# Networking rules

Evidence IDs and source scope are defined in [SOURCES.md](SOURCES.md). Each rule has a corresponding row in [TEST-CASES.md](TEST-CASES.md). Game assumptions are proposed model choices for later Spec Kit validation.

## RouteMaul

### R-01 IPv4 unicast forwarding

Evidence: RFC1812 / 5.2.1 | RFC1812 / 5.2.4.2

A router determines whether a datagram is for local delivery or forwarding. For ordinary nonlocal unicast traffic, destination-based forwarding selects an outgoing path and next hop.

**Game assumption.** Use ordinary transit IPv4 with no IP options, policy routing, NAT, or multicast.

### R-02 Longest-prefix matching

Evidence: RFC1812 / 5.2.4.3 | RFC4632 / 5.1

IPv4 longest-prefix matching selects the installed route with the greatest prefix length among matching destination prefixes. A default /0 matches every destination and loses to any matching more-specific prefix. Route preference among candidates for the same prefix is a separate control-plane decision.

**Game assumption.** One installed route per prefix. ECMP and administrative distance are outside the initial model.

### R-03 Connected routes, static routes, and next-hop resolution

Evidence: RFC1812 / 5.2.4.2 | RFC1812 / 5.2.4.3 | RFC1812 / 3.3.2 | CISCO-ROUTE / Static Routes

For an on-link destination, the destination itself is the IP next hop. A remote destination uses a router next hop. Cisco IOS XE distinguishes configured static routes from active routing entries and removes a static entry when its associated interface fails or its configured next hop becomes unresolvable. Route resolution and Ethernet neighbor resolution are separate checks.

**Game assumption.** An up, addressed interface contributes its connected prefix. A static gateway must resolve directly onto an up connected interface. Treat neighbor reachability as a separate topology condition. Recursive static resolution needs its own design.

### R-04 TTL expiration and routing loops

Evidence: RFC1812 / 5.3.1 | RFC1812 / 5.2.7.3 | RFC1812 / 4.3.2.7 | RFC1812 / 4.3.2.8

A forwarding router decrements IPv4 TTL by at least one. When it reaches zero, the router drops the packet and normally generates ICMP Time Exceeded code 0 for eligible unicast traffic, subject to suppression, rate limiting, and the interface option. TTL limits the lifetime of packets in a routing loop.

**Game assumption.** Decrement by exactly one per forwarding hop. Test an eligible packet with ICMP enabled and rate limiting inactive. Recording a drop reason and delivering an ICMP reply are separate outcomes.

### R-05 Unreachable destination

Evidence: RFC1812 / 5.2.4.3 | RFC1812 / 5.2.7.1

Lookup failure with no matching route, including no usable default, drops the packet. RFC 1812 distinguishes network unreachable from failure to reach a host on a directly connected network. ICMP delivery depends on eligibility and a return path.

**Game assumption.** Report no route and neighbor failure as distinct diagnostic reasons. ICMP tests use eligible traffic and an available return path.

## PacketMaul

### P-01 Ordered ACL matching and implicit deny

Evidence: C9300-ACL / Standard and Extended IPv4 ACLs | CISCO-ACL / Process ACLs

In a Cisco IPv4 packet-filter ACL, the first matching entry decides permit or deny. An applied, populated ACL denies any packet that matches no entry. Prefix length does not change ACL entry order.

**Game assumption.** Distinguish an absent ACL from a populated ACL. Empty and undefined ACL references are outside the initial model.

### P-02 Source and destination prefix matching

Evidence: C9300-ACL / Standard and Extended IPv4 ACLs | CISCO-ACL / Masks

Cisco standard IPv4 ACLs match source addresses. Extended ACLs can constrain both source and destination. IOS wildcard mask bits of zero require a match and bits of one ignore the corresponding address bits.

**Game assumption.** Represent contiguous address matches as CIDR prefixes, including /0 and /32. Cisco wildcard patterns beyond contiguous prefixes are outside the initial model.

### P-03 Protocol, ports, and stateless TCP filtering

Evidence: CISCO-ACL / Extended ACLs | CISCO-ACL / Reflexive ACLs

Extended ACLs can match IP protocol and applicable TCP or UDP source and destination ports. Ordinary static rules evaluate each packet independently. Return traffic needs its own matching permit wherever it encounters an applied ACL. Classic IOS reflexive ACLs are a distinct session-aware feature.

**Game assumption.** Test unfragmented traffic with available transport headers. Omit TCP flag predicates, reflexive rules and connection tracking.

### P-04 Interface direction and connectivity diagnostics

Evidence: CISCO-ACL / Define In, Out, Inbound, Outbound, Source, and Destination | C9300-ACL / Supported ACLs | RFC1812 / 5.2.4.3

In the documented IOS transit router pipeline, an inbound ACL checks arrivals before forwarding lookup, while an outbound ACL checks packets selected for the outgoing interface. Permit lets processing continue. A route or adjacency failure can independently prevent delivery.

**Game assumption.** Use Layer 3 transit ACLs with an explicit interface and direction. Do not generalize to bridged traffic or locally generated packets.

## QueueMaul

### Q-01 Classification and marking

Evidence: RFC2475 / 2.3.1 | RFC2475 / 2.3.3.2 | C9300-QOS / Classification | C9300-QOS / Marking

Classification selects a traffic class from packet fields or existing markings. A configured marker can set or change the DSCP. Class selection and marking are separate operations.

**Game assumption.** One policy class per packet, with a configured default class. The same DSCP can be preserved or rewritten according to policy.

### Q-02 DSCP and differentiated services

Evidence: RFC2474 / 3 | RFC2474 / 4.1 | RFC3168 / 5 | RFC3260 / 4 | RFC2475 / 2.4 | RFC3246 / 2.7 | RFC2597 / 2

The high six bits carry DSCP and the low two carry ECN. A configured Diffserv node maps DSCP to a per-hop behavior. DSCP 0 identifies the default PHB and 46 is the recommended EF codepoint. AF specifies classes and relative drop precedence within each class. Marking alone gives no end-to-end guarantee.

**Game assumption.** Use explicit per-hop DSCP mappings and preserve ECN on a DSCP-only rewrite. AF conformance requires additional queue policy and is outside the initial scheduler.

### Q-03 Egress queues and congestion drops

Evidence: C9300-QOS / Queuing and Scheduling | C9300-QOS / Weighted Tail Drop | RFC2597 / 4 | RFC3246 / 2.5

Packets wait for service when offered load exceeds egress capacity. Finite buffers and drop policy can discard packets. Catalyst 9300 documents weighted tail drop thresholds. AF and EF each have specific loss behavior and assumptions.

**Game assumption.** Use a bounded FIFO per class with deterministic tail drop. Measure waiting slots independently of the in-flight packet. This is a simplified queue, with no claim of AF or Catalyst hardware equivalence.

### Q-04 Strict priority and best-effort service

Evidence: RFC3246 / 2.10 | C9300-QOS / Priority Queues | C9300-QOS / Priority Queue Policer | CISCO-LLQ / LLQ Bandwidth Allocation

A non-preemptive strict-priority scheduler selects an eligible priority packet before a waiting lower-class packet at the next transmission opportunity. Sustained unbounded eligible priority traffic can starve lower classes. EF requires limitation when the mechanism permits unlimited preemption of other traffic. Cisco LLQ with configured bandwidth meters priority traffic under congestion in the generic guide.

**Game assumption.** Treat the unbounded variant as a starvation demonstration. For bounded priority, use an explicit eligibility budget. Catalyst policer and generic LLQ accounting require separate configuration evidence.

### Q-05 Bandwidth allocation and limits

Evidence: RFC2475 / 2.5 | RFC3246 / 2.1 | C9300-QOS / Bandwidth

QoS allocates service across classes within available link capacity. A per-hop rate objective depends on configuration, offered load and the relevant PHB conditions. QoS cannot create additional link bandwidth or guarantee end-to-end performance by marking traffic alone.

**Game assumption.** Use fixed link capacity and measure on-wire service over complete transmission intervals. Ignore framing overhead only in the idealized rate examples.

## Unresolved

Platform-specific recursive static routing and ECMP, fragments and control-plane ACLs, undefined or empty ACL binding, hardware queue thresholds and policer accounting, and multi-domain QoS guarantees need additional source review before adding corresponding mechanics. No device lab verification was performed.
