Test report

Note

Auto-generated from a live ctest run by bench/gen_test_report.py (regenerated in CI before every Pages deploy). Not hand-maintained.

Summary

suites	passing	conformance vectors	wall time	verdict
222	222/222	96	81.36s	✅ all green

By subsystem

category	suites	passing
Codec (L2/L3)	13	✅ 13/13
Substrate (L0/L1)	16	✅ 16/16
Graph (L4)	21	✅ 21/21
Net (FWD plane)	21	✅ 21/21
Transport	9	✅ 9/9
Examples	70	✅ 70/70
other	72	✅ 72/72

Suites

Each row’s description is the test file’s own @brief, read from the source at generation time, and source links to that file. Neither is transcribed into this generator, so neither can drift from the test.

Codec (L2/L3)

suite	result	time	covers	source
byteorder	✅ pass	0.00s	Unit tests for the little-endian (de)serialization primitive (byteorder.hpp).	byteorder_test.cpp
can_frames	✅ pass	0.00s	transport_can PURE framing-layer test (#55).	can_frames_test.cpp
conformance	✅ pass	0.01s	Conformance harness for the seed vectors under tests/conformance/vectors/v1/.	conformance_runner.cpp
frame	✅ pass	0.00s	Frame-codec nesting + length-width tests.	frame_test.cpp
key_view	✅ pass	0.00s	Unit tests for tr::wire::key_view_t (key_view.hpp) — the canonical-key NAME navigation the L4 graph dispatch and ACL-inheritance walks funnel through.	key_view_test.cpp
length_prefix_framer	✅ pass	0.00s	length_prefix_framer unit test — drives the u32-length-prefix reassembly state machine directly (no QUIC connection), the whole point of extracting it from transport_quic / transport_webtransport (finding #4): prefix/body split across chunks, multiple frames per chunk, empty records, over the protocol cap => malformed, over local capacity => backpressure drain + resync (#932), and reset.	length_prefix_framer_test.cpp
rope_decode	✅ pass	0.00s	Differential test for the rope-aware grammar (ADR-0048 §1): wire::validate_rope over a scatter-gather rope MUST reach the exact same verdict as wire::decode over the equivalent flat bytes — for every adversarial split, including splits that fall mid-header, mid- trailer, and mid-payload.	rope_decode_test.cpp
rope_decode_fuzz	✅ pass	0.08s	Rope-source differential FUZZER for the wire grammar (ADR-0048 §consequences: “the differential fuzzers gain a rope-source mode — same bytes split at adversarial link boundaries MUST decode identically to the contiguous case, including mid-header splits”).	rope_decode_fuzz_test.cpp
tlv_arena	✅ pass	0.00s	Terminus arena decoder test (ADR-0041).	tlv_arena_test.cpp
tlv_view	✅ pass	0.00s	Tests for the lazy rope-backed decode view (ADR-0053): tlv_view_t over a scatter-gather rope must (a) agree with the eager decoder node-for-node when fully walked, (b) be actually LAZY — siblings of a corrupt TLV deliver, the corrupt one fails only its own verify(), bytes are shared not copied — and (c) keep its links’ segments alive past the source rope (owning tier).	tlv_view_test.cpp
ws	✅ pass	0.00s	transport_ws PROTOCOL-layer test (#54).	ws_test.cpp
ws_handshake_bound	✅ pass	0.51s	#934 car B — the WS OPENING HANDSHAKE is bounded by a deployment-injected pre-auth budget: an over-budget request is refused BEFORE it is buffered, counted, and closed.	ws_handshake_bound_test.cpp
ws_rx_bound	✅ pass	1.25s	#872 — the WS DATA path is bounded by the injected seam: a peer may not name the receiver’s memory budget, and every refusal is COUNTED.	ws_rx_bound_test.cpp

Substrate (L0/L1)

suite	result	time	covers	source
mem_pool_source	✅ pass	0.00s	Unit tests for the bounded RECYCLING block source (tr::mem::pool_source_t, #597).	mem_pool_source_test.cpp
mem_source	✅ pass	0.00s	Unit tests for the nothrow failable-block seam (mem_source.hpp, #551).	mem_source_test.cpp
mem_source_pmr	✅ pass	0.00s	Unit tests for the std::pmr adapter over the block seam (mem_source_pmr.hpp, #873).	mem_source_pmr_test.cpp
mem_sync_policy	✅ pass	0.02s	The mem::synchronized_pool_t SYNCHRONISATION-POLICY seam (#770, ADR-0060 §2): the critical section is a compile-time policy, and it is the thing that makes the pool safe at a receive seam.	mem_sync_policy_test.cpp
mem_sync_pool	✅ pass	0.09s	ADR-0060 §2 — the thread-safe pool value_backend_: mem::sync_pool_t must serialize concurrent alloc (writer threads) with cross-thread destroy (a segment reclaimed on the reader/subscriber thread that drops the last ref).	mem_sync_pool_test.cpp
path	✅ pass	0.00s	Path parsing / validation (docs/reference/03-addressing.md).	path_test.cpp
path_label	✅ pass	0.00s	The RFC-0027 path-label value, its §5.3 ELEMENT, and the mint table — codec, element grammar, the §12.5 vectors, mint, lookup, retire, SATURATE-AND-RETIRE, and §8.3.	path_label_test.cpp
path_label_codec	✅ pass	0.00s	RFC-0027 car 3 — the mixed-element codec (path_element.hpp) and the origin-side label cache on path_t (§6.1), with §9’s layering discipline asserted rather than assumed.	path_label_codec_test.cpp
path_label_forward	✅ pass	0.00s	The two links, spelled as the RFC-0014 mount runs a real node uses.	path_label_forward_test.cpp
path_label_terminus	✅ pass	0.00s	The inbound link, spelled as the three-segment mount run a real node uses.	path_label_terminus_test.cpp
path_ref	✅ pass	0.00s	PATH_REF (0x14) codec test — the bound-path wire form of RFC-0024 §4.	path_ref_test.cpp
rope	✅ pass	0.00s	rope_t small-buffer storage + the value-consumption accessors (ADR-0053 §6).	rope_test.cpp
rope_cursor_assert	✅ pass	0.00s	rope_cursor bounds preconditions (#916) — the debug-assert parity death tests.	rope_cursor_assert_test.cpp
rope_cursor_release_guard	✅ pass	0.00s	The cursor window guarantee in a RELEASE build (#986) — the gate that cannot pass vacuously under NDEBUG.	rope_cursor_release_guard_test.cpp
substrate	✅ pass	0.00s	L0/L1 substrate tests: refcount lifetime (the canonical intrusive_ptr orderings), zero- copy subview/concat, rope serialization equivalence (the docs/reference/02 proof obligation), the view->TLV cast claim with the lifetime gap M1 left open now closed, and the bounded pool backend.	substrate_test.cpp
substrate_no_atomic	✅ pass	0.00s	L0/L1 substrate tests: refcount lifetime (the canonical intrusive_ptr orderings), zero- copy subview/concat, rope serialization equivalence (the docs/reference/02 proof obligation), the view->TLV cast claim with the lifetime gap M1 left open now closed, and the bounded pool backend.	substrate_test.cpp

Graph (L4)

suite	result	time	covers	source
acl	✅ pass	0.00s	Per-endpoint :acl — storage (#81-A) + core-subset enforcement (#81, ADR-0018/ 0020/0026).	acl_test.cpp
acl_cache_race	✅ pass	0.10s	ACL-cache coherence under a concurrent subtree invalidation (#880, ADR-0078).	acl_cache_race_test.cpp
app_fields	✅ pass	0.03s	RFC-0010 owner app fields: the field descriptor table, :settings.app.* read/write gating, the two-part :schema, container reads, the owner apply seam, and the announce-write convention (§C — a field write never wakes await and never propagates).	app_fields_test.cpp
children	✅ pass	0.00s	#82 — in-band vertex creation via a :children[] SPEC write (ADR-0017, ADR-0021; docs/reference/05 §0x0E).	children_test.cpp
delivery_drops	✅ pass	0.00s	Per-cause delivery-drop counters (graph_t::delivery_drops).	delivery_drops_test.cpp
edge_eviction	✅ pass	0.00s	RFC-0009 §D extended to peer departure — subscriber-edge eviction on link teardown, and edge-slot reuse.	edge_eviction_test.cpp
effective_acl	✅ pass	0.19s	effective_acl_t unit test (ADR-0050, the previously-untested half) — the pure effective- ACL merge semantics driven directly with synthetic ACE lists (no graph, no locks, no wall clock): own-before-ancestors ordering, nearest-first ancestor order, INHERIT gating at merge time, any-present-ACE-closes (even expired), open-by-default over an empty merge, and DENY first-match-per-bit ordering under the full policy.	effective_acl_test.cpp
folded_children	✅ pass	0.00s	L4 fold, Slice 0 — the folded :children projection is byte-identical to the materialized serialize, and is a genuine scatter-gather rope a cursor walks.	folded_children_test.cpp
graph	✅ pass	0.14s	L4 graph-runtime tests.	graph_test.cpp
graph_config_race	✅ pass	5.50s	#1049 — graph_t’s five configuration seams under a setter storm: no reader is ever handed a destroyed target, a torn pair, or a half-mutated container.	graph_config_race_test.cpp
graph_oom_softfail	✅ pass	0.00s	#477 — the store/delivery path’s nothrow soft-fail discipline under OOM.	graph_oom_softfail_test.cpp
graph_pmr	✅ pass	0.00s	#361 §5 — the ADR-0039 §1 injection seam on graph_t: per-write LKV allocations (control block + rope) draw from the constructor-injected std::pmr::memory_resource, and a default-constructed graph keeps the standard heap (zero churn).	graph_pmr_test.cpp
graph_value_backend	✅ pass	0.00s	ADR-0060 — the write-path value byte-buffer seam on graph_t: the copy-store flatten of a branch/field write draws its owned segment from the injected value_backend_ (a mem_backend_t), not the default heap.	graph_value_backend_test.cpp
identity	✅ pass	0.00s	#406 / RFC-0011 — the node identity facet: read <vertex>:identity.	identity_test.cpp
registry_teardown	✅ pass	0.00s	Registry teardown — child_registry_t::erase / remove_child / remove_connection (#494).	registry_teardown_test.cpp
retire	✅ pass	8.01s	graph_t::retire() — RFC-0009 §B/§C/§E.6 vertex retirement (#407).	retire_test.cpp
security_acl	✅ pass	0.00s	security_acl unit test (ADR-0050) — drives the PURE ACL policy seam directly, no graph, no locks, no wall clock: both adapters (allow_only / full), the ACE edge cases (expiry, EVERYONE@, per-bit matching, INHERIT-flag filtering, first-match-per-bit DENY ordering), and the typed parse/build round-trip that replaces the per-test byte builders.	security_acl_test.cpp
subtree	✅ pass	0.00s	RFC-0005 — subtree subscriptions (vertical bubbling), branch-write decomposition, and write-creates.	subtree_test.cpp
subtree_snapshot	✅ pass	0.00s	RFC-0005 §C follow-on — the COMPOSED BRANCH READ: a plain READ of a vertex with ≥ 1 registered child serves the folded POINT tree of its registered subtree’s landed LKVs.	subtree_read_test.cpp
vertex	✅ pass	0.04s	vertex_t verb-interface unit tests — a BARE vertex, no graph_t (the point of the verb seam: the vertex’s storage/readiness/edge/ACL state is testable in isolation).	vertex_test.cpp
vertex_size	✅ pass	0.00s	The vertex RAM-diet regression gate (#361 §8): compile-time ceilings on sizeof(vertex_t) and the hot/cold split invariants, plus runtime probes that prove the cold extension block is NOT allocated for the common default leaf and IS allocated exactly when the identity needs it.	vertex_size_test.cpp

Net (FWD plane)

suite	result	time	covers	source
compact_cache	✅ pass	0.00s	ADR-0062 increment 2 — a warm binding is fast, and it invalidates correctly.	compact_cache_test.cpp
fwd_bind_refusal_unwind	✅ pass	0.00s	#833 — a REFUSED forwarding bind must strand neither the out-label it took nor the egress route retained with it, through the production on_advertise wiring.	fwd_bind_refusal_unwind_test.cpp
fwd_compact	✅ pass	0.49s	RFC-0004 / ADR-0035 slice 4 — the route-handle: ws delivery-compaction, proven over LIVE transport_ws.	fwd_compact_test.cpp
fwd_compact_acl	✅ pass	0.00s	#974 — a label-compacted (COMPACT) delivery is ACL-gated under the INBOUND LINK’s name, exactly like the full-route FWD{WRITE} it is the compaction of.	fwd_compact_acl_test.cpp
fwd_fanout	✅ pass	0.00s	RFC-0004 / ADR-0035 slice 4 (#136) — the PRODUCER remote fan-out.	fwd_fanout_test.cpp
fwd_flatten_backend	✅ pass	0.00s	#730 — every rope flatten fwd_router_t performs draws from its INJECTED mem_backend_t, and an exhausted one is answered by value, never by storing an empty value and reporting success.	fwd_flatten_backend_test.cpp
fwd_frame_view	✅ pass	0.00s	fwd_frame_view unit test — drives the FWD offset-dispatch cluster directly (no router, no transports), the point of extracting it from fwd_router.cpp (the length_prefix_framer precedent): first-dst-seg / op / control peeks over BOTH cursors (contiguous span + adversarially split rope), the shrunk-dst / grown-src head rebuild proved BYTE-EXACT against a reference re-encode, stack_writer clamp-to-empty overflow, and malformed rejects.	fwd_frame_view_test.cpp
fwd_malformed_control	✅ pass	0.00s	The malformed-control-frame rejection surface of fwd_router_t — the guards a peer’s bytes reach first, and which nothing asserted.	fwd_malformed_control_test.cpp
fwd_multihop	✅ pass	0.40s	RFC-0004 / ADR-0035 slice 3 — multi-hop FWD forwarding + zero-copy src accumulation, proven over LIVE transport_ws links.	fwd_multihop_test.cpp
fwd_readvertise_reuse	✅ pass	0.00s	#913 — a re-advertise cycle must burn no downstream label and leak no egress entry.	fwd_readvertise_reuse_test.cpp
fwd_reconnect_selfheal	✅ pass	0.00s	RFC-0004 §E.1 — a MID-CHAIN reconnect self-heals: clear_link sweeps the cross-link ingress bindings whose downstream half crossed the cleared link (#716).	fwd_reconnect_selfheal_test.cpp
fwd_rope_forward	✅ pass	0.00s	ADR-0053 ④b — the FWD forward hop over a MULTI-LINK rope, WITHOUT flattening.	fwd_rope_forward_test.cpp
fwd_sink_race	✅ pass	0.07s	#914 — a router sink’s {fn, ctx} pair is published as a UNIT, so no receive thread can hand a newly installed fn the previous sink’s ctx.	fwd_sink_race_test.cpp
fwd_terminus_reject	✅ pass	0.00s	The terminus resolver’s malformed-FWD rejection surface — where every remote operation lands, and which nothing asserted.	fwd_terminus_reject_test.cpp
fwd_two_mount	✅ pass	0.00s	The TWO-MOUNT route (#419): a dst that crosses two net/<module>/<name> mounts.	fwd_two_mount_test.cpp
mount_routing	✅ pass	0.00s	The per-module mount-routing primitives (ADR-0061 / RFC-0014 S2a).	mount_routing_test.cpp
net_control_plane_race	✅ pass	0.13s	ADR-0063 §3 — control-plane writers are serialized, and the forward reader is not.	net_control_plane_race_test.cpp
op_resolve	✅ pass	0.01s	RFC-0004 / ADR-0035 — op_resolver_t host tests, over the ADR-0041 terminus arena.	op_resolve_test.cpp
op_resolve_view	✅ pass	0.00s	ADR-0053 §7 (3c-ii) — the DIFFERENTIAL ORACLE for the two terminus-resolver instantiations.	op_resolve_view_test.cpp
route_handle	✅ pass	0.06s	route_handle_t unit test (Brick 4, ADR-0038 §3 / ADR-0039): the label state is per- connection — pmr-backed tables with a per-link mutex.	route_handle_test.cpp
transport_vertex	✅ pass	2.08s	#83 Stage-1 — transport/connection as a / vertex (ADR-0027), the SHELL over the live path (ADR-0037 Stage-1).	transport_vertex_test.cpp

Transport

suite	result	time	covers	source
can_tx_pool	✅ pass	0.02s	#383 — can_tx_pool_t ownership/backpressure host suite.	can_tx_pool_test.cpp
tcp	✅ pass	4.60s	M6 TCP transport tests: length-prefix framing over a real localhost TCP stream.	tcp_test.cpp
transport_can	✅ pass	1.81s	#55 (increment 2) — transport_can SocketCAN-binding tests over an in-memory fake link, so they pass in plain Docker with NO kernel CAN (vcan).	transport_can_test.cpp
transport_can_peers	✅ pass	0.33s	ADR-0044 — stateless transport-peer enumeration + transparent per-peer FWD over the CAN bus binding, proven over the in-memory fake link (no kernel CAN):	transport_can_peers_test.cpp
transport_can_vcan	✅ pass	0.00s	#55 (increment 2) — REAL-bus smoke test for transport_can over Linux SocketCAN.	transport_can_vcan_test.cpp
transport_conformance	✅ pass	0.56s	Transport seam-conformance suite.	transport_conformance_test.cpp
udp	✅ pass	1.47s	M5 UDP transport tests: raw frame delivery over a real localhost UDP socket, and an end- to-end two-node FWD delivery through graph_t + fwd_router_t over UDP (the explicit-source- routed net plane, ADR-0040 — no bridge_t/ROUTER).	udp_test.cpp
write_all_eintr	✅ pass	0.10s	#903 — the full-write helpers must RESUME an interrupted write, not truncate.	write_all_eintr_test.cpp
ws_transport	✅ pass	2.78s	#54 — transport_ws SERVER socket-layer tests.	ws_transport_test.cpp

Examples

suite	result	time	covers	source
example_acl_everyone_reserved	✅ pass	0.00s	—	—
example_acl_expiry	✅ pass	0.00s	—	—
example_acl_inherit	✅ pass	0.00s	—	—
example_acl_open_by_default	✅ pass	0.00s	—	—
example_acl_parse_strict	✅ pass	0.00s	—	—
example_acl_policy_profiles	✅ pass	0.00s	—	—
example_acl_right_bits	✅ pass	0.00s	—	—
example_acl_subject_resolver	✅ pass	0.00s	—	—
example_graph_await	✅ pass	0.07s	—	—
example_graph_children	✅ pass	0.00s	—	—
example_graph_handler_vertex	✅ pass	0.00s	—	—
example_graph_read_write	✅ pass	0.00s	—	—
example_graph_register	✅ pass	0.00s	—	—
example_graph_retire	✅ pass	0.00s	—	—
example_graph_stream	✅ pass	0.00s	—	—
example_graph_write_creates	✅ pass	0.00s	—	—
example_in_process_pubsub	✅ pass	0.05s	In-process publish/subscribe over the L4 graph — the M3 P0 node, end to end.	in_process_pubsub.cpp
example_mem_block_array	✅ pass	0.00s	—	—
example_mem_block_source	✅ pass	0.00s	—	—
example_mem_bump_source	✅ pass	0.00s	—	—
example_mem_bump_upstream	✅ pass	0.00s	—	—
example_mem_pool_source	✅ pass	0.00s	—	—
example_mem_size_classes	✅ pass	0.00s	—	—
example_mem_source_resource	✅ pass	0.00s	—	—
example_mem_source_vs_backend	✅ pass	0.00s	—	—
example_net_can_bus_peers	✅ pass	0.00s	—	—
example_net_dial_and_listen	✅ pass	0.20s	—	—
example_net_kind_catalog	✅ pass	0.00s	—	—
example_net_multi_peer_listener	✅ pass	0.40s	—	—
example_net_tcp_stream_framing	✅ pass	0.00s	—	—
example_net_transport_seam	✅ pass	0.00s	—	—
example_net_udp_datagram	✅ pass	0.21s	—	—
example_net_ws_upgrade	✅ pass	0.20s	—	—
example_pubsub_fanout	✅ pass	0.01s	Pub/sub fan-out — one publisher, a growing set of subscribers, and the per-delivery dispatch cost as fan-out scales.	pubsub_fanout.cpp
example_rope_scatter	✅ pass	0.00s	L1 scatter-gather — compose a multi-link rope_t with zero byte copies, then measure the cost of scatter-gather egress vs. the one flatten copy.	rope_scatter.cpp
example_route_child_table	✅ pass	0.00s	—	—
example_route_dst_is_source_route	✅ pass	0.00s	—	—
example_route_label_compact	✅ pass	0.00s	—	—
example_route_label_stale	✅ pass	0.00s	—	—
example_route_multi_hop	✅ pass	0.00s	—	—
example_route_qualified_mount	✅ pass	0.00s	—	—
example_route_reply_home	✅ pass	0.00s	—	—
example_route_terminus_or_forward	✅ pass	0.00s	—	—
example_sub_callback	✅ pass	0.00s	—	—
example_sub_durability_latch	✅ pass	0.00s	—	—
example_sub_retire	✅ pass	0.00s	—	—
example_sub_subtree	✅ pass	0.00s	—	—
example_sub_terminal_delivery	✅ pass	0.00s	—	—
example_sub_unsubscribe	✅ pass	0.00s	—	—
example_sub_unsubscribe_from_dispatch	✅ pass	0.00s	—	—
example_tree_of_ropes	✅ pass	0.00s	The three composition axes, made visible — why a libtracer node is a tree of ropes, not a rope of ropes.	tree_of_ropes.cpp
example_two_node_fwd	✅ pass	0.01s	Two nodes over a wire — an FWD write routed between two graphs, and the end-to-end delivery latency across the “wire”.	two_node_fwd.cpp
example_view_borrow	✅ pass	0.00s	—	—
example_view_device_rope	✅ pass	0.00s	—	—
example_view_pool_backend	✅ pass	0.00s	—	—
example_view_rope_compose	✅ pass	0.00s	—	—
example_view_rope_subrope	✅ pass	0.00s	—	—
example_view_segment_refcount	✅ pass	0.00s	—	—
example_view_subview	✅ pass	0.00s	—	—
example_view_sync_pool	✅ pass	0.00s	—	—
example_wire_arena_decode	✅ pass	0.00s	—	—
example_wire_codec	✅ pass	0.03s	Wire codec deep-dive — build a structured frame, inspect its bytes, and measure encode / decode / round-trip throughput.	wire_codec.cpp
example_wire_decode_refusals	✅ pass	0.00s	—	—
example_wire_lazy_view	✅ pass	0.00s	—	—
example_wire_packed_path	✅ pass	0.00s	—	—
example_wire_path_escape	✅ pass	0.00s	—	—
example_wire_roundtrip	✅ pass	0.00s	L2/L3 wire codec round-trip — build a TLV, encode to bytes, decode back.	wire_roundtrip.cpp
example_wire_structured_vs_opaque	✅ pass	0.00s	—	—
example_wire_tlv_header	✅ pass	0.00s	—	—
example_wire_trailer	✅ pass	0.00s	—	—

other

suite	result	time	covers	source
batch	✅ pass	0.00s	The BATCH record (0x80) — the fold, the derived sample clock, and what the reader declines (RFC-0025 §4.2.1 / §4.1.2, batch.hpp).	batch_test.cpp
bound_forward	✅ pass	4.00s	RFC-0024 car 3 — the bound-path FORWARDER hop and the origin-side bind, end to end.	bound_forward_test.cpp
bound_path	✅ pass	0.00s	RFC-0024 — bound-path routing and minting, at the terminus.	bound_path_test.cpp
bus_can_guard	✅ pass	4.08s	—	—
bus_module	✅ pass	0.40s	The ADR-0044 BUS module as a build-time-closed seam (#375 deliverable 3) — what tr::graph::default_config_t::kBusLinks closes, and what it must NOT disturb.	bus_module_test.cpp
child_rx_churn	✅ pass	0.00s	Receiver-ctx churn (#884): a re-added NAME resolves to its CURRENT tenancy, and remove/re-add cycles do not grow the router’s published receiver chain.	child_rx_churn_test.cpp
collect	✅ pass	0.01s	graph_t::collect() / parked_seam_count() — the explicit end of retirement’s value-seam park (#576, the direction-3 ruling that supersedes ADR-0072).	collect_test.cpp
config_reader	✅ pass	0.00s	#927 / #985 — tr::wire::config_reader_t: the PAIR-CONSUMING SETTINGS walk every transport config parser AND both L4 readers (the creation SPEC, the SUBSCRIBER QoS SETTINGS) share. tr::net::config_reader_t is its alias.	config_reader_test.cpp
conn_add_oom	✅ pass	0.00s	make_connection under a REFUSING allocator: a connection whose link cannot be wired into the router is ROLLED BACK, not published as a live-looking dead connection (#930).	conn_add_oom_test.cpp
delivery_mode_race	✅ pass	0.04s	delivery_mode against a concurrent assign: the sweep-set exclusion and the plain-byte data race (#895, RFC-0008 §B/§C).	delivery_mode_race_test.cpp
edge_cold_half_share	✅ pass	0.00s	#1442 / #1448 — NOBODY deep-copies a subscription’s cold half any more. A republish shares it (so an admission’s allocation count is flat in the edges already on the vertex) and so does the dispatch snapshot (so a WRITE’s allocation count is flat in the remote fan-out).	edge_cold_half_share_test.cpp
edge_publish	✅ pass	0.30s	The published edge array and its edge pin (#635): a snapshot sees the OLD edge set or the NEW one and never a torn mixture, an unsubscribe stops delivering at once, the pin is never held across a dispatch, and nothing is freed under a reader or leaked at teardown.	edge_publish_test.cpp
empty_src_unacked	✅ pass	0.00s	RFC-0004 Amendment 2 (#1502, ruling on #1491) — a zero-length src is “no reply requested”: the terminus applies the WRITE and stays SILENT.	empty_src_unacked_test.cpp
esp_ws_client_dial	✅ pass	2.30s	—	—
esp_ws_client_liveness	✅ pass	8.71s	—	—
esp_ws_client_recv	✅ pass	2.06s	—	—
esp_ws_client_teardown	✅ pass	8.70s	—	—
field_dispatch	✅ pass	0.00s	The field surface’s READ/WRITE dispatch parity guards (#869).	field_dispatch_test.cpp
folded_read_backend	✅ pass	0.00s	#831 — BOTH folded READs’ POINT headers draw from graph_t’s injected value_backend (ADR-0060), not the global heap — and exhaustion degrades by value (BACKPRESSURE), never a throw.	folded_read_backend_test.cpp
grammar_bounds	✅ pass	0.00s	Unit tests for the grammar’s total-encoded-size bound (grammar.hpp, #921).	grammar_bounds_test.cpp
handler_write_alloc	✅ pass	0.00s	#1505 — a HANDLER write allocates ONE block fewer per write than a retaining write, at EVERY link count. The non-retaining role is never the more expensive one.	handler_write_alloc_test.cpp
httpd_ws_admission	✅ pass	0.00s	—	—
httpd_ws_adopted_host	✅ pass	0.00s	—	—
httpd_ws_auth	✅ pass	2.40s	—	—
httpd_ws_close_peer	✅ pass	0.00s	—	—
httpd_ws_counters	✅ pass	0.00s	—	—
httpd_ws_departure_cost	✅ pass	0.01s	—	—
httpd_ws_eviction_lock	✅ pass	0.18s	—	—
httpd_ws_fanout_alloc	✅ pass	0.00s	—	—
httpd_ws_fanout_wait	✅ pass	0.42s	—	—
httpd_ws_fanout_width	✅ pass	0.00s	—	—
httpd_ws_in_call_reply	✅ pass	0.00s	—	—
httpd_ws_keepalive	✅ pass	0.00s	—	—
httpd_ws_send_stall	✅ pass	0.00s	—	—
httpd_ws_session_identity	✅ pass	0.00s	—	—
httpd_ws_teardown	✅ pass	2.12s	—	—
httpd_ws_tx_pool	✅ pass	1.03s	—	—
link_liveness	✅ pass	0.00s	RFC-0014 §4 S5 (#492) — the link-liveness ENGINE drives link_state_t through its DIAL state machine: dormant creation, auto-wake dial on demand (connect_timeout-bounded), self-heal with backoff while a standing binding holds, fail-fast on RECONNECTING, and close-to-dormant on the last release.	link_liveness_test.cpp
link_token_carry	✅ pass	0.00s	#1266 / #1417 — the CARRIED link token, and the four things about it that have to be BUILT and SHOWN rather than asserted.	link_token_carry_test.cpp
lkv_slot	✅ pass	0.03s	ADR-0069 — the LKV slot policies, both of them, in whatever build this is.	lkv_slot_test.cpp
mount_add_oom	✅ pass	0.00s	add_child under a REFUSING allocator: a registry that cannot grow registers nothing, says so, and leaves no ghost child behind (#523).	mount_add_oom_test.cpp
mount_shape_flip	✅ pass	0.02s	#882 — a shape-flipping rebind must never let a forward route a BUS link point-to-point.	mount_shape_flip_test.cpp
mount_width	✅ pass	0.00s	The mount WIDTH lift: a mount of any width registers, resolves, and resolves the SAME way on the FWD plane and the COMPACT plane (#523, #765).	mount_width_test.cpp
net_lock_order	✅ pass	0.20s	RFC-0014 S6 (#492) — the control-plane lock-order invariant: transport_vertex_t never holds ctl_m_ across a call that can re-enter it.	net_lock_order_test.cpp
nonretaining_contract	✅ pass	0.01s	RFC-0008 Amendment 2 (#1506) — the non-retaining-vertex contract, five vectors.	nonretaining_contract_test.cpp
orchestrate_depart	✅ pass	0.70s	#491 / ADR-0073 Consequences — third-party orchestration end-to-end: an orchestrator C wires a flow between two OTHER nodes (B producer → A consumer), then departs; delivery must continue after the departure.	orchestrate_depart_test.cpp
peer_handle_seam	✅ pass	0.31s	#1294 — the peer-receiver seam carries an opaque per-peer HANDLE, not a name string.	peer_handle_seam_test.cpp
plane_isolation	✅ pass	0.00s	#1460 phase (b) — RFC-0025 conformance vector 12 stream/receiver-ring-flood, PLANE-ISOLATION arm: a flood that exhausts the NET-plane store must leave the GRAPH plane still allocating, under the flood rather than after it.	plane_isolation_test.cpp
pool_only_dispatch	✅ pass	0.00s	The SINGLE-MEMBER (LIBTRACER_BACKEND_SET_POOL_ONLY) module-set destroy dispatch (#922, ADR-0047 §2): the folded fast path must still be a fast path, not a correctness dependency.	pool_only_dispatch_test.cpp
propagate_fold	✅ pass	0.00s	RFC-0025 §4.1.2 (Amendment 3, clause 5) — propagate’s FOLD emission mode.	propagate_fold_test.cpp
qos_policy	✅ pass	0.00s	RFC-0022 (as amended): delivery policy is per-subscription; settings_t dissolves.	qos_policy_test.cpp
read_fork	✅ pass	0.06s	The leaf/branch fork of graph_t::read, and the state machine that now answers it without the map lock (#652).	read_fork_test.cpp
reclaim	✅ pass	0.00s	The ADR-0080 reclamation seam — WHEN a retired subscription’s {fn, ctx} pair becomes safe to free (#894), asserted against the policy this build bound.	reclaim_test.cpp
registry_rebind_race	✅ pass	0.00s	#684 — a rebind of a LIVE name must not touch the published mount encoding.	registry_rebind_race_test.cpp
ring_pressure	✅ pass	0.00s	#1460 phases (a) and (c) — RFC-0025 conformance vector 12 stream/receiver-ring-flood, the RING-PRESSURE and RECOVERY arms: at exhaustion the receiver’s ring must decline as a POLICY OUTCOME under §4.4, account every loss, and recover exactly when the pressure does.	ring_pressure_test.cpp
router_drop_stats	✅ pass	0.00s	#1503 step 3 — fwd_router_t exposes its four injected seams, and its cold-path drops are COUNTED instead of silent.	router_drop_stats_test.cpp
send_stall	✅ pass	2.82s	#838 — a host stream send into a stalled peer is BOUNDED, counted, and the peer that keeps stalling is closed rather than blocked on forever.	send_stall_test.cpp
session_anchor	✅ pass	0.20s	#1223 step 2 — an ACCEPTED slot_server_t session holds an identity anchor: a vertex-map slot with a saturating generation, revived in place on slot reuse.	session_anchor_test.cpp
spin_pool_guard	✅ pass	0.30s	—	—
stats_field	✅ pass	0.00s	#1503 step 5 / RFC-0010 Amendment 1 — the node-scoped seam census: read <vertex>:stats.<class>.<seam>.	stats_field_test.cpp
subscribe_toward	✅ pass	0.00s	#739 — fwd_router_t::subscribe_toward: bind a local producer toward a mount-path target through the shared strip-K descent (ADR-0061).	subscribe_toward_test.cpp
target_binding_ablation	✅ pass	0.00s	The #830 ablation: target_canonical_resolves() proves WHICH of dispatch_edge_target’s two resolve spellings a local target edge actually took.	target_binding_ablation_test.cpp
terminus_egress_backend	✅ pass	0.00s	#795 — the terminus FWD{REPLY}’s egress-construction segments (the reply HEAD and, on a mint, the trailing PATH_REF) draw from the router’s injected egress backend, not the global heap — and a refusal degrades by value, never an abort.	terminus_egress_backend_test.cpp
terminus_flatten_backend	✅ pass	0.00s	#766 — the TERMINUS resolver’s rope-tier flattens draw from the router’s injected mem_backend_t too, and a refusal is answered by value rather than by reading a short span.	terminus_flatten_backend_test.cpp
test_support	✅ pass	0.19s	The shared test runner, under test by itself (#874).	test_support_test.cpp
transport_alloc_softfail	✅ pass	6.24s	#848 — the transport EGRESS framing path never aborts on a tight heap: a refusing allocator must produce a DROPPED FRAME AND A LIVE NODE, never abort().	transport_alloc_softfail_test.cpp
try_grow_race	✅ pass	0.03s	The nothrow tr::detail::try_* growth helpers under a REFUSING allocator (#923, and #850 folded into it): an allocation failure must come back as false, never as a bad_alloc crossing their noexcept boundary into std::terminate.	try_grow_race_test.cpp
try_probe_window	✅ pass	0.00s	#981 — the two try_* call sites that MIGRATED off the -fno-exceptions probe window really draw from the injected block_source_t, and answer its exhaustion by value.	try_probe_window_test.cpp
twai_backpressure	✅ pass	0.41s	—	—
twai_rx_ingress	✅ pass	0.71s	—	—
write_ctx	✅ pass	0.00s	#375 — a HANDLER’s on_write receives the writer’s SUBJECT (tr::graph::write_ctx_t).	write_ctx_test.cpp
write_fault_class	✅ pass	0.00s	#948 — a malformed send/sendmsg call is NOT a dead socket, and must not silently swallow the rest of the frame.	write_fault_class_test.cpp

Warning

No source could be resolved from the CMake build graph for: bus_can_guard, esp_ws_client_dial, esp_ws_client_liveness, esp_ws_client_recv, esp_ws_client_teardown, example_acl_everyone_reserved, example_acl_expiry, example_acl_inherit, example_acl_open_by_default, example_acl_parse_strict, example_acl_policy_profiles, example_acl_right_bits, example_acl_subject_resolver, example_graph_await, example_graph_children, example_graph_handler_vertex, example_graph_read_write, example_graph_register, example_graph_retire, example_graph_stream, example_graph_write_creates, example_mem_block_array, example_mem_block_source, example_mem_bump_source, example_mem_bump_upstream, example_mem_pool_source, example_mem_size_classes, example_mem_source_resource, example_mem_source_vs_backend, example_net_can_bus_peers, example_net_dial_and_listen, example_net_kind_catalog, example_net_multi_peer_listener, example_net_tcp_stream_framing, example_net_transport_seam, example_net_udp_datagram, example_net_ws_upgrade, example_route_child_table, example_route_dst_is_source_route, example_route_label_compact, example_route_label_stale, example_route_multi_hop, example_route_qualified_mount, example_route_reply_home, example_route_terminus_or_forward, example_sub_callback, example_sub_durability_latch, example_sub_retire, example_sub_subtree, example_sub_terminal_delivery, example_sub_unsubscribe, example_sub_unsubscribe_from_dispatch, example_view_borrow, example_view_device_rope, example_view_pool_backend, example_view_rope_compose, example_view_rope_subrope, example_view_segment_refcount, example_view_subview, example_view_sync_pool, example_wire_arena_decode, example_wire_decode_refusals, example_wire_lazy_view, example_wire_packed_path, example_wire_path_escape, example_wire_structured_vs_opaque, example_wire_tlv_header, example_wire_trailer, httpd_ws_admission, httpd_ws_adopted_host, httpd_ws_auth, httpd_ws_close_peer, httpd_ws_counters, httpd_ws_departure_cost, httpd_ws_eviction_lock, httpd_ws_fanout_alloc, httpd_ws_fanout_wait, httpd_ws_fanout_width, httpd_ws_in_call_reply, httpd_ws_keepalive, httpd_ws_send_stall, httpd_ws_session_identity, httpd_ws_teardown, httpd_ws_tx_pool, spin_pool_guard, twai_backpressure, twai_rx_ingress. Those rows fall back to a hand-written description, which is exactly the drift this resolution exists to remove.

How every suite is verified

Beyond this Release pass, the same suites run under three more configurations in CI (core-ci.yml), and the net forward path carries an absolute allocation gate:

configuration	what it proves
Release (this page)	functional correctness, byte-exact wire behavior
ASan + UBSan	no leaks, no undefined behavior, no buffer overruns
TSan	the lock-free LKV + concurrent forward paths are race-free
GCC-13 + GCC-15	the toolchain floor + the ESP on-silicon compiler
16KB zero-heap gate	the FWD forward hop allocates 0 bytes (bench_forward_heap, ZEROHEAP_MAX=0; ADR-0038/0039)

Cross-implementation conformance (C++ / TypeScript / Rust agree on every vector) and the live latency/throughput numbers are on the Performance page.