Reference corpora for agents that answer from evidence
A public registry of pinned, licenced, measured corpora — what an agent should retrieve against when the answer has to be current rather than memorised. Every entry names its domain, its sources and pins, its licence verdict, and what the retrieval tests actually scored.
bench/corpus-tasksan agent working on read-mostly ANN queries this for the forest-of-trees build and on-disk format
an agent working on columnar memory queries this for the Rust arrow arrays, kernels and IPC paths
an agent working on pure-Go KV stores queries this for its LSM-with-value-log design and transaction implementation
an agent implementing a VPN data plane queries this for the wireguard protocol implementation and its async runtime use
an agent working on web content encoding queries this for the static dictionary and transform handling
an agent reviewing C code queries this for version-pinned vulnerable functions and their fix diffs, permissively-licensed projects only
an agent building a TLS-serving reverse proxy queries this for cert automation and config-reload design
an agent doing capability RPC queries this for the serialisation layout, promise pipelining and the security hardening history
an agent triaging an active campaign queries this for the alert's IOCs, required actions and mitigations
an agent triaging findings queries this for whether a CVE is known-exploited in the wild, with ransomware-use and due-date fields
an agent reviewing crypto call sites queries this for real CWE-327/328/347-class advisories showing the misuse and the corrected construction
an agent writing an HTTP client queries this for protocol handling, auth negotiation and the protocol-selection seams
an agent implementing multi-raft queries this for the sharded log and snapshot plumbing
an agent working on query execution queries this for how a real embeddable OLAP engine does join ordering, expression rewrites and memory-aware operators
an agent answering consumer-credit or mortgage questions queries this for what Regulation Z / TILA actually says — higher-priced APR triggers, ability-to-repay/QM, rescission windo…
an agent answering federal tax questions queries this for what the Treasury regulations actually say — de minimis safe harbor, estimated-tax safe harbor, home-office exclusive use …
an agent answering workplace/employment questions queries this for what 29 CFR actually says — OSHA guardrail height, FMLA 12-month/1250-hour eligibility, FLSA overtime, youth-empl…
an agent building a JS-adjacent tool queries this for the spec's abstract operations and completion-record semantics
an agent building a data plane queries this for filter-chain, xDS and overload-manager internals
an agent operating a distributed system queries this for etcd's failure/restore/compaction semantics and tuning guides
an agent working on coordination stores queries this for raft integration, watch semantics and backend compaction
an agent working on ANN queries this for index construction, IVF/HNSW composition and GPU paths in the reference implementation
an agent prioritising remediation queries this for a CVE's EPSS score and percentile to rank patch order
an agent designing zero-copy serialisation queries this for the vtable layout and verifier approach
an agent administering a BSD system queries this for the handbook's exact rc/ZFS/jail/networking procedures
an agent auditing dependencies across ecosystems queries this for GitHub-reviewed advisories with CVSS and patched ranges
an agent building a GitHub integration queries this for the machine-readable schema of a huge production REST API (the exemplar OpenGraph/OpenAPI corpus)
an agent on call for a SaaS queries this for real alert definitions, escalations and mitigation playbooks from a large production fleet
an agent reviewing Go code queries this for vulnerable symbols with fix diffs (extends go-vulndb's symbols into code records)
an agent reviewing Go modules queries this for Go-team-curated advisories with symbols and fixed versions
an agent reviewing code queries this for the review standard's concrete what-to-look-for and clangers
an agent building accessible UI queries this for the component's exact accessibility criteria and edge cases
an agent building a GraphQL server queries this for execution, defer/stream and introspection semantics
an agent building lakehouse tables queries this for spec semantics and the reference metadata/manifest handling
an agent implementing or debugging a protocol queries this for what the RFC actually says at clause level (TLP permits unmodified redistribution)
an agent working on time series queries this for TSM/flux paths and downsampling semantics
an agent writing validators queries this for the dialect's exact keyword semantics (2019-09 vs 2020-12 differences bite)
an agent implementing a Kafka client queries this for the protocol guide's exact request/response schemas and error codes
an agent working on log systems queries this for the log-cleaner, controller and replication protocol internals
an agent learning LSM internals queries this for the original clean implementation (memtable/sstable/manifest split)
an agent building an event loop queries this for the polling primitives and async threading patterns
an agent working on memory-mapped B+trees queries this for LMDB's page/cursor/COW mechanics
an agent needing fast block compression queries this for the block format and decoder loop
an agent writing or reviewing detections queries this for technique ids, tactics, detection data-sources and procedure examples
an agent reasoning about attack surface queries this for attack-pattern enumerations with prerequisites and mitigations
an agent classifying a finding queries this for the CWE definition, its parents/children and detection notes
an agent configuring TLS termination queries this for the recommended cipher/protocol sets per server and their security/cost tradeoffs
an agent implementing an MQTT client/broker queries this for packet formats, QoS state machines and keepalive semantics
an agent designing a REST surface queries this for the naming/paging/versioning/error rules with examples
an agent implementing msgpack queries this for the format handling and unpack edge cases
an agent implementing a NATS client queries this for the wire protocol, verbs and semantics text
an agent working on pub/sub queries this for subject-tree routing, jetstream and clustering internals in readable Go
an agent working with HTTP/2 queries this for flow control, HPACK and priority internals in the reference C library
an agent working on web servers queries this for the event-driven request cycle and module seams (note: very large)
an agent designing a control or wipe procedure queries this for what SP 800-53/61/63/88/207 actually require, clause-level
an agent hardening infrastructure queries this for NSA/CISA joint guidance specifics (MTIG, memory-safe roadmaps, KPMs)
an agent generating or validating an API description queries this for the OpenAPI schema's exact keywords and resolution rules
an agent working on secret stores queries this for the barrier, seal and lease internals of the open vault fork
an agent hardening an OpenBSD host queries this for the pf FAQ's rule grammar, upgrade procedures and faq-page hardening steps
an agent debugging TLS/ASN.1 queries this for the reference provider's handshake, BIO and provider internals
an agent writing host detections queries this for battle-tested osquery SQL by objective (suid diffs, persistence paths, launch items)
an agent pinning dependencies queries this for affected-version truth across Go/PyPI/npm, resolved to one record per vulnerability
an agent wiring telemetry queries this for the exact semantic conventions and proto field semantics
an agent reviewing an application's security posture queries this for the ASVS requirement and its verification level
an agent implementing an auth/upload/crypto feature queries this for the concrete cheatsheet: header lists, code shapes, test ideas
an agent planning a web test pass queries this for test categories, procedures and expected results
an agent reading/writing parquet queries this for encodings, logical types and the thrift metadata spec
an agent working on LSM stores queries this for a modern Go LSM with delete/RangeLock semantics
an agent writing SQL against Postgres queries this for exact function semantics, GUC defaults and MVCC behaviour by version
an agent working on planners queries this for the real cost model, join ordering and executor nodes
an agent designing for failure queries this for how real outages of the same shape started, cascaded and were found
an agent instrumenting a service queries this for metric semantics, recording-rule patterns and the query language's exact behaviour
an agent working on TSDB queries this for the head-block/WAL/head-compaction paths and PromQL evaluation
an agent generating or debugging protobuf code queries this for wire format, descriptor handling and the conformance suite
an agent reviewing Python code queries this for vulnerable functions with fix diffs, PYSEC-linked
an agent reviewing Python dependency trees queries this for PYSEC advisory records with fixed versions
an agent implementing QUIC queries this for congestion control, stream state and 0-RTT handling in idiomatic Rust
an agent working on message brokers queries this for queue/exchange routing and mirrored-queue semantics
an agent implementing raft queries this for the log/ready/progress state machine the tikv fleet runs on
an agent vetting regex handling queries this for real ReDoS advisories with the pathological pattern and the bounded rewrite that fixed it
an agent working on LSM stores queries this for compaction, WAL and recovery internals at production scale
an agent designing a Rust library API queries this for the naming/convertibility/error conventions with their reasoning
an agent reviewing Rust code queries this for vulnerability precedent — identity-resolved RUSTSEC/GHSA/CVE advisories with affected functions and fixed versions
an agent implementing TLS in Rust queries this for the state machine, cipher dispatch and certificate verification
an agent hardening TLS queries this for the defensively-structured C state machine and its testing approach
an agent stepping into an SRE task queries this for the curriculum's concrete networking/Linux/DB ops explanations
an agent needing cheap compression queries this for the format and the reference encoder decisions
an agent depending on SQLite's exact behaviour queries this for datatype, locking and pragma semantics
an agent working on embedded DB internals queries this for the btree module, pager and the famously commented test harness
an agent scripting an unfamiliar long-tail tool queries this for its example invocations (popular tools are memorised — G3 value is the tail only)
an agent containerising or configuring an app queries this for the factor's exact statement and rationale (G3 borderline: well-known prose — keep only if G7 shows retrieval value)
an agent answering UK-law questions queries this for what the current statute actually says — Consumer Rights Act remedies, GDPR-adjacent DPA 2018 duties, Building Safety Act dutie…
an agent formatting dates/numbers/names queries this for locale-specific CLDR rules and plural/data tables no model recalls
an agent building against RESP queries this for command semantics, ranges and error cases (the open post-relicence fork of the Redis docs)
an agent working on in-memory KV queries this for the event-loop, object encoding and replication paths of the open fork
an agent working on layout/selector internals queries this for the csswg draft's algorithm text (the resolution and cascade algorithms are unmemorisable prose)
an agent building browser-adjacent tooling queries this for the living standard's exact normative statements (html/url/fetch/dom)
an agent working on columnar layouts queries this for how clickhouse does codecs, layout and vectorised scans
an agent working on full-text search internals queries this for how lucene/tantivy/quickwit/meilisearch/sonic/elasticsearch do BM25, merge policy and the ES wire
an agent working on storage engines queries this for how sled/fjall/redb do WAL, crash recovery and compaction
an agent working on vector search queries this for how qdrant/usearch/instant-distance/hnswlib do HNSW construction, quantisation and filtered kNN
an agent reviewing an API design queries this for the rule, its intent and propulated exceptions from a fleet of thousands
an agent working on deflate queries this for the modernised algorithms and SIMD paths
an agent tuning compression queries this for frame/block formats, window and level internals