A Sourced Investigation · by Michael Polzin
Big Tech
Accountability
Cloud economics, the SaaS lock-in machine, and the people and places that pay for it — told entirely from the public record. Every claim links to its source. Opinions are labeled. Nothing here alleges a crime.
In what world would it be true?
You already hold a model of how this economy works. Before the evidence, guess. The distance between your guess and the record is the surprise — and surprise is exactly what updates a map. Drag each slider to your honest estimate, then reveal.
What this evidence is made of
Every claim in the 12-part series cleared a legal/ethics gate; the newer human-cost, planetary-cost and systemic sections hold to the same standard and are under review (see Method). Here is the exact mix — and how much is anchored to primary documents you can open yourself.
Six ways into one world
The receipts, the science behind them, the meaning underneath, and the way forward — held to one standard of evidence.
Where the value goes
The cloud is the most profitable machine in the history of business. Here is how big the numbers are, how one assumption moved billions, and who ends up holding the gains.
The numbers at stake
The largest dollar figures pulled directly from primary-sourced facts in this investigation, drawn to scale. Each links to the claim it comes from.
One assumption. Billions in profit.
In 2023 Alphabet changed a single accounting estimate — how long a server is assumed to last — from four years to six. Drag the slider to see how that one number moves reported profit, using Alphabet's own disclosed 2023 figures.
server life
Illustration. Holds Alphabet's depreciable asset base constant and scales from the company's disclosed figures: the 4→6 year change cut FY2023 depreciation by $3.9B and raised net income by $3.0B ($0.24/share). Source: Alphabet Form 8-K, Exhibit 99.1 (SEC, 2024-01-30).
The cloud runs on people you never see
Behind every "AI safety" filter and every clean SaaS dashboard is a workforce — often offshore, often contracted, often paid a fraction of a living wage to absorb work no one else will. The figures below are documented. The conclusion is labeled as opinion. You decide.
Opinion · Michael Polzin The author's interpretation — expand to read
"The system was built to make the cost of this labor invisible — and to keep its value flowing to the people furthest from the harm. No bad people. A bad system, working exactly as designed."
A data-center economy has a physical bill
Compute is not weightless. It is electricity, water, and land — drawn from real grids and real communities, increasingly to train and run AI. Here is what the meters say.
The system is editing your map of the world
An attention economy is engineered to maximize engagement, not your understanding — and what we attend to becomes who we are. This is the most contested terrain in the investigation, so the science is labeled carefully: established findings as fact, the disputed causal claims as exactly that.
Opinion · Michael Polzin The author's interpretation — expand to read
"We carry our ancestors as priors — the model the world built in us over generations. A feed tuned to the next swipe overwrites that model with noise. Rebuilding an accurate map of the world is how a mind, and a people, finds its way home."
You are not the first to say this
Scholars, regulators, engineers and whistleblowers have documented this machine for years — on the record, under their own names. And the record also shows what it has cost some of them to speak.
The verified math — and the honest fences
The most provable part of the program, and the most carefully bounded. Universal Natural Intelligence is standard active inference, evidence-classed from a verified mathematical core to a 100-animal bestiary — with the speculative frontier marked "not claimed."
What faith does — and what it cannot settle
The gentlest, most clearly-labeled wing. Read through active inference, faith and ritual look like tools for keeping a shared map coherent. This studies that function — names no deity, and settles no theology.
Guardrail "This framework can neither prove nor disprove God. It studies how belief, ritual and meaning operate in human communities, and stops there."
Grow the World
The regenerative pole — and the open invitation. Calibrated signals and cooperation grow the world; this is where the receipts point, the anthem plays, and every frontier claim is published as a testable hypothesis.
Every frontier claim, with its falsification test
A claim earns its place by being testable. Here is the register — each one published with the specific result that would prove it wrong. Bring that result, and the record updates.
Every claim, every source
The full verified dataset behind the series — 12 parts, each claim labeled fact, allegation, or opinion, each fact linked to a primary or reputable source. Search it, filter it, check it.
How this was sourced
This investigation runs on one rule: if a sentence of consequence can't be tied to a source you can open, it gets cut — not softened.
Traces to a primary source — a regulatory filing, court docket, company disclosure, or peer-reviewed study — or to at least two independent reputable outlets.
Attributed to whoever made it, on the record, with the filing or statement cited. We never restate it in our own voice as settled fact.
Clearly labeled as Michael's interpretation, containing no embedded false fact. You can filter it out entirely in the Receipts.
We do not assert that any identifiable person or company committed a crime. The human-cost and planetary-cost sections apply the same standard: documented wages, settlements, court rulings, filings and peer-reviewed measurements as fact; harsher characterizations carried only as labeled opinion or quotes attributed to named sources. Where companies dispute or contextualize a point, we link their own public statements. Found an error or something unfair? Reach Michael Polzin, or call 262.914.2929 — corrections are made promptly and noted here.
Check the receipts the browser just checked
Every dataset on this site is hashed at build time with SHA-256, and your browser verifies those hashes on load. The panel below is live — if anyone modified a file after the build without re-publishing, you will see TAMPERED right here, not buried in a log.
Help us build the real ledger
The hash check above is what we could ship today — useful, but partial. Luis Fernando Martinez Chavez (Founder & Principal Architect, High ArchyTech) laid out what an honest evidence explorer actually requires. We are publishing his words and asking the community to help build it — properly.
“Michael, respect the lack of ego. To build a true evidence explorer, you must separate UI scaffolding from cryptographic truth. Evidence cannot live on ephemeral compute. Here is how you architect the ‘home’ and ‘design’ at the application layer:
The Home (Stateful Ledger): Get off serverless for data ingestion. Evidence requires strict, append-only non-repudiation. You need a dedicated CP (Consistency/Partition Tolerance) architecture, a Write-Ahead Log (WAL) or an immutable ledger.
The Ingestion (Cryptographic Receipts): The system generating the event must sign the payload natively (e.g., Ed25519). Your backend does exactly one thing: it verifies the signature and synchronously fsyncs it to disk.
The Design (Read-Only Verification): Your Vercel frontend becomes a pure read-only viewer. But the browser doesn’t just blindly trust the API. The frontend fetches the payload AND the signature, mathematically verifying the hash client-side before rendering. If the hash fails, the UI instantly flags: ‘Tampered.’
At Layer 7, your job is to cryptographically prove to the user that the receipt hasn’t been altered by a single byte since execution.”
Where we are, honestly
- What works today: SHA-256 hashing at build, manifest pinned to a git commit SHA, client-side SubtleCrypto verification on every load, a visible tampered state. The "design" layer Luis describes — read-only viewer that flags mismatches — is now live above.
- What is still missing — the home and the ingestion: claims are authored in JSON and committed to git. There is no native Ed25519 signing at the point of ingestion. There is no append-only WAL, no CP-architected ledger, no fsync-on-write backend. Hash-pinning verifies that what we served matches what we built; it does not prove a source signed the original payload, and it does not survive a CDN compromise that replaces both data and manifest.
- What we will not pretend: this site, today, is a journalism-grade evidence viewer with a verification layer bolted on. It is not yet a cryptographic ledger. Calling it one would be exactly the kind of overclaim the rest of this site is built to refuse.
If you can help, here is the work
- The Home — append-only ledger. Stand up a CP-leaning store (immudb, QLDB, a WAL-fronted Postgres, or a Merkle-anchored equivalent) that holds every claim revision with non-repudiation. Cryptographic engineers, database architects.
- The Ingestion — Ed25519 receipts at source. Key management for the author and contributing investigators, sign-at-source tooling for claim payloads, a verifier that fsyncs before acknowledging. Applied cryptography, key-handling UX.
- The Design — verifier hardening. The current browser verifier reads its manifest from the same origin as the data, so a CDN compromise defeats it. We need out-of-band manifest publication (a signed transparency log, IPFS pinning, mirrored Merkle roots) and a verifier that warns when the trust chain narrows to a single host. Web-platform engineers.
- Threat model and review. An adversarial read against the threats this actually defeats and the ones it does not, with concrete attack walkthroughs. Security researchers, journalists who have run a target.
- The covenant. Whatever we build belongs to the public record, not to any one person or company. Help us write that covenant so contributors know that up front.
If any of this is your craft, please book a conversation, message Michael on LinkedIn, or call 262.914.2929. This explorer was built in the open by one author; it can only become the world’s ledger if it stops being one author’s project. Bring the better architecture, and the record gets stronger.
The readers who made it sharper
This investigation improves when people who know the field push back. Here are two observations from Brad Wolfe that changed how this site presents its evidence — with appreciation.
“The strongest material is where the documentation is unambiguous — the no-poach settlement record, the Alphabet server life disclosure, the SEC filings. Those are facts that speak for themselves and do not require the reader to accept a framing to find them useful. The data is the argument. It does not need a narrator.”
“The CFO certifying financials built on AI-assisted models, without documented decision rights or audit trail, is already signing something they may not be able to defend. Most have not done that math yet. When they do, the internal forcing function arrives. Whether that happens before the SEC letter or after is the only variable left.”
Brad's first observation led directly to the collapsible opinion framing on this site — interpretive quotes now require a click to expand, so the facts carry the argument on their own. His second points at the next frontier: AI governance liability, the SOX 302 equivalent that is not yet written. Spotted something off, or have a signal that sharpens the record? Bring it.
Where this investigation is least certain
An honest map shows its own fog. These are the claims we hold with the lowest confidence, or that are labeled allegation or opinion rather than fact — surfaced on purpose, drawn from both datasets. Strong claims don't fear a list like this.
Transparency & limits
Two datasets, two stages of review
The 12-part series (The Receipts) has cleared a documented legal/ethics gate. The human-cost, planetary-cost and systemic sections are newer — rigorously sourced and labeled by the same taxonomy, but still awaiting that formal gate review. We tell you that rather than hide it.
What is contested is labeled contested
Whether trauma is inherited across generations through the germline, and whether social media caused the teen mental-health decline, are genuine open scientific debates — we cite the skeptics. Characterizations of the AI build-out as a "bubble" are the attributed opinions of named analysts, not our verdict.
Sources, dated and preferred primary
We prefer primary documents — filings, dockets, disclosures, peer-reviewed studies. Where a publisher blocked automated access, we cite a faithful republication and say so. Every figure carries a date and may drift as the world moves.
Confidence is our own estimate
The percentage beside each claim is the author's calibrated estimate, not a measurement of truth. The calibration ledger above foregrounds the lowest-confidence and contested claims on purpose — an honest map shows its own fog.
Projections and assembled figures
Some numbers are forecasts (IEA 2030 power, 2027 water, e-waste) or sums the author assembled from underlying sources (the 658,617 layoff total; the top-10% stock share computed from two Fed series). These are labeled in context, not presented as single official statistics.
What we never do
We do not assert that any identifiable person or company committed a crime — not as fact, not as "opinion." Settlements resolve claims without an admission of guilt; we never present one as a verdict. Where a company answered on the record, we link its words. Found an error? Reach Michael — corrections are logged.
This is a hypothesis.
Falsify it.
Everything here is offered as a calibrated signal, not a final verdict. That is the scientific posture, and it is the honest one: a theory earns its standing only by surviving every serious attempt to break it.
So break it. If you can show a figure is wrong, a quote misattributed, a claim stronger than its source — bring the better-sourced signal and the record is corrected. That correction is not a defeat; it is the entire point. It is how a shared map of the world gets more accurate, and how the people living inside it stop paying for a distorted one.
Where the evidence is uncertain, we have said so. Where it is opinion, we have labeled it. Nothing here accuses a person of a crime. Until a claim is falsified, the documented record stands — and setting the record straight is where the healing starts.
Speak in calibrated signals. Falsify what you can. Bring it to Michael →