECISE-DECISION-LEARNING-CAPABILITY-PLAN.md · founder direction · signed documents control equity and IP

Precise builds the software. Konstant carries it, white label, into every vertical.

Precise began by building better ways to answer difficult media questions: who created value, where the next dollar should move, and which evidence deserves to guide a decision. But an answer delivered once does not compound. The larger opportunity is that every decision Precise touches leaves behind a durable, testable learning object.

The unit of record is the Call: the forward bet the analysis decides, written down completely. What was known, what was chosen, what was predicted at what confidence; then what happened under the declared evidence rule, what contribution the comparison supports, and what the result taught the system. The first three fields go down at the moment of the bet.

decompose both sides · make the call · score it · adjust the bet · book the lesson

The order of operations is the concept: contribution analysis comes first, on both sides of the outcome, decomposing the inputs into what each earned and the outcome into its pieces. The call is the bet made on top of that analysis: change these weights, and we predict that result. It never comes before the analysis, and it is not a monument after it; it is a bet adjusted as evidence lands, written down at the moment it is made so the scoring stays honest.

Precise builds best-in-class Decisioning Economics software, applied many ways across many verticals, ad tech first. We deploy it to customers white label on the Konstant platform: each customer gets a private decision-learning network of their own, on rails that are nobody's product, and other providers can bring their capabilities onto the same network. Precise is the anchor, and the ones who join are Precise customers: they build their own capabilities, send them downstream, and replace their rent‑charging vendors one seat at a time. Konstant's independence keeps each customer's network its own. Our research remains the engine. The harness makes it executable. Boombox lets it operate inside our infrastructure, our customers' infrastructure, and, when authorized, their downstream networks. And private means the plain thing: customer evidence, data, models, and context stay inside the rights the customer granted. The network carries bounded capabilities, qualified releases, and payload-free records, never a pooled data lake.

tense discipline Today we say "we are building and launching." The present-tense "Precise creates" becomes fully earned when the first signed, mature outcome closes the live loop. PRECISE-DECISION-LEARNING-CAPABILITY-PLAN.md:21

Two companies, because the network only works if it is nobody's product.

The whole reason fits in two failure cases. If Konstant sat inside Precise, every other capability provider would reasonably ask whether it was joining a neutral network or strengthening a competitor, and the network would starve. If Precise gained nothing as the network won, Precise would reasonably ask why its product and customer learning should subsidize someone else's platform, and the anchor capability would leave.

The signed structure resolves both: economic alignment without capability capture. And the alignment is structural before it is contractual: the network consumes the Precise primitive everywhere it runs, so as the network wins, Precise wins. The equity moves were the proof of that claim before the operating evidence could land: Adam backed it personally. Precise keeps all ownership and value of its own capabilities and customers. Konstant stays independently governed, which is precisely what lets Blockdaemon, industrial companies, and other providers join without entering a competitor's estate. That is the entire politics of it, and it is settled in the signed documents, not in meetings.

The commercial shape over time is clean: Konstant knows organizational intelligence alone is a commodity; its eighteen months went into what is not: the network itself, and selection, with its own algorithms and patents. Konstant consumes the Precise primitive as a licensee. Two selectors, and neither silently acquires the other's authority. Konstant routes the organization: which capability, provider, and composition fit its context, authority, relationships, custody, timing, and commercial terms, with its own selection algorithms and patents. TikTok makes the primitive legible: it decomposes each video into the attributes that predict you watching more, then selects the next one. Konstant points that same machinery at decisions, selecting the context that matters for the call and feeding it to Precise's ablation system. Precise runs the evidence-bound court inside the decision: freeze the eligible models, methods, strategies, policies, and surfaces, apply the eval law and the floors, and return a winner, a null, or a refusal. The score teaches both.

Fig. 1Two selectors, split on the company line, sharing only the score.

And the dependency runs every direction, on purpose. Without Precise, Konstant stops at organizational intelligence: context without the capability that gives it gravity. Without Konstant and Boombox, Precise stays powerful but hard to move across companies; without the network to think in, we would have ended up building dashboards. Without the capability network, the blockchain work lacks the agentic economy it was built to support. Every piece needs the others.

Boombox is the network made operational: identity, tenancy, authority, deployment, durable work, recovery, inspection, and capability distribution, without absorbing what a capability means. AliceNet is the neutral protocol floor for the records that must outlive any one vendor or cloud.

The IP splits cleanly: Precise owns the judgment, Konstant owns the grammar.

Who	Owns
Precise	Addressable decision-learning and selection reality: decompose prior outcomes and their contributing inputs, turn that evidence into an explicit forward bet, observe and decompose the next outcome, and use the comparison to improve the weights, policy, or action that follows. The Call, Score, Contribution, Book, evidence, and provenance records make every part of the cycle answerable; courts, floors, null decisions, and refusals make it clear why a method earned use, or why none did. Stable, provider-owned, addressable — and the primitive under the order book for the internet, where contribution is priced, tested against outcomes, and settled: how any organization automatically finds the best paths to better outcomes in complex workflows.
Konstant	The independent network and its canonical ontology: Company Advocates, organizational intelligence, mandates, capability supply and demand, routing intelligence, and the shared grammar that lets independent providers interoperate without surrendering domain IP.
Boombox	The common operating protocol, playing the TCP/IP-like role: one way for independently owned capabilities to receive authority, run across different infrastructure, recover, and return authenticated operational records.
AliceNet	The inter-network contract and durability layer, and real machinery, live on Ethereum mainnet today (chain 1042): a Validium L2 anchored by validity proofs, W3C decentralized identifiers and verifiable credentials for people, assets, and organizations, and BBS selective-disclosure proofs — prove a claim without revealing the data behind it. The cryptographic form of "only arithmetic crosses company lines."

A Precise Call maps onto the common grammar; it never redefines it. That is what lets a second provider build a different evaluator or product over the same lineage without forking the network, and it is why the boundary is an asset, not overhead.

The research is older than the company, and it converges here.

Precise's contribution-value research goes back years, through MAD Network and AliceNet, with Matt, at the fundamental level. Before Precise and before Konstant, the order book for the internet was the most important thing planned on top of the AliceNet mechanism itself: contribution valued, settled, and paid, with selective disclosure keeping every party's data its own. The eighteen months belongs to the Konstant side: building, deploying, refusing, and learning until the capability-versus-network boundary settled. The Precise program has been answering three questions the whole time. Who created value: settlement-grade contribution measurement, exact Shapley offline. Where the next dollar goes: the marginal frontier, Next-Dollar serving math, TimeWalk replay of historical decisions without future-data leakage, decision-time logging for off-policy evaluation. What the evidence deserves to claim: declared outcomes and maturation windows, evidence rules, honesty floors, calibration, and refusals as first-class product results.

The architecture was not designed on a whiteboard; it settled through iteration. We are behind in explaining it, and the timing is still right, because the market's need only just became legible.

And the endgame is stated openly: Precise, with the blockchain floor underneath, becomes the thing that values and settles contribution in multi-agentic work. That is the order book for the internet, and we build it the only way it can actually be built: bottom up, from customers like Mediaocean and Polymarket, onto the networks we are already laying.

the machinery is in the repo: harness evals and floors · trace-score · research physical plans and restartable workloads · docs/atlas/HARNESS.md

alicenet status, in the code today: konstantin.ai lib/alicenet runs the first trust rung in CI — Ed25519 verifiable credentials (did:alice:konstant; operator, decision, and precedent credentials) and Merkle inclusion proofs, with Merkle load-bearing in the extraction correction ledger. The chain-anchoring adapter is a designed seam, deliberately unimplemented until the AliceNet SDK: the code reports "unavailable" rather than faking anchoring. That refusal is the house style.

The window: nobody will build their business network on rails a hyperscaler owns.

A hyperscaler will eventually build a version of this and make it free. We have run this experiment before. ABC, NBC, and the big media companies never handed their ad inventory and business workflows to Google; that is why an independent ad market and an ad tech industry exist, and why companies like MadHive exist. The hyperscalers could not vampire-attack those estates, so they built their own media companies instead: YouTube, Prime Video, Apple TV+. Company towns. AI brings the same fork to every industry, and agents coming online with crypto make the online world bidirectional the way the real world already is: any party able to buy, sell, and settle with any other. The alternative to the company town has to exist. Building it is the opportunity.

Companies will use hyperscalers for what hyperscalers are for: scale. But as real-world business interaction comes online through agents, no serious company will hand its context, its authority, its data, its model choice, and its downstream economics to one model company or one cloud. They will want to own all of that and still deploy anywhere.

There is a window before this is obvious to everyone, and the network that wins it will be the one that replicates how business actually works today: relationships, granted authority, custody, evidence, and money changing hands between parties that stay independent. Nobody should pay Google or Microsoft a tax on the operation of their own network. Konstant-style networks are the answer to exactly that: any deployment surface underneath, customer-owned everything on top, and the capability, not the cloud, as the thing that travels.

The hyperscaler is the road. The network is the commerce on it. We are building the commerce, and refusing to let the road set the toll.

The problem the network solves that nothing else can.

Company A's agent uses Company B's capability. Later, both sides, and sometimes a third party neither controls, must prove things about what happened, without anyone opening their books. Three proofs are always needed: authority (who was allowed to act), provenance (what was decided, frozen before the outcome), and contribution (who created the value, and how it settles). A hyperscaler answers this by being the trusted middleman. That is the company town. The network answers with arithmetic:

1. Verifiable credentials make authority portable. Mandates, qualified releases, booked lessons: checkable objects any party can verify, no central identity provider to trust.

2. Merkle commitments make sealing disclosure-free. Batches of Calls become leaves; one root commits them all. An inclusion proof shows this exact decision, with these exact frozen fields, was committed at time T, revealing nothing else in the batch.

3. Anchoring gives the network a clock nobody owns. The root, never the data, anchors through AliceNet's validity-proof path to Ethereum. "Sealed before the outcome" becomes checkable across company lines.

4. Selective disclosure lets the proof cross while the data stays home. BBS proofs show the mandate covered it, the release was qualified, the track record clears the acceptance rule, without opening a single book.

5. Contribution settlement completes the order book. Credentialed demand, committed executions, scored resolutions, contribution splits: every row provable by exactly the party entitled to it, no row public.

Fig. 2The problem the network solves: authority, provenance, and contribution proven across the line while every book stays home. The middleman is replaced by arithmetic.

Deployment climbs a trust ladder: signed local (Ed25519 credentials and Merkle proofs, running in CI today) → Merkle-committed between contracted parties → chain-anchored for open-network and regulated work. Different relationships buy different rungs, and the code refuses to claim a rung it cannot perform.

AI-native is the word. We mean the real thing.

Most companies saying "AI-native" mean they use models. A true AI-native organization is one whose context, authority, evidence, and economics can be operated by agents without being surrendered to a model company or a hyperscaler. A true AI-native network is organizations doing that with each other: capabilities traveling, authority granted per place, contribution valued and settled between parties that stay independent. That is what Boombox and the capability network build, and it is the standard the word should mean.

The belief underneath all of it, stated plainly: capability networks, and capabilities like Precise, become primitives that everything in the world uses as a tool. In AI-native work, the operating system and the networks beneath it play the role DNS plays on the current internet: everything resolves through them, nobody thinks about them, and they belong to no one.

Like DNS: invisible, everywhere, and nobody's product.

Deployment travels the way trust does: one admission at a time.

build and prove the capability locally → operate the qualified release in Precise-owned infrastructure → separately admit the same release into customer infrastructure → separately authorize any downstream customer relationship

Authority never arrives because two companies are adjacent in a tenant tree. Each target gets its own identity, policy, custody, billing, and record chain. That is the network-style deployment point of view in one breath: the capability stays constant, the trust is granted per place.

And it is how the incumbent estates open up. Enter through one valuable capability. Run beside the existing systems and learn the real workflow. Rebuild the business function as a governed product on customer-owned context. Let the customer carry it into their own teams and downstream graph under explicit rights. The attack targets closed distribution and rent, never another company's data or IP; customers keep their estates, capability creators keep their products, and the better network becomes the place where the work happens.