# KIP-0032: Proxy Ingress Adversarial Parity Hardening ## Status Implemented in Milestone 26. ## Purpose Milestone 24 defined the concrete local proxy ingress contract. Milestone 25 implemented a deterministic in-memory local proxy ingress prototype. Milestone 26 hardens that prototype before local egress and relay bridge modeling begins. The goal is to prove the local ingress model rejects malformed request event sequences, unsafe descriptor shapes, lifecycle abuse, queue pressure failures, reset/error leakage, mapping collapse, and generated-backend drift. ## Boundary This KIP covers deterministic local ingress hardening only. It does not add local proxy egress, target dialing, sockets, OS proxy configuration, DNS, TLS, VPN, WebSocket, CDN behavior, deployment, or external network behavior. Fixtures and reports store safe classes, counters, buckets, hashes, and conclusions only. ## Adversarial Corpus `internal/localproxyingressadversary` defines the adversarial corpus. Scenario classes include malformed event ordering, missing or duplicate open events, duplicate event IDs, data before open, data after close, reset-before-open, target-error lifecycle abuse, queue overflow, descriptor abuse, fixed mapping controls, cross-request leak controls, and generated-backend drift controls. Every scenario defines expected accept/reject counts, lifecycle class, mapping class, failure bucket, trace-hygiene expectation, and readiness impact. ## Descriptor Abuse Hardening Descriptor abuse cases model endpoint-like, DNS-like, host-header-like, cloud-metadata-like, credential-like, payload-like, oversized, Unicode, control character, and path-like input classes. The validator rejects those classes with redacted error text and keeps accepted descriptors synthetic and bounded. ## Lifecycle Hardening Lifecycle stress checks reject invalid transitions such as direct close from created, accepting after terminal state, reset then data, close then data, target error before descriptor, duplicate reset, and terminal reopen attempts. The lifecycle gate fails if any invalid transition is accepted. ## Queue And Pressure Hardening Pressure checks exercise exact-limit queue usage, queue overflow, per-request event overflow, many small requests, large request classes, reset under pressure, target error under pressure, and collapsed unbounded controls. The pressure gate requires bounded queues, safe rejection of overflow, mapped backpressure, and no payload or secret logging. ## Reset/Error Isolation Reset/error checks verify that reset and target-error events affect only the intended synthetic request. Control scenarios intentionally leak reset or descriptor details and must be detected. ## Mapping Collapse The mapping collapse scanner detects fixed target binding, fixed stream mapping, fixed lifecycle pattern, fixed error/reset buckets, ignored backpressure, invalid targets mapped as valid, padding-only variation, and generated backend mapping drift. Healthy fixture sets pass while collapsed controls fail. ## Generated/Interpreted Parity Generated modules now emit `localproxyingressadv` constants and tests. The parity report compares adversarial classification, accept/reject counts, lifecycle/pressure/reset/error buckets, collapse conclusions, readiness, and trace-hygiene flags across interpreted and generated paths. ## Readiness Report The M27 readiness report checks: - contract compliance - descriptor abuse resistance - lifecycle integrity - queue pressure safety - backpressure mapping - reset/error isolation - mapping diversity - generated/interpreted parity - trace hygiene - fixture stability The passing decision is `go_for_local_proxy_egress_model`. ## Mutants M26 adds mutant identifiers for descriptor abuse acceptance, data-before-open, data-after-close, terminal reopen, unbounded queue growth, ignored backpressure, reset/error cross-request leakage, descriptor leakage, fixed mapping, missed collapse, bad readiness decisions, payload/secret leaks, and generated backend drift. ## Audit Gates The default audit and `kcheck localproxyingressadv` include: - `localproxyingressadv_corpus_validation` - `localproxyingressadv_descriptor_abuse` - `localproxyingressadv_lifecycle_hardening` - `localproxyingressadv_pressure_hardening` - `localproxyingressadv_reset_error_isolation` - `localproxyingressadv_mapping_collapse` - `localproxyingressadv_generated_backend_parity` - `localproxyingressadv_m27_readiness` - `localproxyingressadv_trace_hygiene` - `localproxyingressadv_mutant_detection` - `localproxyingressadv_fixture_drift` ## Commands ```bash go run ./cmd/kcheck localproxyingressadv --quick go run ./cmd/kcheck localproxyingressadv --full --out testdata/audit/localproxyingressadv.json go run ./cmd/kcheck localproxyingressadv generate --out testdata/localproxyingressadversary/adversarial-corpus-golden.json --force go run ./cmd/kcheck localproxyingressadv verify go run ./cmd/kcheck localproxyingressadv compare --old testdata/localproxyingressadversary/adversarial-corpus-golden.json --new testdata/localproxyingressadversary/adversarial-corpus-golden.json go run ./cmd/kcheck --quick --status STATUS.md ``` ## Fixtures Committed fixtures live under `testdata/localproxyingressadversary/`. They include the adversarial corpus, descriptor abuse report, lifecycle report, pressure report, reset/error isolation report, mapping collapse report, parity report, and M27 readiness report. ## Limitations The milestone hardens deterministic local ingress behavior only. It does not prove behavior for concrete proxy clients, operating system proxy APIs, target networks, or deployed relay paths. M27 moves into adaptive path modeling so future runtime work can reason about volatile candidate-path viability before local egress and relay bridge modeling. ## Next Milestone M27: adaptive path model and candidate taxonomy.