dsh-web-search-ddg

LIMITATIONS

Known limitations

These limits define when the provider is a poor fit. - **Anomaly detection is signature-based** — a 2xx that carries no result markup counts as throttling and trips the failure budget: the HTTP 202 anomaly, and any status answered by a page with no `#links` container at all (live-verified: every served page — organic results, fuzzy matches, the true no-results page — carries the container; only the challenge does not). A challenge served over HTTP 200 that still carries `#links` but no rows reads as a genuinely empty query. - **No retry inside the provider** — a single transient failure surfaces to the caller as a structured error; only repetition (the budget) makes the provider step aside. Deciding whether to retry one failed query is the caller's job. - **Scraped markup is brittle** — drift that removes the `#links` container now raises a transient failure instead of a silent empty result, but drift that keeps the container and only renames row classes still parses to zero sources and reads like a poor query. - **A row with a blank title or unusable URL is dropped** — there is no portable value to map, so fewer sources than requested can return. - **`truncated` is never `true`** — the provider caps its usable output at `request.maxResults` (or the configured default), so the seam never sees an over-return and never flips the flag. A caller cannot tell that a page held more results than it received; request more rows if you need to know. - **No `publishedAt`, no `content`** — the static markup carries neither, so publication-date filtering and provider answers are unavailable; the keyed backends do expose them. - **Query shaping is not exposed** — region, safesearch, time and type filters, and paging have no provider-neutral service fields to hang on yet; the config surface covers `endpoint`, `maxResults`, and the failure-budget knobs. - **Abort classification is error-shape-based** — only a `DOMException` named `AbortError`, or an already-aborted signal, maps to `WEB_ABORTED`; an abort carrying a custom reason surfaces as `WEB_PROVIDER_ERROR`. - **Unauthenticated scraping carries terms-of-service risk** — high-volume deployments should use a licensed or first-party backend.