Audit & release notes

A design system isn't done at launch — it's done when it stays right under pressure. This page is the public artifact of every maintenance pass since v1.0 shipped — each finding, decision, and fix, in the open. Newest first.

Why this page exists

Most of the systems I work on live behind NDAs. I can't show the dashboards, the monitoring panels, or the operator tools that pay the bills. Tiny Wire is the public-safe twin — same constraints, same density, same warm operational tone — built so the process is visible without leaking a single line of client work.

This audit is the proof. It shows what shipping a v1.0 and then maintaining it actually looks like: catching the regressions, defending the decisions, and writing down enough context that the next person (or the next quarter's me) doesn't have to relitigate them.

v1.6 — Receipts

2026-09-20 · documentation and record pass, zero breaking changes.

Every earlier release audited what the CSS does. This one audits what the documents say. A record layer built alongside the system — a page per consumer, per promotion candidate, per sameness call — got checked against the four consumer repos by script, and the script disagreed with the documents eight times. What follows is mostly claims that were true when written and quietly stopped being true.

One thing shipped in lib/: the stat tile, as Card parts (.card-value, .card-delta), because three consumers and both docs pages had each built it from inline styles. Everything else is the paperwork catching up — and then being wired so it cannot fall behind again. The counts on the README, the introduction, the components page, this page's tiles, and the hub's tables are rendered from the source files now. Typing one is overwritten.

v1.5 — Reserve green

2026-06-14 · semantic-color pass, zero breaking changes (back-compat aliases kept).

Building a conceptual attestation prototype on the system surfaced the finding the manual eye had normalized: green was doing double duty — the brand accent and the semantic "healthy / attested / safe" signal. When the brand mark, primary buttons, focus rings, and the "this is safe" chip all share one green, the trust signal stops popping. v1.5 reserves green strictly for a new --success family (attested / approved / safe / healthy / running) and moves brand + all interaction onto the existing cobalt — so the rarer green now reads as a real verdict. Token names are unchanged and the old .alert-brand / .banner-brand / .tag-brand remain as aliases, so the prototype inherited the entire shift through the semantic tokens — no per-screen edits. A free win fell out of it: moving brand off green closed the longest-deferred a11y finding (brand-on-white text, 3.54:1 → cobalt 8.3:1).

v1.4 — Live demos

2026-06-13 · docs-experience pass, no token or class changes.

A structured UX critique (FLOWIE) found the docs were full of false affordances — demos that looked interactive but did nothing — plus wayfinding gaps and a scroll order that didn't track the nav. All fixed without touching the CSS-only system: the docs gained a thin vanilla-JS layer so every demo is honest (tabs, accordion, popover, command search, dialog/sheet now work), code blocks copy in one click, long pages get an auto table of contents + back-to-top, Foundations gets token search, and every page's scroll order now equals its sidebar order. The orphaned sidebar-rail page and the dead "view diff" links were fixed too. Full list in the changelog.

v1.3 — Pedigree pass

2026-06-12 · additive, zero breaking changes.

A systematic rigor + signature pass. The neutrals consolidated into one coherent warm-neutral ramp; a single cobalt accent replaced the two overlapping blues and now owns focus; the warning brown became a disciplined amber; and the whole set was restructured into a three-tier primitive → semantic → component architecture — values only, every public token name preserved, so downstream prototypes inherit it unbroken. Type moved to a 1.2 modular display scale with em tracking and a mandatory tabular-numerals token; the default radius sharpened 10→8; floating overlays dropped their solid border for a hairline and let elevation do the lifting. Four issues surfaced and were fixed in the same pass:

v1.2 — Sidebar rail

2026-06-11 · additive, zero breaking changes.

Added .sidebar--rail — the collapsed, icon-only form of the app sidebar (56px, driven by --sidebar-w). It reuses every existing .sidebar* rule and token: labels hide, icons center and grow to 20px, the active state is unchanged. See the Sidebar component and the rail demo; the full entry lives in the changelog. Scope was the rail only — the deferred a11y findings below remain queued for a later release.

v1.1 — Findings

Twelve issues, grouped by theme. Click a filter to scope the list. Every fixed finding links to the line that changed.

The live checker's five findings from the same release, F-013 – F-017, are on the A11y check page and keep those IDs. The audit's later findings start at F-018 — one ID space across both pages.

How I audit

Three passes, three goals. I run this same loop on every system I maintain, internal or public.

Decisions

A design system is as much about what you refuse to add as what's in. These came up during audits and were decided; each says when. The findings above are the checked part of this page — this section is the judgment part.

What's next

Not a roadmap. The candidates table in the Design System Hub is what's next — every row is something a real app built, counted, and sorted by how many apps built it. Deferred findings above keep their IDs until they close.

Every release ships with a page like this. The point isn't the changelog — every repo has one of those. The point is to make the reasoning visible: what changed, what didn't, and what I refused to add.