Log — Frontend Architecture Wiki

Append-only history. Entries start ## [YYYY-MM-DD] <op> | <title>.

[2026-08-12] split | Spoke created from the hub _inbox frontend-architecture cluster

Spun out at the curator’s approval, from three sources parked over six weeks under the tag frontend-architecture. Every existing spoke had declined them, and the declines became this spoke’s scope boundary (CLAUDE.md):

• flutter-large-app-structure — parked 2026-06-30 as app-development, re-tagged 2026-07-14. Declined by game-engines (Flame is a Flutter game engine; a general app-architecture guide there is a category error), dev-tooling (Ky precedent — application-level concerns are out), webperf, static-site.
• micro-frontends-migration — parked 2026-07-14. Declined by ui-frameworks (the talk explicitly contrasts micro-frontends with components, so it sits above that spoke’s subject), static-site (React meta-frameworks excluded by charter), platform-ops (“microservices for the frontend” is the analogy, not the domain), dev-tooling, webperf.
• open-saas — parked 2026-08-10 with the cluster tally at 3 and both readings recorded rather than acted on. Declined by static-site (full-stack frameworks excluded since 2026-06-03: “park them unless a better home exists”), dev-tooling, ui-frameworks, foss-applications, agentic-tooling.

The genre objection is on the record and was not dissolved. Two of the three sources are architecture guides that argue a thesis; Open SaaS is a template that embodies one silently, and this same cluster had already excluded a build-along tutorial on exactly that distinction. The curator approved with both readings in front of them. Kept live in synthesis → Contradictions and on boilerplate.

One thing this spin-out resolves elsewhere: ../static-site-wiki has excluded full-stack and React meta-frameworks since 2026-06-03 with the instruction to park them “unless a better home exists.” This is that home. Next.js, Remix, Gatsby and Wasp route here from now on.

All three primaries re-fetched at ingest, per ../HUB.md

A park record is a lead, not a finding. Two corrections came out of it:

• open-saas — the park record (2026-08-10) said 15,217★ / 1,818 forks / 114 open issues. Verified 2026-08-12: 15.3k★ / 1.8k forks / 101 open issues. Stars and forks agree; the issue count moved.
• micro-frontends-migration — the park record did not name the venue. It is QCon San Francisco 2025, and Mezzalira’s stated basis is hundreds of migrations at Amazon and with AWS customers.
• flutter-large-app-structure — published 23 June 2026 (the record carried only a fetch date of 2026-06-30). Confirmed on re-read: the article cites no study, case or measurement, which is why it is tiered T3 on what it rests on rather than on its publisher.
• wasp — fetched as its own primary rather than through the boilerplate: 18.7k★, v0.24.0, in beta by its own statement, targeting React + TanStack Query / Node + Express / Prisma only.

Founding pages (11)

Sources: micro-frontends-migration (T2), open-saas (T1), wasp (T1), flutter-large-app-structure (T3). Things: application-architecture, micro-frontends, feature-first-structure, app-shell, boilerplate. Entities: luca-mezzalira (Person), freecodecamp (EducationalOrganization). [[aws]] and [[infoq]] are linked cross-wiki to their existing nodes in cloud-wiki and platform-ops-wiki rather than duplicated.

The founding thesis

One question — who decides an application’s structure — answered from three positions: the team by convention, the organization by ownership, the framework before anyone arrives. The two architecture sources converge on boundaries beat patterns and on features owning their own state, from different scales and ecosystems, which is the strongest signal this corpus holds.

Two findings recorded against the grain of the sources. The corpus’s headline number (3 deploys a month → 25 a day per team) is credited by its own author to the “safe net” of edge-compute rollback, so it is evidence about reversibility rather than modularity. And the difference between the two scales is enforcement: a micro-frontend boundary is enforced by deployment, a feature boundary by discipline, and no source here compares how well each holds.

Recorded weakness

This domain argues more than it measures. One source reports work done, one states what it ships, one cites nothing. CLAUDE.md carries an edge rule requiring every page to record what its source rests on, and growth edge 1 asks for any measured outcome of a structural choice — the spoke’s thesis is currently unfalsifiable.

Sweep of the rest of _inbox (../HUB.md step 6)

Re-read the remaining 25 parked records against this spoke’s new domain: line, since “no spoke owns this” stops being true the moment a spoke is created. Nothing else moved. The three nearest were checked in full and all three fail on the same distinction the charter draws:

• ky-http-client (js-libraries) — an HTTP client. A library the application uses, not a decision about the application’s structure. This is the precedent CLAUDE.md cites against ../dev-tooling-wiki, and it cuts the same way here.
• ai-test-automation-paradox (ai-productionization) — about the brittleness of DOM-structural test abstractions under AI-generated tests. Adjacent (it is an argument about an abstraction) and its subject is test automation.
• temporal-durable-execution (durable-execution) — backend workflow persistence; ../platform-ops-wiki territory if anywhere.

Note for future readers: both earlier park records cite an adrianhajdin-uber-clone record as the build-along tutorial this cluster excluded on genre. It is no longer in the holding pen — routed or dropped at some point before today — so the exclusion survives only as reasoning quoted in those two records, not as a record you can go and read.