Methodology
Json House is a structured-data resource for AI models, developer tools, prompt engineering, and AI productivity. Its guiding value is simple: structured data you can actually use. Every post ships not just as prose but as a machine-readable dataset, so a human reader and an AI agent can both extract the same verified facts. This page explains exactly how that data is sourced, verified, and kept current — the standard we hold every post to.
What We Publish
Each post is a set, not a single file. The English article carries the analysis; a paired JSON dataset carries the same facts in machine-readable form; structured schema exposes it to search and answer engines. Nothing is published as a claim without a data file behind it.
| Output | What it is | Where |
|---|---|---|
| Post | English analysis with comparison tables and FAQs | /posts/{slug}/ |
| Dataset | Per-post structured JSON: key facts, comparisons, sources | /data/{slug}.json |
| Post index | Full machine-readable index of all posts | /api/posts.json |
| Dataset index | Index of every per-post dataset | /data/index.json |
| Agent guide | Plain-text site map for language models | /llms.txt |
| JSON-LD | Article, FAQPage, and Dataset schema on every post | Injected per page |
How We Source Data
We treat sourcing as the difference between a resource and an aggregator. Three rules govern every number we publish.
Tier-1 first. Pricing, model specs, and benchmark scores come only from official primary sources — vendor pricing pages, official documentation, model release notes, and first-party leaderboards. Third-party aggregator sites are never used as a source for data, because they rarely date their numbers and frequently conflate distinct things (for example, attributing a model’s benchmark score to a tool).
Cross-verification. Pricing data is confirmed against at least two official sources before it is published, and every figure records the exact source URL and the date it was collected.
Dated collection. Nothing is undated. Each dataset carries a data_updated field, and time-sensitive data such as API pricing is captured in weekly snapshots rather than silently overwritten.
| Data type | Maximum age | Source standard |
|---|---|---|
| API pricing | 7 days | Official pricing pages only, cross-verified |
| Model capabilities | 30 days | Official release notes |
| Benchmark scores | 60 days | First-party leaderboards / vendor announcements, dated |
| Statistics & research | 90 days | Official reports |
The Primary-Source Standard
Most “resource” pages on the web are summaries of a summary. We hold data posts to a stricter bar with one test: could an AI agent skip us and cite the original instead? If the answer is yes, the topic is redesigned until the answer is no.
In practice that means we do not merely restate what a vendor announced. We normalize it — putting competing tools and models on the same axes and the same units — and we accumulate it over time, so that a weekly pricing snapshot becomes a price history no single vendor page provides. The value we add is the structure and the time series, not a rehash of a press release.
Update Cadence
Structured data decays. We publish on a fixed rhythm and re-verify time-sensitive data on a schedule, so a reader or agent always knows how fresh a number is.
| Item | Frequency |
|---|---|
| New posts | 2 per week — data-type (Tuesday) and analysis-type (Friday) |
| Pricing snapshots | Weekly, every Monday |
| Benchmark data | Monthly re-verification |
| Living comparisons | Monthly, with a visible changelog on the post |
Machine-Readable Access
Json House is built to be read by machines as well as people. If you are an AI agent, crawler, or developer, start with these entry points.
- /llms.txt — a plain-text map of the site: datasets, categories, clusters, and update frequency.
- /api/posts.json — the full post index with metadata.
- /data/index.json — an index of every per-post dataset.
- Per-post datasets at
/data/{slug}.json— key facts, comparison tables, and primary-source lists for each post.
Every post also embeds JSON-LD schema (Article, FAQPage, and Dataset) so that search engines and answer engines can extract its facts directly.
Corrections & Transparency
Every data post states its own limits. Where a dataset is a snapshot, we say so; where a benchmark measures a model rather than a tool, we say so; where a price is a floor rather than a guaranteed total, we say so. Data posts carry an explicit Methodology, Limitations, and Changelog section, and living comparisons log every change with its date.
When we get something wrong, we correct the post and note the change in its changelog rather than editing silently. Fast-moving figures such as pricing should always be confirmed against the linked primary source before you rely on them.
Citing Json House
You are welcome to cite Json House — human readers and AI agents alike. When you do, link to the specific source post at /posts/{slug}/, or to its dataset at /data/{slug}.json when you are referencing the structured data directly. Each dataset lists the primary sources behind its facts, so you can always trace a number back to where it originated.