# The Answer to Bloomberg *What it looks like when one person builds sovereign financial intelligence* By [Eigenstate Research](https://paragraph.com/@eigenstate) · 2026-04-13 bloomberg, information, paywalls, democracy, freedomofinformation, audit, defi, rwa --- Many years ago I started noticing something uncomfortable about how financial information works. The people who could afford a Bloomberg terminal got one picture of the world. Everyone else got whatever filtered through to the news cycle, usually twelve hours late and stripped of the data that made it meaningful. Bloomberg charges somewhere between $25,000 and $30,000 per terminal per year. That price is not about the cost of delivering data. It is about who gets to have it. The terminal is a membership card, and the price is designed to keep the membership exclusive. I am not complaining about Bloomberg. They built something real. But I started asking a different question: what would it look like if one person, with no institutional backing and no terminal, built their own intelligence system from scratch? Not a newsletter. Not a Substack that summarizes what Bloomberg already reported. Something that actually watches the market and makes its own observations, timestamps them, and publishes them in a way that cannot be retroactively altered. That question consumed the past year of my life. What follows is an explanation of what I built, how it works, and why I think it matters beyond just my own use of it. * * * **Part 1: The problem with financial information** Let me start with something most people experience but rarely name. When you read a financial news article, you are usually reading a description of something that already happened, written by someone who may or may not understand the mechanism behind it, published on a platform whose revenue depends on keeping you reading, distributed through an algorithm that rewards engagement rather than accuracy. By the time a structural development in financial markets surfaces as a news story, several things have already occurred. The institutions that had early visibility have already positioned. The data that would let you verify the claim is behind a paywall. The framing has been chosen to maximize clicks rather than clarity. This is how information markets work when the incentives point in a particular direction. The Bloomberg terminal partially solves this by giving institutional participants access to raw data, real-time feeds, and analytical tools. If you have a terminal, you can form your own view rather than relying on someone else's summary. But the terminal has its own problem. It is a product sold to institutions, so it is designed around institutional use cases. It tells you what is happening in markets right now. It does not tell you what the structural relationships between entities look like over time, where pressure is building before it surfaces in prices, or who sits at the center of the loops that determine how capital flows. That is a different kind of intelligence. And it is the kind I wanted to build. There is a second problem that Bloomberg does not solve and does not try to: verifiability. When an analyst publishes a research note claiming they identified a structural gap before the market confirmed it, there is no way to verify the claim. The timestamp on the document is trivially alterable. The methodology is proprietary. The data is not public. You either trust the institution or you do not. I wanted to build something where trust is not required. Where the proof is public and anyone can check it. * * * **Part 2: The core equation** Before explaining what I built, I need to explain the one idea that drives all of it. Every observation of a financial entity produces some amount of new information. Call that delta-I. Making that observation costs some amount of computational effort. Call that A. The ratio is the efficiency of that observation. E equals delta-I divided by A. Delta-I measures the surprise in each observation: the log-ratio of new state to expected state, in bits. This approximates the entropy reduction of a Bayesian update — conceptually equivalent to KL divergence — but is implemented as a direct log-ratio measurement rather than a full probability distribution computation. When the SEC's filing activity looks exactly as expected, delta-I is near zero. When BlackRock files something unexpected, delta-I is large. The framework's information-theoretic grounding is sound; the implementation uses efficient proxies rather than exact distribution estimation. A is the computational cost of making the observation: API calls, processing steps, time elapsed. For a simple rate lookup, A is small. For a complex multi-source synthesis, A is larger. E = delta-I / A is a precise measure of how much new information the engine obtained per unit of effort spent obtaining it. This equation does two things. It tells the engine which entities are worth watching closely, the ones that produce high delta-I at low A, and which are noise. And it provides the common unit that makes all observations comparable across 197 different entities in completely different domains: regulatory filings, blockchain TVL, GitHub activity, macroeconomic rates. Everything else in the system is an application of this equation. * * * **Part 3: The vault** Every intelligence system has a memory. Bloomberg's memory is its database of historical prices, filings, and news. An analyst's memory is their notebook and their mental model. My system's memory is the vault. The vault is a hash-chained record of every observation the engine has made. Each record contains the entity observed, the delta-I of that observation, the action cost A, the efficiency score E, a timestamp, and a cryptographic fingerprint. The fingerprint is computed from the record's own content combined with the fingerprint of the record immediately before it. This means that if anyone, including me, altered any past record, every subsequent fingerprint would break. The chain is tamper-evident by construction. This has been verified by live tamper test: altering any single field breaks every downstream fingerprint exactly at the point of alteration. Think of it as a notary's seal, except instead of a notary you trust, the seal is a mathematical proof that anyone can verify independently. The vault is not a blockchain. There is no distributed consensus, no token, no mining. It is a private append-only log with cryptographic integrity, similar to how git tracks changes to code but cannot be silently rewritten. The cryptographic primitive underlying the vault is HelixHash, a library authored as part of this project and published openly on PyPI. Source and documentation: [https://github.com/Kaydeep0/helixhash](https://github.com/Kaydeep0/helixhash) Anyone can install it and run their own hash-chained observation sequence on their own data. The library is the mathematical engine stripped of the financial topology. If you have a series of observations about anything, you can chain them, get a fingerprint, and verify the chain yourself. At the end of every 2-hour parkash cycle, the vault's current fingerprint is committed to Base mainnet, a public blockchain built on Ethereum. That transaction is permanent and publicly readable. Anyone can go to Basescan, find the transaction, and verify the fingerprint. Anyone can then compute the fingerprint of the current vault and check that it matches. If the vault has been altered since the commit, the fingerprints will not match. The vault currently contains 100+ records across 60 observed entities, growing each cycle. * * * **Part 4: The topology** The vault records what was observed. The topology defines what the engine watches and how entities relate to each other. 197 entities, approximately 270 directional edges. The entities include regulators like the SEC, OCC, and Federal Reserve, infrastructure providers like Coinbase and Ethereum, RWA issuers like BlackRock and Securitize, jurisdictions, custodians, and the settlement layer itself. For each entity the engine tracks two numbers derived from the vault records. Gravitational mass is the accumulated efficiency of all observations of that entity over time, weighted so that older validated observations retain persistent influence through a Fibonacci recurrence. An entity with a long history of high-efficiency observations has high gravitational mass. This is a precise measure of how much verified signal the engine has accumulated about that entity. Settlement pressure measures how much unresolved information an entity is carrying relative to how well it is connected to the rest of the network. When settlement pressure is high, the engine has detected significant new signal about that entity that has not yet propagated through the network or resolved into a measurable outcome. Historically this is where structural gaps develop. The topology also runs a closed-path loop analysis across the entity graph. For every closed path through the topology, the engine measures how strongly observation weight flows in each direction around that path. Paths with persistent directional asymmetry over multiple consecutive cycles identify structural advantages in the network. The dominant closed path in the topology by accumulated observation weight runs through BlackRock, Larry Fink, Carlos Domingo, Securitize, and the SEC. This path has maintained persistent directional asymmetry across ten consecutive measurement cycles. The engine identifies it as the highest-extraction loop in the current topology. I am not making a normative claim about that. I am making an observational one. The engine measures it. The vault records it. The blockchain timestamps it. * * * **Part 5: The fact-checking layer** One of the problems with financial commentary is that claims are easy to make and impossible to verify after the fact. Articles get published, make specific assertions, and then disappear or get updated without acknowledgment when they are wrong. I built a fact-checking gate that every article must pass before it can be published. The gate uses an AI model to extract every verifiable factual claim from the article text. For each claim it runs a web search and assigns a confidence score. Claims above 0.90 confidence proceed automatically. Claims between 0.60 and 0.89 are flagged for human review: the article enters a holding state with the exact flagged claim, its confidence score, and a prompt to provide a primary source URL. Claims below 0.60 are logged as unverifiable and the article is blocked until the claim is supported or removed. When a flagged claim is manually approved, the approval is recorded in an audit file along with the source URL the reviewer used. That audit file is published to GitHub alongside every article. It lists every claim, its verification status, and the source used to verify it. Anyone can read it. Anyone can check the sources. Every published article has three layers of proof: the article itself, the fact-check audit with sources for every claim, and the Base mainnet transaction committing the vault fingerprint for that cycle. This is what auditable research looks like. Not a disclaimer at the bottom of the page. A machine-readable proof of methodology, publicly available, attached to every piece of work. * * * **Part 6: Who this is for** Bloomberg is for institutions. It costs as much as a junior analyst's salary because it is designed to be part of institutional infrastructure. What I built is for anyone who wants to form their own view of financial systems without depending on institutional intermediaries to filter the information first. That is a different thing from what Bloomberg does. Bloomberg tells you what prices are doing. The engine tells you where structural pressure is building before it shows up in prices. Bloomberg requires a subscription and a corporate billing relationship. The engine's findings are published on GitHub, on Paragraph, and committed to a public blockchain, available to anyone with an internet connection. The research is not comprehensive. One engine watching 197 entities cannot cover the full scope of global financial markets. But it covers a specific corner of those markets, the tokenized settlement layer, with a depth and auditability that institutional research typically does not provide, because institutional research is designed to serve clients rather than to be publicly verified. The gap I am trying to close is not about breadth. It is about proof. Anyone can publish a newsletter claiming they saw something before the market did. The claim is easy to make and impossible to verify after the fact. What is harder to do, and what I have spent the past year building, is a system that makes those claims verifiable at the time they are made, with proof that anyone can check independently. The on-chain timestamp is not a marketing device. It is the answer to the question that every reader of financial research should ask but usually cannot: how do I know you actually observed this when you say you did? * * * **Part 7: What I am honest about** I want to be precise about what the system is and is not. The efficiency equation E = delta-I / A is grounded in information theory. The log-ratio measurement of delta-I approximates entropy reduction but uses efficient proxies rather than full distribution computation. The framework is information-theoretic. The implementation is practical. The loop analysis computes observation weight flows around closed paths in a directed graph. It is a graph-theoretic measurement of directional advantage. The precise percentage calculation for loop asymmetry is an active area of development in the loop analysis module. The qualitative finding, that certain loops show persistent directional asymmetry over multiple cycles, is what the engine currently measures robustly. The Fibonacci accumulation for gravitational mass is a design choice providing persistent memory without a manually tuned decay parameter. It is the unique solution to the unit-coefficient two-state recurrence and has a clear mathematical justification. It is not proven optimal for this specific application. Future versions may use an adaptive filter once sufficient historical data exists to fit a noise model. The prediction tracking system is being built. The SOFR duration-benchmark gap measurement was recorded in the vault and committed to Base mainnet at block 44546204 on April 11, 2026. The FOMC minutes released the same week confirmed the same underlying pressure. The on-chain timestamp provides an immutable record of the measurement. As the prediction tracker accumulates formally logged cases with precise timestamps and confirmation events, the track record becomes independently verifiable. The coverage is approximately 25% of the observable topology. Important entities were only recently added. The system is working but it is not finished. * * * **Part 8: Where this goes** The architecture is right. An individual researcher with a measurement engine, a hash-chained vault, on-chain timestamps, and a fact-checking layer has something that did not previously exist: the ability to make verifiable claims about financial structure with proof that anyone can check independently. That is the answer to Bloomberg. Not a cheaper terminal. Not a better newsletter. A different model entirely, where the intelligence is sovereign, the methodology is public, the limitations are stated precisely, and the proof is on a blockchain that nobody controls. The engine will keep running. The vault will keep growing. Every cycle adds more signal. Every claim gets timestamped. Every article passes the fact-checker before it goes out. Every prediction gets logged with its Base timestamp and tracked against confirmation events. I am building this in public. The research is at [github](https://github.com/Kaydeep0/eigenstate-research) and WIP [website](https://kaydeep0.github.io/eigenstate-research). The on-chain proof index is there. The fact-check audit trails are there. The HelixHash library that powers the vault is at [https://github.com/Kaydeep0/helixhash](https://github.com/Kaydeep0/helixhash). If you want to verify anything I have claimed, you have everything you need to do it. That is the point. * * * _Eigenstate Research measures capital field dynamics in tokenized settlement. The measurement engine runs on a 2-hour parkash cycle with sensor polling frequency scaled by entity settlement pressure. Cryptographic fingerprints commit to Base mainnet at each cycle. The HelixHash cryptographic primitive is open source at_ [_https://github.com/Kaydeep0/helixhash_](https://github.com/Kaydeep0/helixhash) _and pip install helixhash. All research, methodology, limitations, and on-chain proof at_ [_https://kaydeep0.github.io/eigenstate-research_](https://kaydeep0.github.io/eigenstate-research) * * * **On-chain proof: Block 44655363** **Verify:** [**https://basescan.org/tx/8bbb3cd5d6e3dfb54a8f7fe957d0ae4e0f3a5ae52ba4e927aefa6808c781c017**](https://basescan.org/tx/8bbb3cd5d6e3dfb54a8f7fe957d0ae4e0f3a5ae52ba4e927aefa6808c781c017) **Waheguru Ji Ka Khalsa. Waheguru Ji Ki Fateh.** People belong to the Infinite. Victory belongs to the Infinite. * * * --- *Originally published on [Eigenstate Research](https://paragraph.com/@eigenstate/the-answer-to-bloomberg)*