TileDB

PASS consensus from 3 PASS.

The supplied URL returned an app shell; its live Academy mirror provides installation and the complete dense-array flow. A single page carries the whole path, install via conda install -c conda-forge tiledb-py or pip install tiledb, then Dim/Domain/Attr/ArraySchema with sparse=False, tiledb.Array.create(), A[:] = data under tiledb.open(..., "w"), and A[:]["a"] to read back, with no step left to inference, though only after the agent discovers that the canonical academy host serves none of it. The Arrays Quickstart documents pip and conda installation, then walks through dimensions, domain, attribute, schema, array creation, write, and read in a single unambiguous code path.

FAIL consensus from 1 PARTIAL, 2 FAIL.

It documents $100 for six months but omits storage rates and API quotas; the pricing link shows deployment pricing. Free-tier credits, validity and expiry are stated precisely ($100 for 6 months, remaining credits retracted on downgrade), but the Pricing page this sentence defers to contains no consumption charge, no usage calculator and no occurrence of 'credit', 'slicing' or 'egress', it prices seats and vCPUs instead, while a sweep of all 423 sitemap pages returned zero hits for 'rate limit', 'quota', 'per GB' or 'per TB'. Free-tier credits ($100, 6 months) and expiry are clearly documented, but the per-unit compute and storage rates promised on the credits page are not published on the pricing page, which only shows annual organization pricing. No API rate limits are documented anywhere.

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

The error page catches and prints TileDBError, but it provides no correction or retry sequence. The tutorial catches tiledb.libtiledb.TileDBError and prints it but stops there, so correcting and retrying required cross-referencing the Configuration page for rest.retry_http_codes/rest.retry_count and the API Usage page for .retry(), and no page in the corpus documents an error code list or the word 'diagnostic'. Configuration-based retry for HTTP 429 is documented (add 429 to rest.retry_http_codes, configure retry_count, delay, and factor). However, no documentation shows the runtime pattern of catching tiledb.TileDBError, extracting diagnostics, correcting the operation, and retrying it. Only preventive configuration is covered.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The token page omits token-copy instructions, while separate tutorials assume TILEDB_REST_TOKEN already contains the generated value. Token creation and expiry are documented cleanly under Accounts, but the Arrays Cloud tutorial runs only inside TileDB Cloud and reads an auto-injected TILEDB_REST_TOKEN, so the external-client path had to be recovered from the Tables section, which authenticates with a different call (tiledb.cloud.login(token=token)) than the Arrays page (tiledb.Config({"rest.token": ...}) plus tiledb.default_ctx). The Get Started page links to the API tokens page (UI creation of scoped, expiring tokens) and the Basic TileDB Cloud tutorial shows end-to-end configuration: tiledb.Config with rest.token, tiledb.default_ctx, and tiledb:// URI access.

PASS consensus from 3 PASS.

The stable API documents Array.create and tiledb.open, which match the methods and modes used by the current quickstart. The API Reference page links TileDB-Inc/TileDB-Py to a reference built from TileDB-Py 0.36 against a PyPI latest of 0.36.1, carrying a runnable minimum example under from_numpy and full entries for every symbol the quickstart uses (Dim, Domain, Attr, ArraySchema, Array.create, open), with no contradiction between the two. The API Reference page links to the official TileDB-Py docs and the Arrays Quickstart provides a complete dense-array example that matches the pattern shown in the Basic Dense tutorial.