Methodology

How Dracon knows what it says it knows. Capture-dated data makes the freshness claim real; the selection gate decides which vendors get a row; the AI-assistance pipeline shapes the editorial layer. This page is the trust surface for those three.

Capture-dated data

Every number on the tools surface says when it was captured. That claim is machine-kept, not marketing: pricing is captured from vendor pricing pages on a known date, and the benchmark numbers come from the studio's own warp runs against real test buckets — primary data, not vendor self-reports.

The rule is simple: no number claims freshness it does not have. A cell without a measurement is printed as such (the page says "seed placeholder, real capture pending") rather than dressed up with a fake date. The raw data files are public in the repo (web/tools/src/lib/data/), one catalog per surface, overwritten by the capture scripts when a new run lands.

The selection gate

The tools are not a coupon dump. The selection gate is one rule: would the studio spend its own money on this? The studio runs this site, its games, its music, and its AI Hub on real infrastructure, and the storage compare lists the S3-compatible object-storage vendors it actually uses or could plausibly use — Bunny, Cloudflare R2, Backblaze B2. If a vendor is not plausible first-party infrastructure, it does not get a row.

That rule kills the link-dump problem before it starts. There are no affiliate rows and no tracked merchant links anywhere on the platform — the full list of refusals lives on the how we make money page.

The AI-assistance pipeline

Editorial pages are drafted with AI assistance. The pipeline is three layers, each with a distinct source of truth:

1. AI-assisted draft. The studio uses AI to draft the prose framing, the comparison tables, and the FAQ Q/A pairs. The AI is not the source of truth for any number — every price, capture date, and benchmark figure comes from primary capture, not from the AI.
2. Primary-data capture. The captured catalogs are the source of truth for any number on the page. Price lines, capture-date stamps, and benchmark rows are all rendered from the captured data files, overwritten by the capture scripts.
3. Human editorial review. A human in the studio reviews the AI draft against the captured data, fixes any framing that drifts from it, and signs off before the page is published. The page carries a capture-date stamp and an AI-assistance disclosure caption.

The disclosure line at the bottom of every editorial page reads Drafted with AI assistance · Capture-dated primary data · Reviewed by the Dracon team. The line is the same across every editorial page because the pipeline is the same — a different disclosure line per page would be evidence of a different pipeline, not a stylistic choice.

Frequently asked, and the honest answer

How current are the prices on the storage compare?

The Object Storage Compare shows a capture date next to the numbers. Prices are captured from vendor pricing pages by the studio on that date, and the capture script is re-run when a vendor changes pricing. No number on the tools surface claims freshness it does not have — unmeasured cells say so on the page.

Where do the benchmark numbers come from?

From the studio’s own runs: the capture script drives MinIO’s warp benchmark suite against dedicated test buckets on each provider, from the same Dracon dev VPS and network path, back-to-back. The results are written into the public data catalog with a capture-date stamp. Until real runs land, the page prints a visible note that the cells are seed placeholders — we do not dress placeholders up as measurements.

How do you decide which vendors are listed?

The selection gate is one rule: would the studio spend its own money on this? The studio runs real infrastructure, and the storage compare lists the S3-compatible object-storage vendors the studio actually uses or could plausibly use: Bunny, Cloudflare R2, Backblaze B2. Competitors of our own products are not listed.

How are editorial pages written?

Editorial pages are drafted with AI assistance. The drafting flow is: AI-assisted draft → capture/research against the primary data (pricing pages, benchmark runs) → human editorial review → published with a capture-date stamp and an AI-assistance disclosure caption. The captured data is the source of truth for any number; the editorial layer is the source of truth for the prose framing and the ranking.

What does "capture-dated" mean on a row?

It means the number is a primary-data capture: we recorded it on the date shown, from the source named on the page, and the raw data file is public in the repo (web/tools/src/lib/data/). A row without a capture date is marked as such rather than given a fake date.

Why disclose AI assistance?

Because the 2026 search/AI-overview landscape treats undisclosed AI-assisted content as a quality signal, not a "human-only" signal. The honest position is the AI-assisted drafting + primary-data capture + human review pipeline. Hiding it would be the unverified claim this site is built to refuse.