Skip to content

Risk register

Risk: APIM spans many tiers, gateways, protocols, portals, integrations, and fast-moving AI features.

Response: generated inventories, dated parity snapshots, stable/preview separation, phased releases, and no undated blanket parity claim.

Risk: specs describe shapes better than policy ordering, errors, routing, timeouts, and edge behavior.

Response: authorized Azure differential harness, backend recorder, trace capture, regional/tier metadata, and permanent fixtures for every discovery.

Risk: parsing is easier than reproducing .NET conversions, exceptions, collections, regex, JSON/XML, crypto, and date behavior.

Response: project-owned typed AST, member-level inventory, golden vectors executed in .NET and Azure, differential fuzzing, explicit unsupported nodes, and an optional sandboxed reference worker for diagnosis.

Risk: generated clients may hard-code ARM hosts, audiences, API versions, or polling behavior differently by language/version.

Response: P0 four-SDK spike before broad implementation, pinned versions, documented endpoint injection, and older API projections rather than query-version aliasing.

Risk: buffering, policy caches, portal assets, analytics, GraphQL, gRPC, and many services can erase Go’s memory advantage.

Response: streaming by default, bounded body stores, immutable snapshot sharing, LRU caches, retention limits, heap benchmarks, and release memory budgets.

Risk: management writes race with requests, rate limits, quota, cache, portal publishing, and distributed gateway configuration.

Response: transactional writes, compile-before-publish, immutable snapshots, atomic counters/SQLite transactions, revisioned config, race tests, and failure injection.

Risk: XML, WSDL, GraphQL, protobuf, OpenAPI, C# expressions, portal content, and linked imports expose denial-of-service and injection paths.

Response: depth/size/time limits, disabled external XML entities, controlled fetcher, CSP/sanitization, fuzzing, loopback default, secret redaction, and helper isolation.

Risk: the open-source portal evolves independently and contains a large UI, content format, publishing workflow, and authentication surface.

Response: pin upstream commits, separate content/runtime contracts from UI, test self-hosted interoperability, use Playwright visual/workflow fixtures, and review license implications before code reuse.

Risk: Premium/v2, workspaces, regions, networking, and AI integrations cost money or require scarce permissions.

Response: scheduled budgeted suites, reusable fixtures, sanitized goldens, tier-specific test subscriptions, cleanup audits, and blocked-external parity state.

Risk: accidental use of non-public Microsoft implementation details would undermine the project.

Response: grounding manifest, public-source citations, fixture provenance, contributor rules, no leaked/internal artifacts, and review of every behavioral claim that is not directly documented.

R11 - Local substitutions overclaim fidelity

Section titled “R11 - Local substitutions overclaim fidelity”

Risk: network, scale, region, monitoring, and availability simulations could be mistaken for real infrastructure guarantees.

Response: distinguish wire/state parity from physical implementation, document substitutions, expose local topology, and never market the emulator as a security, performance, resilience, or billing simulator.