Part 0 - Overview#
Contents
Two entry points to the same argument, at different compression ratios.
| Read if | |
|---|---|
| Thesis | You want the argument in five steps, with links into the evidence for each |
| The compressed version | You will read nothing else, and want the conclusion plus what to do about it |
| Notation | Recurring shorthand (master asymmetry, inelastic complement, …) |
| Dependencies | Claim → pages that rest on it (for scoring misses) |
| Scenarios | Best / base / worst joint worlds (stress tests of the chain, not a fan that replaces it) |
The spine#
Everything in this corpus descends from two sentences colliding:
Capability grows fastest where verification is cheap. Value accrues to whatever intelligence cannot manufacture.
The first is a claim about the technology, forced by data exhaustion pushing the frontier onto verifiable-reward training. The second is a claim about economics, forced by inference cost collapse making cognition itself abundant.
The first explains the uneven progress across every domain - why math and code raced ahead of taste and judgment. The second explains where the money goes - to energy, land, licenses, distribution, trust, and physical presence rather than to the technology or to most of the firms adopting it.
Together they produce the document's least intuitive conclusion: the transition is real, large, and mostly invisible in the aggregate statistics, because competitive markets pass the gains to consumers and the concentrated harms land on cohorts and regions rather than on national averages.
Note what the spine does not claim. It does not say capability stops growing where verification is expensive - it says growth there is slower and gated on buying cheaper ground truth (data names the escape routes). And it does not say the scarce-complement list is fixed - Uncertainty 7 exists precisely because that list has a half-life. The spine is a claim about gradients, not endpoints, which is why it can be checked quarterly (Part VII) rather than only at the finish line.
Each sentence has a designated adversary. Uncertainty 5 attacks the first: if learned verifiers get good enough, the cheap-verification gradient flattens and the domain ordering in Part III reshuffles. Uncertainty 7 attacks the second: intelligence plus capital can, over time, manufacture substitutes for several items on the complement list, and every distributional prediction in Part IX is exposed to how fast that happens. A reader who wants to break the corpus efficiently should start with those two pages; everything else is downstream.
Why two compression ratios#
Compression is not a convenience here; it is a test. An argument that survives being cut to one page (thesis) and then to one paragraph (compressed) has a spine; one that only works at full length is a mood. The two short forms also serve as commitment devices - they state the conclusion baldly enough that a miss cannot be reinterpreted afterward, which is what scoring needs. If the long form and the short form ever disagree, the short form is the claim and the long form has drifted; that check belongs to the Review step of every autoresearch round.
Reading orders#
- Linear - follow the Next link at the foot of each page, starting with Thesis. Runs about 90 minutes.
- Argument-first - Thesis → Part I → Part II → stop. That is the half that carries the argument.
- Sceptic-first - Part VI and Part VIII before anything else, to see where it is weakest and how the numbers were made.
- Operator-first - Compressed → Game 3 → Part VII, for what to build and what to watch.
Each order has a failure mode worth knowing before you pick. Linear front-loads substrate detail and can bury the economic punchline for a reader who only wanted the conclusion. Argument-first stops before Part VI, so it delivers the case without the counter-case - do not cite the corpus from that read alone. Sceptic-first inverts that risk: uncertainties read as fatal when you have not yet seen the structure they are attacking. Operator-first optimizes for action and quietly assumes the framework is right; an operator who skips the dependency index will not notice when a miss upstream invalidates the thing they are building. There is no order without a blind spot - the orders exist so you can choose which one you accept.
Site navigation#
On future.lulzx.space: Find (bottom bar, or / on desktop) searches all pages via catalog.json. Theme toggle is t. Arrow keys page when prev/next exist.
When the short forms disagree with the long form#
Prefer compressed / thesis as the claim and treat the long page as drifted until reconciled in a Review step. That is not a style preference - it is the commitment device that makes scoring work.