THE CORTEX holds what a fixed window cannot — a stack, a debt, a plane, an identity, a counter — with no learned parameters and no gradient path. Two of the five are genuinely runnable down the centre: the VETO scope-matcher accepts a closer iff it matches the top open scope, and the DRAIN emits exactly the outstanding closers at once. The honesty is the whole game — I-13 keeps a measured DEAD register, so this board shows which rules actually execute and which never fire.
source I-13, THE CORTEX — five rules (rev 2 teaching brief §6; H1.0-FINAL freeze), David Lee Wise / ROOT0 / TriPod LLC. Rendered from the brief’s own register, not quoted.
A fixed attention window is bounded: past a horizon it forgets. THE CORTEX is the keeper-room that holds five pieces of state a window drops — a stack of open scopes, a debt of outstanding closers, a plane already paid for, an identity, and a counter.
It is deterministic and bounded: same input, same state, with no learned parameters and no gradient path. Nothing here is trained — the rules are wired, not fit.
(1) VETO LIT — the only legal closer is the one matching the top open scope, or none.
(2) −I + DRAIN LIT — pressure grows as 1.5·ln(1+age); THE DRAIN emits every outstanding closer at once.
(3) DEPTH mech — refuse a plane already paid for.
(4) IDEMPOTENCE DEAD — I am I; a no-op cannot repeat.
(5) ADDRESS DEAD — the substrate writes it.
The math of all five is real. Which ones the shipped binary actually calls is a separate, measured question — answered right and below.
"Zero parameters" means no learned parameters and no gradient path — it does not mean no constants. The rules carry tuned constants: 1.5 (pressure gain), min(8) (the idempotent clamp), min(12) (the depth cap).
So the claim is honest but narrow: the CORTEX learns nothing and back-propagates nothing — yet three numbers were chosen. That is convention, not measurement, and it stays amber.
Feed a stream of scopes. Openers ( [ { < push; closers ) ] } > ask to close. The question the CORTEX answers, token by token: is this closer the one matching the top open scope?
Anything that is not a bracket is skipped. Edit freely — every verdict below is computed on the spot.
VETO walks the stream, keeping a stack of open scopes. A closer is accepted iff it matches the top — otherwise it is vetoed (refused, stack untouched). This is a stack, not one slot per bracket, so no reachable input silently breaks the claim.
| # | tok | action | stack after |
|---|
THE DRAIN emits exactly the outstanding closers, LIFO, leaving the stack empty. Below: the debt, the −I pressure per open scope (amber — the 1.5 gain is tuned), and the drained closers.
| open scope | age | 1.5·ln(1+age) | closer |
|---|
In the shipped binary only VETO and DRAIN have call sites — LIT for those two. The other three are present in source and never invoked (see right).
IDEMPOTENCE can never fire. In the shipped rule self.last is never assigned, so it compares against a permanent None and the guard always returns true — the de-dup branch is taken unconditionally and the rule performs no work.
ADDRESS is a flat sensor. self.run is never assigned → sensor dim 5 = constant 0. It reads the same forever.
This is I-13’s own measured register. The mechanisms are real; the wiring is not. Kept visible, never laundered.
"All five rules run." Cut. Only VETO + DRAIN execute in the shipped binary; three have no call site — DEAD, from I-13’s own register.
"Zero parameters means no constants." Kept, corrected. It means no learned params + no gradient path; 1.5 / min(8) / min(12) are tuned.
"A single slot per bracket is enough." Cut. Nested scopes need a stack — one slot loses depth and a reachable input breaks the match.
The red move: let VETO accept a non-matching closer. An open scope then closes against the wrong opener and the stack corrupts — the witness (below) turns red.
VETO’s law: only the matching closer, or none. Break it and a mismatched stream reports "clean" while the correct matcher refuses — the attack is real and it is caught.