v0.8.0

The size guarantee becomes a standing ratchet (ceilings re-baseline to measured x1.25 when they bind; core re-baselined once to 147,500), with honest comparative positioning — a capability matrix in the README, a dated size survey in the gallery, and small consumer-reported fixes (renderStats instance pass-through, checker-limitation and vendoring documentation).

Context

The frozen core ceiling had reached 17 bytes of headroom and was forcing byte-archaeology on every core-file change; the ratchet keeps walls deliberately close (placement decisions stay forced) while ending the archaeology. Positioning was rewritten against a mid-2026 survey of the minimal-framework landscape — the differentiator stated as surface-per-byte plus enforcement tooling, with measured gzip numbers replacing the absolute "tiny" claim.

Breaking

• The size guarantee changed: the core JS byte ceiling is no longer frozen. Every tier ceiling (core included) now follows a STANDING RATCHET: it is set to the tier's current measured weight x1.25, and re-baselines to a new measured x1.25 in review whenever a legitimate addition makes it bind. The old "the core ceiling is never raised" promise is retired; the core ceiling re-baselined once, from 118,000 to 147,500 bytes. Walls are kept deliberately close so adding a capability still forces a real placement decision.

Features

• New README section: Scope and non-goals. Spells out that tinymoon is a fixed-identity, hand-coded, web-only framework and states its explicit non-goals (server-driven UI trees, per-deployment theming, cross-platform native rendering, and app-platform features), so readers can tell at a glance whether it fits their use case.
• New README positioning: a comparative size story and a capability coverage matrix. The docs now frame tinymoon by surface-per-byte rather than absolute smallness: honest measured numbers (raw ~396 KB, gzip ~128 KB for the whole kit), a metric-discipline note (tinymoon's gzip weight vs competitors' min+gzip), two hedged comparisons (about three Bootstraps; roughly 2x a hand-rolled no-build stack), and a durable capability matrix versus view microframeworks, Mithril, Alpine, htmx, Web Awesome, and Bootstrap.
• renderStats now accepts already-built createStat instances, or any mix of instances and config objects, not only config objects. Passing an instance no longer throws "label is required" -- an entry carrying an .el is mounted as-is (preserving references, bindings, and a pre-set trend), and the row's destroy still tears it down.

Fixes

• Documented two conformance-checker details consumers rely on. The external-url rule matches inline URL literals only: a URL passed through a variable (const u = "https://..."; fetch(u)) is neither caught nor allowlistable, so keep external URLs inline at the load site. And vendored framework assets pass by hash-identity in a plain directory with no manifest, but under third_party/ the quarantine's pin requirement applies to them too.