Deduplicated Fallback Query Uses Wrong Relevance Threshold

Deduplicated fallback search queries can accidentally trigger the wrong relevance threshold and reject valid results when the requested query matches the publisher fallback.

Quick solution: Identify the fallback query by its configured value rather than checking inequality with the requested query. Here is a minimal implementation pattern in TypeScript:

function resolveSearchPolicy(requestedQuery: string, fallbackQuery: string, currentQuery: string) {
    const isFallback = currentQuery === fallbackQuery;
    const threshold = isFallback ? RELAXED_THRESHOLD : STRICT_THRESHOLD;
    return { isFallback, threshold };
}

Prerequisites: Ensure your search client preserves structured retrieval counters and logs whether a request originated from a primary or fallback path. For further background on parsing specifications accurately during search logic refactors, consult the CommonMark Spec.

The Root Cause in Query Deduplication

When a source requests the same workstation query used by the publisher fallback, search deduplication properly avoids a duplicate HTTP request. However, fallback detection previously compared the incoming query against the original source query. Because they matched, the single search executed under the stricter topic threshold instead of the relaxed fallback rule. This misclassification caused the search client to reject every unused candidate despite successful API responses.

When building reliable background workers or handling edge cases like Automating Medium Draft Workflows, strict deduplication without proper state preservation can inadvertently drop fallback paths. When evaluating similar performance limits in edge environments, review platform guidelines such as the Cloudflare Workers Limits.

Implementing the Repair

The fix updates fallback identification to rely on its explicitly configured query string rather than checking whether the requested query differs from the fallback query. This design maintains search deduplication efficiencies, preserves the two-download ceiling, enforces raster validation, checks URL and hash uniqueness, and respects the intended fallback relevance rule.

Verification and Limits

A local test run covering 247 checks, typechecking, coverage thresholds, a Wrangler deploy dry-run, and a no-write local Workflow passed successfully. A dedicated regression test now verifies that exactly one search and one selected fallback result occur when the requested and fallback queries match. A live API check confirmed 26 relevant unused candidates among 80 total results, successfully selecting a valid JPEG image that decoded cleanly and was absent from prior URL and hash history. Note that remote release checks remain pending and the underlying image pool is finite.

Verification Checklist

• Confirm that the current search query is evaluated against the configured fallback string rather than the raw source input.
• Verify that structured retrieval counters increment correctly for both primary and fallback paths.
• Run local regression tests to ensure deduplicated requests apply the relaxed relevance threshold.
• Check that raster validation, URL uniqueness, and hash checks continue to function without regression.