Evidence-first tooling for reliable web retrieval.
PN Labs builds small, auditable tools for scraper, crawler, browser-automation, and web-retrieval systems where reliability matters more than raw proxy count.
Our working rule is simple:
Observe first. Attribute carefully. Measure usable outcomes.
Deterministic HTTP/proxy outcome classification that avoids turning every 403, 429, 451, timeout, or transport failure into an automatic “bad proxy” conclusion.
- evidence-first attribution;
- proxy-path vs endpoint-health separation;
- local-only, zero runtime dependencies;
- privacy-conscious public inputs and outputs.
Local benchmarking for retrieval policies using operator-facing metrics such as:
- usable success rate;
- requests per usable result;
- rotations per usable result;
- latency distribution;
- cost per usable result.
The goal is not to prove that “more proxies” are better. The goal is to measure whether a policy actually produces better usable outcomes.
- Evidence before attribution — an HTTP status is an observation, not automatically a root cause.
- Usable-result first — success is defined by the workload, not by proxy count or request count.
- Fail closed under ambiguity — uncertain evidence stays uncertain instead of becoming a confident health signal.
- Data minimization — public tooling should not require credentials, private endpoints, production captures, or customer data.
- Small, testable components — classification, measurement, and control policy stay separable.
We are exploring integrations and design-partner benchmarks for legitimate, authorized web-retrieval workloads with measurable reliability problems: unstable proxy pools, blind rotation, mixed target/proxy failure attribution, and high requests-per-usable-result.
If you have a reproducible workload or failure pattern that fits that scope, open a sanitized issue in the relevant repository. Please do not post credentials, private infrastructure details, raw production logs, or sensitive customer data.
PN Labs · reliability tooling for web retrieval