The entry's own README (the types/ precedent): the experiment in prose, the
transfer mechanism (Diag16 carries all four corners and all four X-squares;
extraction and injection are bit-identical copies; B learns the diagonal at
doubled scale), the seven forced claims with the closed-form curves, why
bit-determinism makes them theorems rather than measurements, the reproduction
commands, and the honest boundary: nothing that writes a bounded list reaches
a binary today (trap.md + capability cells own the gap), so the probes stand
as the machine-checkable spec until list writes lower.

@@ -498,3 +498,3 @@
         ("if_expr",        "chain",   "M16 arms in one `match`",                &c.match_arms,       16,  "4 x M16 - an else-if chain is the branching pressure a `match` carries, and it is a REPETITION rather than recursion"),
         ("match_expr",     "arms",    "M16 arms in one `match`",                &c.match_arms,       64,  "2 x `data_decl`'s declared case limit: a `match` needs one arm per case, plus guarded arms over the same case"),
-        ("list_lit",       "items",   "M20 items in one list literal",          &c.list_items,       32,  "2.3 x M20"),
+        ("list_lit",       "items",   "M20 items in one list literal",          &c.list_items,       128, "2 x M20 - the corpus worst case is ml_othello's full 8x8 board literal (64), one headroom doubling"),

@@ -0,0 +129,4 @@
+
+```sh
+# 1. The app typechecks with ZERO diagnostics (no trap block — the fibonacci shape):
+src/rust/target/debug/ckc check tests/corpus/ml_othello --stdlib tests/corpus/stdlib