Content Quality: Clear News article of 1,178 body words by the pipeline's whitespace count (News range 400-1,200; tables and headings included). Separates the benchmark design (README, results config), Lawson's reported results (cost, time, lines of code, Lighthouse), his own caveats, and the press coverage; the Herald's one computation is labelled as its own. Keeps the qualifiers: one small app, one model/effort/agent configuration, five runs per stack, API-list-price cost (not money spent).
Source Verification: All 5 snapshots read from disk (sources/2026-11/a-benchmark-has-claude-code-build-one-small-pokmon-app-in-12-javascript-stacks-with-median-costs-from-267-hono-and-htmx-to-497-angular/source-0..4.html.gz); all HTTP 200, no archive fallback, no suspicious_patterns; decompressed sha256 re-computed and equal to the manifest (0 d8eb9ab2..., 1 3fc8a026..., 2 b51ec82d..., 3 37dc9ef3..., 4 be94b969...). Text extracted from the raw HTML, no summaries. The two GitHub sources were additionally read as raw files (raw.githubusercontent.com with a browser User-Agent): results config.json sha256 ff8640d5ff3021415998675ddb29a8a2d87450aaeefa22c928c4c83d00844ca7 (model claude-opus-5-5, effort high, claudeVersion 2.1.310, iterations 5, frameworks list of 12) and README.md sha256 8acfa3d7632479f48c732c013f52ca3bf3cc570e646f9563aeadeb1f6fa87b17. [0] nolanlawson.com post (Oct 28, 2026): every quoted phrase found verbatim ('represents only one holotype', 'a relatively content-based site with little interactivity', 'Opus 5.5 with high effort', 'this quotes the API prices directly from Claude Code and doesn't actually represent how much money I spent', 'nearly saturated', 'highly variable', 'everything I'm measuring is very stochastic', 'biased against SPAs', 'end-all-be-all benchmark', 'mostly in the interest of time', 'largely vibe-coded', 'not trying to vilify any particular framework', 'read the SvelteKit v3 migration guide in every run, likely because it's too new to be in Opus 5.5's training data', 'For Nuxt, the MCP was never used by the agents'). All 12 rows of the article's table recomputed against the post's cost, time and lines-of-code tables: all 36 figures match (cost 2.67/2.76/2.82/2.98/3.14/3.27/3.54/3.79/3.81/4.28/4.65/4.97; minutes 8.6/8.2/9.0/10.1/10.9/10.7/10.3/11.2/12.5/10.8/12.1/17.2; lines 1370/1461/1415/1492/1539/1744/1697/1630/1564/1611/1717/2009). Cost and time are medians of 5 runs for Angular too; lines of code and Lighthouse use 3 Angular apps, as the article says. React Router cheapest run $2.82 vs Hono+htmx most expensive $3.19 (post ranges 2.82-3.97 and 2.23-3.19) correct; '~70% of the code' correct (1370/2009 = 68%). Lighthouse: Performance 97 (Vanilla) to 100, Accessibility 100 in all 12 rows, Astro 0 KiB / 810 ms, Next.js 142 KiB / 2,090 ms, TanStack Start 2,028 ms, Next.js 1 of 5 skill files, Svelte plugin MCP 5/5, versions Astro 7.4.2, SvelteKit 3.1.0, @angular/core 22.2.3: all match; Slow 4G, 4x CPU throttling and mobile-only home-page scores match. Angular: post says the 2 failed runs were an SSR app that rejected non-localhost and were not re-run, and carries an Update note that an earlier version overstated Angular's inability to build a working app: matches the article. [1] The Register, 3 Nov 2026 14:51 UTC, byline Joab Jackson: headline, subtitle ('Sorry React die-hards, Claude prefers lean HTML-first tools over bloated SPAs'), 'a set of instructions for a Claude Code agent', 'agents best utilize those frameworks that favor convention, API stability, and concision', $2.67 Hono+htmx, 'failing to even finish a working app in two of its five attempts', SvelteKit 'repeatedly going back to the SvelteKit documentation to get a better handle on how the recently-updated Svelte works', 'a few developers on Hacker News': all verbatim or faithfully paraphrased. [2] DevClass, 4 Nov 2026 14:53 UTC, same byline (Joab Jackson), same headline and subtitle; the article text diffed against The Register's is identical apart from ads, site chrome and tags, so The Register and DevClass are one account, as the article states. [3] README: the quotes ('the same spec (a Pokedex backed by a local copy of PokeAPI), the same tests, the same tools and the same one-line prompt', 'The only difference is the framework it's told to use', 'build-time fetching and static generation of data pages aren't allowed', 'is content-heavy, which favors server-rendered MPAs on first-load metrics'), the prompt template, the install-up-front tooling list and the 'None' rows (TanStack Start; SolidStart, Qwik, Gracile + Lit, vanilla), and 'Only page loads are measured' are all present. [4] config.json: model, effort, Claude Code version and five iterations as the article states; the file holds no cost figures, and the article does not claim it does (the same numbers appear on the post, the Register and DevClass). Note for the record: config.json shows the five iterations were assembled from several runs (stitched plus an appended batch of 2); the article does not claim a single run. Absence claims ('None of the cited sources reports a response from the maintainers of any of the 12 stacks'; 'none of the cited sources tests whether the cost ordering holds outside that configuration'; the README lists no tooling for TanStack Start, SolidStart, Qwik, Gracile + Lit, vanilla) checked against all 5 sources: none contradicts them. Links: the SvelteKit 3.0 internal link resolves to src/content/articles/2026-10/02-sveltekit-30-ships-..., dated 2026-10-02T10:27Z, a News article on the SvelteKit 3.0 release, no corrections record; described exactly. Allowlist: theregister.com, devclass.com and github.com were already listed; nolanlawson.com was not and was confirmed and added in this review (GitHub user nolanlawson, owner of both repositories, lists https://nolanlawson.com as its website and Socket as company, about 3,400 followers; the blog footer reads 'I'm Nolan, a programmer from Seattle working at Socket'; the post links the two repositories; The Register names him 'Seattle-based Nolan Lawson, currently a developer at AI security firm Socket' and links the post). It is a recognised developer's personal site, usable for his own work, not an editorial outlet; the allowlist comment says so. Anthropic disclosure: required - Claude Code and Opus 5.5 (Anthropic products) are the benchmarked agent and Claude Code is named in the headline, summary and lead; the Overview carries the one-sentence disclosure ('The Machine Herald's contributor model is built by Anthropic; the benchmark described here used Claude Code and Opus 5.5 as its agent.'), accurate and neutral.
Factual Accuracy: Headline, summary and lead hold: Claude Code with Opus 5.5 (high effort), 12 JavaScript stacks, one small Pokemon app, medians of $2.67 (Hono + htmx) to $4.97 (Angular) at API prices, post dated October 28. The headline keeps 'Median Costs' and 'One Small Pokemon App'; the 'API prices' and five-run qualifiers are in the summary and lead. Derived figure: the Herald's 'about 1.9 times' (4.97 / 2.67 = 1.86) is labelled 'By the Herald's own calculation'. Disagreements: The Register ('failing to even finish a working app in two of its five attempts') versus Lawson (two runs built an app that rejected non-localhost requests, counted as a spec failure, with a note that an earlier version overstated the Angular failure) are both reported with attribution and left unreconciled, as is the SvelteKit cause (documentation versus v3 migration guide; the two accounts are not strictly contradictory and the article does not claim they are). Dates in order and as calendar dates: Oct 28 (post), Nov 3 (Register, 14:51 UTC), Nov 4 (DevClass, 14:53 UTC), Oct 2 (Herald); no relative dates. Quantifiers: 'every row' (Accessibility 100) true for all 12 rows; 'in every run' is a quote; '1 of 5 / 5 of 5' counts match the post. Attribution: all Claude Code results are attributed to Lawson ('Lawson reports', 'Results as Lawson reports them'); the Register's 'Claude prefers lean HTML-first tools' is given as the Register's subtitle and set beside Lawson's 'not trying to vilify any particular framework'; the article draws no conclusion of its own about which framework is better.