docs: ratify v1 architecture and threat model (#2 )

Closes #2
2026-07-16 04:11:24 +02:00
7 changed files with 478 additions and 0 deletions
@@ -76,3 +76,6 @@ The initial service does not provide:
 ## Project status
 Rendezvous is currently in its initial design and bootstrap stage. The first implementation should establish the contracts, directory leases, LiteNetLib mediator, client SDK, thin test client, and a three-party integration test before either game depends on it for production connectivity.
 The ratified v1 boundaries, trust decisions, privacy rules, safety budgets, and
 threat model are indexed in [the architecture documentation](docs/architecture/README.md).
@@ -0,0 +1,113 @@
 # ADR 0001: v1 control-plane boundaries and domain
 - Status: Accepted
 - Date: 2026-07-16
 - Tracking: #2
 ## Context
 Rendezvous must help two game peers discover and attempt an authenticated direct
 connection without becoming a game server, an identity provider, or a gameplay
 traffic service. The HTTP API and UDP mediator share short-lived state and must
 agree on authorization, endpoint freshness, and tenant scope.
 ## Decision
 V1 is one ASP.NET Core deployable with separable directory, join-authorization,
 endpoint-registry, NAT-mediator, and operations modules. Modules communicate
 through application interfaces, not through transport DTOs or one another's
 storage implementation. Contracts and the client SDK remain independently
 packageable.
 The service is a connection control plane. A successful join authorization only
 grants permission to attempt a direct connection. The game host remains the
 final authority for player identity, capacity, bans, admission, and gameplay.
 Rendezvous success is reported only after the host accepts a valid connection
 ticket and LiteNetLib establishes the authenticated peer connection.
 ## Domain glossary
 | Term | Definition | Lifetime and exposure |
 | --- | --- | --- |
 | `SessionListing` | Bounded public discovery data for one hosted game session. | Visible only while its lease and host presence are fresh. Never contains endpoints or credentials. |
 | `Lease` | Renewable capability controlling the lifetime of a listing. | Secret, host-only, expires unless renewed. |
 | `HostPresence` | Authenticated observation of the host's local and public UDP endpoints from its gameplay socket. | Internal, short-lived, never returned by browsing. |
 | `JoinAttempt` | Authorization linking one client attempt to one compatible session. | Internal and short-lived; it is not authoritative game admission. |
 | `PunchCapability` | Opaque, one-time credential scoped to attempt, role, tenant, and expiry. | Sent only to its intended peer; consumed at the UDP mediator. |
 | `ConnectionTicket` | Compact signed credential presented to the host during the direct connection. | One-time, short-lived, and scoped to the attempt and host. |
 IDs are opaque and tenant-scoped. They are never canonical player, entity, or
 world identities.
 ## Trust boundaries
 ```mermaid
 flowchart LR
    Browser["Untrusted browser/client"] -->|"HTTPS: browse/join"| Proxy["Reverse proxy"]
    Host["Game host"] -->|"HTTPS: register/renew"| Proxy
    Operator["Privileged operator"] -->|"separate authenticated route"| Proxy
    Proxy -->|"normalized HTTP + trusted forwarding metadata"| Service["Rendezvous service"]
    Host -->|"host gameplay UDP socket"| Mediator["UDP mediator module"]
    Browser -->|"client gameplay UDP socket"| Mediator
    Mediator <--> Service
    Service -->|"read keys; never list or log values"| Secrets["Secret provider"]
    Service -.->|"future authenticated state protocol"| Store["Future shared store"]
    Service -->|"redacted events and aggregate metrics"| Ops["Observability systems"]
 ```
 - Public HTTP input is hostile even after TLS termination. The proxy may be
  trusted to terminate TLS and supply forwarding metadata only when its source
  address is allowlisted; forwarded headers from other sources are discarded.
 - Public UDP input is hostile even when structurally valid. HTTP-supplied
  endpoints are claims, never proof. Public response targets come only from an
  authenticated UDP packet's observed source. A private local candidate may be
  carried inside that packet only under ADR 0002's bounded same-LAN rules.
 - Operator routes use a separate authentication policy and network exposure.
  Operator access does not bypass tenant scoping, audit, or secret redaction.
 - The client SDK is convenience code in an untrusted process. Server decisions
  never rely on client-side validation or secrecy.
 - Game hosts are authoritative only for their own gameplay admission. A host
  cannot enumerate or mutate another game/environment tenant.
 - The secret provider is trusted with long-lived key material. The application
  receives only the minimum named key version it needs.
 - A future shared store is a distinct authenticated boundary. Moving state to it
  does not make stored input trusted and requires a new availability ADR.
 ## Connection data flow
 ```mermaid
 sequenceDiagram
    participant H as Game host
    participant R as Rendezvous HTTP
    participant M as Rendezvous UDP mediator
    participant C as Game client
    H->>R: Register listing (publisher authorization)
    R-->>H: Lease + host-presence capability
    H->>M: Presence from gameplay UDP socket
    M->>R: Store observed endpoint and freshness
    H->>R: Renew lease
    C->>R: Browse compatible visible listings
    C->>R: Request join attempt
    R-->>C: Client punch capability
    R-->>H: Host attempt/capability via authenticated poll or stream
    H->>M: Host capability from gameplay UDP socket
    C->>M: Client capability from gameplay UDP socket
    M->>M: Validate scope, freshness, expiry, and replay state
    M-->>H: Introduce verified client endpoints + connection ticket
    M-->>C: Introduce verified host endpoints + connection ticket
    C->>H: Direct LiteNetLib connect + ticket
    H->>H: Validate and consume ticket; apply game admission
    H-->>C: Authenticated peer connection or rejection
 ```
 The mediator does not forward normal gameplay packets. A connection attempt
 that times out or is rejected returns a typed outcome to the caller.
 ## Consequences
 - Directory and mediator can ship together without erasing their module boundary.
 - Contracts cannot expose server storage or LiteNetLib implementation types.
 - Tests must cover the three-party host/service/client flow; an HTTP-only test is
  insufficient evidence of a successful connection.
 - Splitting modules into processes requires an explicit protocol, shared-state
  ownership, deterministic mediator routing, and a superseding ADR.
@@ -0,0 +1,88 @@
 # ADR 0002: publisher trust, discovery, compatibility, and fallback
 - Status: Accepted
 - Date: 2026-07-16
 - Tracking: #2
 ## Context
 Dedicated servers can protect provisioned credentials. Public game binaries
 cannot. Discovery also needs rules that prevent accidental cross-game joins and
 make the meaning of a successful authorization precise.
 ## Decision
 ### Publisher trust modes
 Each `GameId` and `EnvironmentId` is provisioned policy, never caller-created
 free text. V1 supports two visibly distinct publisher modes:
 1. **Managed dedicated host.** A provisioned workload principal authenticates
   with a rotatable credential held outside the game binary. It is scoped to
   allowed games, environments, regions, and listing limits. Public or unlisted
   discovery may be enabled by policy.
 2. **Player-hosted session.** A short-lived publisher grant is minted by a
   game-owned backend and is scoped to one game, environment, host, and expiry.
   Rendezvous does not interpret it as player identity. If a game has no grant
   issuer, it may opt into anonymous unlisted hosting with strict address and
   concurrency limits; anonymous sessions can be joined only through an opaque
   share code and never appear in public browsing.
 A reusable credential embedded in a downloadable client is not authentication
 and is rejected as a provisioning design. Responses and metrics expose the
 publisher trust mode so operators and games can apply different policy without
 claiming anonymous hosts are authenticated identities.
 ### Discovery and metadata
 - `Public` listings can appear only in tenant-scoped compatible browsing.
 - `Unlisted` listings never appear in browse results and require a random,
  unguessable share code. Unlisted does not mean private; join authorization and
  host admission still apply.
 - Browser responses contain display data only. They exclude raw endpoints,
  internal IDs, lease credentials, punch capabilities, and connection tickets.
 - Metadata is treated as hostile data. It is schema/budget validated, stored and
  returned as data, and never rendered as markup by the SDK or TestClient.
 ### Compatibility and address families
 - `NetworkProtocolVersion` must match exactly in v1. `BuildVersion` is bounded
  display/diagnostic text and never overrides protocol compatibility.
 - `GameId` and `EnvironmentId` must match exactly. Region is a browse filter and
  preference, not a compatibility escape hatch.
 - IPv4 direct connection and NAT punching are required for v1.
 - Contracts carry an address-family discriminator. IPv6 direct connections may
  use observed global IPv6 endpoints when both peers support them, but IPv6 NAT
  traversal is not a v1 release requirement.
 - Public candidates are derived only from the authenticated UDP packet's source.
  For same-LAN attempts, that packet may additionally claim at most one private
  unicast candidate per supported address family. A local claim is scoped to the
  capability and is introduced only to the opposite role in the same authorized
  attempt after both roles contribute. Loopback, link-local, multicast,
  unspecified, documentation, and otherwise invalid destinations are rejected.
  The SDK bounds probes per introduced candidate and lets callers disable local
  candidates. HTTP-supplied endpoint claims are never introduced.
 ### Authorization and fallback
 Join authorization means only that Rendezvous permits a scoped connection
 attempt. It does not reserve a game slot and does not authenticate a player to
 the game. The host validates and consumes the connection ticket, then applies
 its own capacity, ban, identity, and gameplay rules.
 The SDK returns a typed outcome including success, cancellation, timeout,
 incompatibility, stale host, service rejection, host rejection, and transport
 failure. A game may provision an optional dedicated fallback endpoint. The SDK
 reports it but never connects without an explicit caller decision.
 Gameplay relay is not part of v1. It remains a separate future service whose
 need is evaluated from privacy-safe measured direct-connection failures.
 ## Consequences
 - A player-hosted game needs a game-owned grant issuer for public discovery.
 - Anonymous player hosting is useful for direct invitations but makes no user
  identity claim and receives the strictest quotas.
 - Games remain responsible for presenting unsafe user-authored text safely.
 - The exact-match v1 rule favors predictable interoperation over flexible
  version ranges; a later compatibility scheme must be versioned explicitly.
@@ -0,0 +1,137 @@
 # ADR 0003: state, privacy, availability, and safety budgets
 - Status: Accepted
 - Date: 2026-07-16
 - Tracking: #2
 ## Context
 V1 needs safe defaults before contracts and stores make them difficult to
 change. The initial deployment is deliberately single-active and in-memory, so
 its restart and availability behavior must be honest.
 ## Decision
 ### State and lifecycle
 All directory, lease, presence, attempt, capability, ticket-consumption, and
 rate-limit state is ephemeral and held behind atomic store interfaces. V1 has
 one active writer/service instance. A second instance may be a cold standby but
 must not accept public traffic concurrently.
 ```mermaid
 stateDiagram-v2
    [*] --> Registered: authenticated register
    Registered --> Visible: fresh lease and fresh UDP presence
    Visible --> Registered: presence becomes stale
    Visible --> Visible: lease renew + presence refresh
    Registered --> Expired: lease expires
    Visible --> Expired: lease expires
    Registered --> Revoked: host or operator revokes
    Visible --> Revoked: host or operator revokes
    Expired --> [*]
    Revoked --> [*]
 ```
 Restart loses all ephemeral state, used capabilities, and listings. Readiness is
 false until HTTP, UDP, policy, key material, and the state store are ready. SDK
 publishers use jittered backoff and re-register after a restart; old credentials
 remain invalid. The service drains by refusing new registrations/attempts,
 allowing a bounded completion window, then cancelling remaining work.
 No horizontal scale is supported until shared atomic state and deterministic
 mediator routing exist. A shared-state design is triggered when any of these is
 true:
 - one measured supported node cannot sustain 150% of the 30-day peak load;
 - the approved availability target exceeds what single-active operation can meet;
 - planned maintenance without listing loss becomes a product requirement; or
 - a region needs more than one active mediator endpoint.
 Relay remains independently triggered only when a representative real-network
 canary shows direct-connect failure high enough to justify its privacy, abuse,
 bandwidth, and operating cost.
 ### Initial time and size budgets
 These are enforceable v1 ceilings, not suggestions. Contract issue #4 may lower
 them but must not raise them without security review.
 | Budget | V1 ceiling |
 | --- | --- |
 | HTTP request body | 16 KiB after content decoding; compressed request bodies are rejected in v1 |
 | Listing metadata | 4 KiB encoded JSON, at most 32 keys; key 64 UTF-8 bytes; scalar value 256 UTF-8 bytes; nesting depth 3 |
 | Browser page | 100 listings and 256 KiB encoded response; opaque cursor; stable bounded sort |
 | UDP datagram accepted | 1,200 bytes; oversized or fragmented application payloads are dropped without response |
 | Opaque HTTP credential | 1,024 bytes encoded |
 | UDP capability or ticket | 768 bytes encoded, with the complete datagram still at most 1,200 bytes |
 | Clock skew | 30 seconds maximum when validating issued/not-before/expiry times |
 | Lease lifetime | 60 seconds; renewal accepted from 30 seconds; no client-selected extension |
 | Host presence freshness | 20 seconds |
 | Join attempt lifetime | 30 seconds |
 | Punch capability lifetime | 30 seconds and one successful use per role |
 | Connection ticket lifetime | 20 seconds and one successful host consumption |
 | Graceful drain | 30 seconds maximum |
 All work queues are bounded. Initial per-instance ceilings are 1,024 concurrent
 HTTP requests, 4,096 queued UDP datagrams, and 10,000 active join attempts.
 Overflow is rejected or dropped early with a metric; it never creates an
 unbounded task, allocation, log entry, or retry loop.
 For an endpoint that has not proved possession of a valid capability, the UDP
 mediator sends no response. Once both valid peer contributions exist,
 authenticated mediation sends at most one introduction datagram to each peer.
 The combined response bytes caused by the completing contribution must be no
 more than twice that contribution's bytes, giving zero unverified amplification
 and at most 2.0 verified byte amplification. Protocol padding or a smaller
 response enforces the byte ratio. Responses are sent only to endpoints observed
 from the corresponding authenticated gameplay socket, never to an arbitrary
 HTTP-supplied address.
 ### Supported and capacity profiles
 The development profile is functional, not a production capacity claim. The
 initial production candidate is one Linux instance with 2 vCPU and 2 GiB RAM,
 targeting 25,000 visible listings, 10,000 active attempts, 200 HTTP requests per
 second, and 2,000 UDP datagrams per second while staying below 70% sustained CPU
 and 75% memory. Issue #18 must measure and publish the actual supported profile;
 production is blocked if the target is not met or the documented profile is not
 reduced accordingly.
 The initial single-active service objective, after the real-network canary, is
 99.5% monthly successful availability for valid in-profile requests, excluding
 announced maintenance. In-profile latency objectives are p95 <= 200 ms for HTTP
 and p95 <= 100 ms from the second valid UDP contribution to both introduction
 datagrams. These are service objectives, not guarantees of NAT traversal.
 ### Data classification and retention
 | Data | Classification | Retention and handling |
 | --- | --- | --- |
 | Raw public/local endpoints | Sensitive network data | In memory only while the lease/attempt requires it, then deleted within 10 minutes; never logged or exported as metric labels |
 | Listing display metadata | Public-untrusted or unlisted-untrusted | In memory for the active lease; audit stores only schema/result and a listing ID, not metadata values |
 | Lease/capability/ticket/key material | Secret | Opaque random credentials are retained only as keyed digests; signed credentials retain verification keys and consumption IDs, not issued plaintext; plaintext is returned only at creation and is never logged or traced |
 | Principal and tenant IDs | Internal identifiers | Audit retention 30 days; access-controlled and never used as high-cardinality metric labels |
 | Security/audit event | Confidential operations data | 30 days online, access-controlled; contains action, coarse result, tenant, principal, and correlation ID, but no raw endpoint or secret |
 | Diagnostic attempt record | Sensitive diagnostic data | Disabled by default; when explicitly enabled, redacted record retained at most 24 hours; raw endpoints remain excluded |
 | Aggregate outcome/capacity metrics | Operational aggregate | 13 months; only bounded dimensions such as game, environment, region, trust mode, and typed outcome |
 Logs use allowlisted fields rather than after-the-fact redaction. Correlation IDs
 are random and are not credentials. Error responses are stable and do not reveal
 whether a cross-tenant resource exists.
 ## Owner decisions required before production
 Implementation can proceed with the baseline above. Production remains blocked
 until the owner records:
 - the actual secret-provider and key-custody system for each environment;
 - which games may enable anonymous unlisted player hosting;
 - deployment regions, data-processing jurisdiction, and approval of the stated
  30-day audit/13-month aggregate retention periods;
 - the per-game dedicated fallback endpoint policy;
 - the measured supported profile and whether the 99.5% single-active objective
  is sufficient or shared-state/high-availability work must be brought forward.
 These are configuration and launch decisions, not permission to weaken the
 tenant, replay, endpoint-verification, or secret-handling controls.
@@ -0,0 +1,14 @@
 # Rendezvous architecture decisions
 These records define the v1 architecture baseline. A later change to a ratified
 decision requires a superseding ADR and corresponding contract/test updates.
 - [ADR 0001: v1 control-plane boundaries and domain](0001-v1-control-plane-boundaries.md)
 - [ADR 0002: publisher trust, discovery, compatibility, and fallback](0002-publisher-trust-and-connection-policy.md)
 - [ADR 0003: state, privacy, availability, and safety budgets](0003-state-privacy-availability-and-budgets.md)
 - [Threat model](../security/threat-model.md)
 - [Security promise and test matrix](../security/control-matrix.md)
 These decisions intentionally leave gameplay authority, player identity,
 simulation, persistence, social features, skill matchmaking, and gameplay
 traffic with each game. Relay is future evidence-driven scope, not part of v1.
@@ -0,0 +1,27 @@
 # README security promise and test matrix
 Tracking: #2
 This matrix turns each security and lifecycle promise in the README into an
 enforceable control and planned evidence. Issue numbers refer to the delivery
 backlog where the control is implemented and verified.
 | README promise | Enforceable control | Planned evidence |
 | --- | --- | --- |
 | Per-game credentials and signing keys | Provisioned principals and versioned keys are scoped to game/environment; secrets come from a provider and never a public binary. (#5) | Cross-tenant authorization tests, rotation/overlap/revocation tests, and secret scans. |
 | Short-lived, single-purpose tokens resistant to replay | Issuer fixes audience, tenant, attempt, role, issued/expiry times, nonce, and key ID; store atomically consumes nonce/ticket. (#4, #6, #10) | Golden vectors; expired, future, mutated, wrong-role, wrong-tenant, and concurrent replay tests. |
 | Strict payload, metadata, and token size limits | ADR 0003 ceilings are checked before allocation/deserialization and again at domain construction. (#4, #15) | Boundary/property tests, malformed corpus, and allocation-aware fuzzing. |
 | Registration, query, and introduction rate limits | Layered per-address, principal, tenant, and global token buckets with bounded queues and stable retry guidance. (#15) | Limit partition/isolation tests and overload/soak profiles. |
 | Lease expiry removes abandoned servers | Visibility and join eligibility atomically require a fresh lease and fresh authenticated presence. (#6, #7) | Fake-clock expiry, renew/expire race, restart, and stale-host join tests. |
 | Validate game, environment, room, and protocol boundaries | Every identifier is a validated type; store keys and authorization decisions include server-derived tenant scope; protocol is exact-match in v1. (#4-#10) | Contract, tenant-isolation, incompatible-version, and confused-deputy tests. |
 | Structured audit events without secrets or reusable credentials | Allowlisted audit schema excludes metadata values, raw endpoints, tokens, and key material; event volume is bounded. (#16) | Captured-log/audit assertions and credential canary scans. |
 | Public endpoint observation | Only authenticated UDP packets from the gameplay socket establish public endpoint ownership; bounded private local candidates follow ADR 0002 and HTTP claims are never introduced. (#11) | Spoofed-source, arbitrary-target, private-range, and same-LAN/external tests. |
 | Authenticated join and punch tokens | Join issuance rechecks compatible visible listing; mediator validates scoped one-time capabilities; host consumes signed ticket. (#10-#12) | Deterministic three-party success, rejection, replay, mismatch, and timeout tests. |
 | Clear timeouts and failure results | SDK owns explicit deadlines/cancellation and returns a closed typed outcome set; NAT introduction alone is not success. (#12, #13) | Fake-clock deadline/cancellation and host-rejection tests. |
 | Isolation by game, environment, protocol, and region | Tenant and protocol are mandatory exact filters; region is bounded policy/filter data and cannot override tenant compatibility. (#5, #8, #15) | Cross-product browse/register/join isolation tests. |
 | Operational health, metrics, logging, administration, and rate limiting | Separate liveness/readiness, bounded privacy-safe metrics/logs, authenticated operator controls, and overload signals. (#15, #16, #27) | Authorization matrix, redaction tests, dashboard queries, and failure-injection checks. |
 | Service leaves gameplay path after direct connection | Mediator handles only presence/capability/introduction messages and has no gameplay forwarding API. (#4, #11) | Contract/API review, UDP unknown-message drop tests, and end-to-end traffic-path assertion. |
 | Direct traversal is not guaranteed and requires fallback | SDK distinguishes traversal failure from service/host rejection and only returns configured fallback data for caller choice. Relay is absent from v1. (#13, #20, #24) | Typed-outcome tests and TestClient scripted fallback scenarios. |
 Release readiness requires the linked implementation tests to exist and pass;
 the design documents alone do not satisfy the security promise.
@@ -0,0 +1,96 @@
 # Rendezvous v1 threat model
 Tracking: #2
 ## Scope and assets
 This model covers the public HTTP API, public LiteNetLib-compatible UDP mediator,
 operator API, client SDK, game host integration, reverse proxy, secret provider,
 observability pipeline, and the proposed future shared store. Gameplay traffic
 after direct connection and game-owned identity/admission systems are outside
 the service boundary, but their handoff is in scope.
 Assets include tenant isolation, service availability, signing and publisher
 keys, lease and connection credentials, raw endpoints, unlisted share codes,
 listing integrity, audit integrity, and the guarantee that Rendezvous does not
 turn into a reflector or private-network probe.
 ## Actors and assumptions
 - Anonymous Internet attackers can send arbitrary HTTP and UDP traffic, spoof
  source addresses where their network permits it, scrape listings, and create
  many identities or addresses.
 - Malicious publishers possess credentials only for their assigned tenant and
  may submit hostile metadata or attempt to target arbitrary endpoints.
 - Malicious clients can obtain legitimate join credentials for sessions they can
  see and may replay, race, mutate, or share those credentials.
 - A compromised game client and its SDK are fully attacker-controlled. No
  reusable secret in them is trustworthy.
 - Operators are privileged but fallible. Their actions are authenticated,
  constrained, and audited.
 - The reverse proxy, secret provider, and build/release pipeline are trusted
  dependencies. Their compromise is considered and mitigated but cannot be
  completely contained by the application.
 ## Abuse paths and controls
 ```mermaid
 flowchart TD
    A["Attacker input"] --> H{"HTTP or UDP?"}
    H -->|HTTP| V["Authenticate when required; validate tenant, schema, size, and rate"]
    H -->|UDP| U["Parse bounded datagram; validate capability before response"]
    V --> S{"Allowed and in quota?"}
    U --> E{"Capability valid, fresh, scoped, unused, and endpoint observed?"}
    S -->|No| R["Stable bounded rejection"]
    E -->|No| D["Silent drop + bounded aggregate metric"]
    S -->|Yes| State["Atomic ephemeral state transition"]
    E -->|Yes| State
    State --> O["Allowlisted audit event; no secrets/endpoints"]
 ```
 | Threat | Example | Required prevention/detection | Planned evidence |
 | --- | --- | --- | --- |
 | Spoofing and reflection | Forged UDP source causes traffic to a victim | No response before valid capability proof; send responses only to observed authenticated sources; at most two responses and <=2.0 verified byte amplification | Packet-level spoof/reflection tests and amplification accounting |
 | Private-network probing | Publisher supplies `127.0.0.1`, link-local, or another victim as a same-LAN candidate | Accept only bounded private-unicast claims inside a scoped authenticated UDP contribution; reject prohibited ranges; disclose only to the opposite role in that attempt; bound SDK probes | Endpoint classification matrix and three-party adverse tests |
 | Capability/ticket replay | Reuse a captured token to repeat introductions or connect | Short expiry, role/tenant/attempt scope, atomic one-time consumption, bounded skew, key rotation | Concurrent replay and post-expiry tests with golden vectors |
 | Cross-tenant access | Game A browses, renews, or joins Game B | Server-derived principal scope on every lookup and atomic mutation; indistinguishable not-found response | Tenant isolation tests across every endpoint/store operation |
 | Listing spam and scraping | Flood registrations or enumerate public sessions | Trust-mode quotas, per-principal/address limits, bounded pages/cursors, rate limits, aggregate alerts | Rate-limit, cursor-tamper, and sustained-load tests |
 | Metadata injection | Control characters or markup attack logs/UI | UTF-8/schema/size validation; store as data; exclude values from audit; SDK does not render markup | Malformed Unicode/JSON corpus and TestClient safe-display tests |
 | Credential theft | Secret appears in log, URL, metric, crash, or package | Credentials in headers/bodies only; allowlisted logging; secret-provider indirection; no credential metric labels | Log-capture tests, repository/package scans, rotation exercise |
 | Parser/resource exhaustion | Oversized, nested, fragmented, or high-rate input | Fixed ceilings, bounded parsers/queues/concurrency, early rejection/drop, no input-sized logging | Fuzz/property corpus, allocation limits, overload tests |
 | Stale or crashed host | Dead listing remains joinable | Both lease and recent authenticated presence required; atomic expiry; join rechecks freshness | Fake-clock lifecycle and join-race tests |
 | Clock manipulation | Token accepted outside intended lifetime | Server-issued timestamps, monotonic elapsed-time for local expiry, <=30 s wall-clock skew | Boundary and clock-jump tests |
 | Operator misuse | Unauthorized enumeration/revocation or secret exposure | Separate strong auth/network policy, least privilege, tenant scope, immutable audit, secrets never readable through API | Authorization matrix and audit completeness tests |
 | Reverse-proxy confusion | Forged forwarded address bypasses limits | Trust forwarding headers only from allowlisted proxies; direct traffic uses socket peer | Forwarded-header spoof tests |
 | Store race | Renew/revoke/expire/replay operations interleave | Compare-and-swap/transactional interfaces and deterministic outcomes | Parallel race tests with a fake clock |
 | Dependency/supply-chain compromise | Malicious or drifting package/build output | Central pinning, lock files, reproducible builds, vulnerability review, signed release provenance | Locked clean restore, dependency audit, artifact verification |
 | Availability attack | Valid-looking traffic fills CPU, memory, queues, logs | Layered quotas, bounded queues/tasks, graceful overload, readiness/drain, capacity alerts | Load/soak/resilience gates and forced saturation tests |
 ## Security invariants
 The implementation and its tests must preserve these invariants:
 1. No UDP response is sent to an endpoint that has not presented a valid scoped
   capability from that observed endpoint.
 2. No browse response contains an endpoint, secret, internal attempt ID, or
   credential.
 3. Every state lookup and mutation includes server-derived game/environment
   scope; caller-supplied scope alone is never authoritative.
 4. A listing is visible and joinable only while both lease and presence are
   fresh at the atomic decision point.
 5. A capability or ticket can cause at most one successful state transition for
   its intended role and attempt.
 6. Join authorization never bypasses host-owned final admission.
 7. Input cannot create unbounded memory, work, response bytes, metric labels, or
   log volume.
 8. Raw endpoints and secrets never enter normal logs, traces, audit payloads, or
   metric dimensions.
 ## Residual risk
 Direct traversal cannot work through every NAT, firewall, carrier, or platform
 policy. Rate limiting cannot eliminate distributed abuse. A compromised trusted
 proxy, secret provider, operator identity, game grant issuer, or host credential
 can act within its granted scope until detected and revoked. Unlisted share
 codes can be disclosed by recipients. These risks are communicated as typed
 outcomes and operational signals rather than hidden behind a success claim.