The whole book runs on one science lane, standalone-readable here. Every term is defined in the Notation & Glossary; this page is how the pieces fit. (Reframing, not a claimed solution to the free-will problem — see the brake at the end.)
1. The one free thing
There is a deep result (Čencov, 1972): the geometry of distinguishing one state of the world from another is unique — fixed, not chosen. On that fixed substrate sit your dynamics (D, what you do) and the mechanism you are entangled with (M). Exactly one thing is left free: the reference (Y) you check yourself against — in plain terms, what you align to, your true north. Everything in this guidebook is an operation on that one free choice. (And note already the shape of the whole book: Y is your reference whatever it is — a good north or a bad one. An idol is a Y too; the chapters that follow are the anatomy of choosing well or badly.)
2. Free will, not fate — and it’s a result, not a slogan
Because the substrate geometry is time-reversal symmetric, it forces no arrow of time at all. A “forward,” a fate, a destiny, only appears when a boundary breaks the symmetry — and which direction counts as forward is itself a choice of reference. So there is no rail. Fatalism — the attractor, the prophecy, the inevitable singularity — is a created direction laundered as a found one. Free will is precisely the choice of the reference.
3. The reference and the penalty
Take two channels, D and M, that both bear on some outcome. Condition on a reference Y and measure how much D and M still share once Y is accounted for: the explaining-away penalty I(D;M|Y). When Y is a genuine, independent common reference, the penalty drops to zero — D and M are fully explained through Y, the clean three-point geometry. When Y is coupled to the very D and M it is supposed to referee, the penalty is forced positive. The penalty measures how badly your chosen reference fails to be an independent reference.
4. Why a low penalty is not safety — the detection gap
A created or captured reference makes the penalty read low, because the system grounds in it perfectly. So a low penalty does not mean grounded; it means grounded or cleanly captured, and the penalty alone cannot tell which. Pushed to its exact form: conditioned on the reference you actually use, the penalty falls to ≈0 whether that reference is read or created — it is structurally blind to provenance (and the naive rule “minimise the penalty to be safe” reads lowest at total capture). The meters that do sort read from outside the loop: transparency-spent (how much the mechanism still tracks the external referent) and the cut-survival itself. This is forced: you cannot validate a channel from inside it. A true reference and a captured one both produce coherent, low-penalty experience. Network science meets the very same wall and proved it a theorem — the homophily–contagion confound: from the correlations alone, influence and shared-trait are generically inseparable. Every culture that engaged a guidance channel hit this wall and built external apparatus to get around it.
5. Provenance over coherence — the one move that survives
Since content cannot distinguish read from captured, the discriminator has to be origin: is the reference’s source independent of the act of consulting it, verified from outside? That is the one move that survives the detection gap — and it is cross-culturally vetted (the traditions that lasted all sort by provenance, not by whether a sign “came true”), re-derivable from the penalty (a captured Y is, by construction, a function of D and M, so only an origin-check ever exposes it), and the same escape causal inference itself converges on: the way past the homophily–contagion confound is not a cleverer look at the data but an intervention. The runnable form is cut the loop: perturb the consultation and watch whether the reading tracks an external domain or just the apparatus. It now carries a validated instrument — the cut-ratio: the part of the signal that survives the cut is the read reference, the part that collapses with it is the created one, and the ratio between them reads the provenance on constructed ground truth where the answer is known in advance.
6. The one free move, and the three registers
Collect the mechanics: substrate fixed (Čencov), dynamics (D, M) given, arrow not owned ⇒ the sole free degree of freedom is the choice of Y. That is the whole content of free will — not act-selection (wiring-determined) but what you ground in. So the operations catalogue is the anatomy of this one move: the reference surrendered (idolatry), counterfeited (the ouroboros/oracle), protected (the bounds), restored (resurrection), or left un-authorable (the apophatic apex). And the same move shows up in three registers with no copying between them — temple, clinic, laboratory — which is why this is one architecture, not one tribe’s sermon.
Brake. This is a reframing — localizing the one freedom at the reference (Y) plus the cut-the-loop provenance test — not a claimed solution to the free-will problem (prior lineage: James on will-as-attention, Frankfurt, Frankl, compatibilism). The Čencov/Fisher substrate result fixes the metric, not any numerical constants. Secular re-inventions are convergence, not a confirmation count. The Fisher substrate also has a living cousin in cognitive science — the free-energy principle / active inference (Friston), itself built on Fisher-information geometry — and the honest relation is prior-art lineage in the same kind (an information-geometric account of mind), cited and not claimed over: the same mathematical neighbourhood, a different object (prediction-error minimisation, not the I(D;M|Y) provenance penalty). Reach no further than the cousinhood.
Read in order: ← The Frame · Contents · Chapter 1: Idolatry →
Want the philosophy first, in voice? The Frame (Part 0) sets the stage this page runs the machine for.