2026-08-07 — Truncation announces its size

Built (2026-08-07) — item (3) from the in-situ wishlist, the interim form. The value renderer's caps stay as they are; what changed is the silence: a cut rendering now carries a footer line — ;; truncated — 9872 elided · (res 8) is the whole value — produced by a mirror of the printer's own walk (counted collections counted exactly for free, an uncounted seq probed one already-realized cell past the cap, below-*print-level* subtrees reported hidden, not sized). The inline ... n more rendering the wishlist sketched was considered and declined: it needs a custom bounded printer, which is precisely the machinery §6's shape will retire the cap with — the footer line joins the eviction announcement instead, same doctrine, and nothing gets thrown away when shape lands. Verified live through the MCP connection: (vec (range 10000)) → "9872 elided" (the wishlist's own example number), (take 200 (range)) → "uncounted elision" with (count @(__r/res 9)) → 200 behind it. One trap re-confirmed while testing: an eval returning an infinite seq never settles — deep grounding walks the return value — so the infinite case cannot reach the renderer at all; the honest uncounted case is a finite lazy seq. Suite green: 72 tests, 258 assertions.

The trap got its teaching line the same day: the manual's grounding paragraph and the eval description now both say it — grounding walks the whole of each form's value, so a returned infinite seq never settles; work with them freely ((def nat-ints (range)) returns the var, (take 5 nat-ints) a bounded value), return them bounded. Verified live before writing: the def-and-take shape settles cleanly through the connection.