AT&T measurement catalogue

Every measurement_id the rXg produces, every protobuf ID the webhook accepts, and the gap between the two.

Scope: this is the off-rXg contract only. Everything below concerns what leaves the rXg for AT&T. The same ingested data separately drives the rXg's own view of its fleet — AP online/offline state, client counts, radio state, health notices — through entirely different code, and that on-rXg path uses none of these protobuf IDs. See §1 of the project context before concluding that a metric "is not collected" because it is absent here.

How this was built. Producer IDs extracted from every measurement_id: literal under console/app/lib/telemetry/{cisco,ruckus} on rgnets/rxg@32c0dfea95 (branch hbh/kpi-update). Protobuf IDs extracted from ATT_PROTOBUF_MEASUREMENT_TYPES in notifications.yaml on rgnets/att-templates@0e44313f27 (branch hbh/combined-new-schema). The three tables below are set operations over those two lists, so they are exhaustive as of 2026-09-17 — not a curated selection.

Set	Count	Meaning
Map entries	71	Keys in `ATT_PROTOBUF_MEASUREMENT_TYPES` (69 distinct ints — two are double-assigned).
Delivered	31	Produced by a vendor and mapped. This is what AT&T actually receives.
Dropped in transit	12	Produced by a vendor, absent from the map, silently discarded by the webhook's `select`.
Reserved	40	Mapped, but no code emits them. Aspirational or awaiting a producer.

1 Delivered — reaches AT&T today

Accessor columns name the method on Telemetry::Cisco::Grpc::Grpc / Telemetry::Ruckus::Mqtt::Mqtt that supplies the value, so you can trace any number back to its computation. Radio metrics carry source_subtype 0 and a band source_subcategory; switch and port metrics carry source_subtype 1.

ID	measurement_id	Cisco	Ruckus	Backing accessor	Notes
Radio — phase 1 — source_subtype 0, extra = radio slot ID
111	client_count_per_ap	Y	Y	client_counts	Distinct client MACs per radio. Ruckus gates the whole client-derived block on this being non-zero.
112	channel_utilization_avg	Y	Y	channel_utilizations	Ruckus source field is `airtime`.
113	throughput_per_client_avg	Y	Y	ap_to_client_throughput_avgs	Downstream only, despite the name. Webhook divides by 10⁶ → Mbps.
114	aggregate_throughput_avg	Y	Y	ap_to_client_ + client_to_ap_throughput_avgs	Sum of both directions. Webhook divides by 10⁶ → Mbps.
115	rssi_avg	Y	Y	rssi_avgs	Ruckus reads `receiveSignalStrength`. See the SNR/RSSI caveat below.
116	snr_avg	Y	Y	snr_avgs	Ruckus reads `client["rssi"]` — the vendor's field names are crossed. See caveat.
117	mcs_avg	Y	Y	mcs_avgs	Derived, not reported: `StatsTask::McsComputation` back-solves an index from rate + PHY + NSS + width. Computed last because it depends on the others.
122	noise_avg	Y	Y	noise_floors	Cisco falls back to a second source: `noise_floors \|\| noise_floors2`.
123	radio_band_primary_channel	Y	Y	current_channels
124	radio_channel_width	Y	Y	channel_width_avgs	An average width is sent, not the configured one.
126	radio_power_level	Y	Y	tx_powers	Shares 126 with `radio_gain` — deliberate, commented in the map as "Map radio_gain to RADIO_POWER_LEVEL".
126	radio_gain	Y	Y	gains	Same int as above. AT&T cannot distinguish the two without `measurement_id`.
Switch — phase 2 — source_subtype 1, switch-level, Cisco only
4000	switch_version	Y	—	svalue: version	String metric — travels in `svalue`.
4001	switch_poe_budget	Y	—	poe_budget	Collides with the reserved `temperature` key, also 4001. Almost certainly meant 4002.
4003	switch_poe_usage	Y	—	poe_usage
4005	switch_stp_mode	Y	—	stp_mode	e.g. `stp-mode-rapid-pvst`.
4006	switch_routes	Y	—	routes	List metric — repeated `Route` message in the proto.
4007	switch_vlan_stp_states	Y	—	vlan_stp_states	Repeated `VlanStpState`. The webhook's `map_stp_state` helper folds blk/lis/lrn/fwd/dis into the port-status enum.
4008	switch_vty_lines	Y	—	vty_lines	Repeated `VtyLine`.
4009	switch_cpu_load	Y	—	cpu_loads
4010	switch_memory_usage	Y	—	memory_usage
Port — phase 2 — source_subtype 1, extra = port BIA MAC, Cisco only
5000	port_rx_packets	Y	—	unicast_pkts_ins + multicast + broadcast	Summed in Ruby from three separate counters.
5001	port_tx_packets	Y	—	unicast_pkts_outs + multicast + broadcast	Likewise.
5002	port_rx_discards	Y	—	discards_ins
5003	port_tx_discards	Y	—	discards_outs
5004	port_duplex	Y	—	duplexes
5005	port_uptime	Y	—	port_uptimes
5006	port_poe_usage	Y	—	port_poe_usages
5007	port_tagged_vlans	Y	—	svalue: tagged_vlans	String metric. Keyed with `bia_address: nil`, unlike its neighbours.
5008	port_access_vlan	Y	—	access_vlans	Same nil-BIA keying.
5009	port_native_vlan	Y	—	native_vlans	Same nil-BIA keying.

2 Dropped in transit — produced, then discarded

The rXg computes all twelve of these and hands them to the webhook, which drops them because ATT_PROTOBUF_MEASUREMENT_TYPES has no matching key. No log line, no error — the select just narrows the array. Note that this includes every byte counter and every port speed/status value, three of which schema_notes.md documents as already having assigned protobuf IDs from the old scheme.

measurement_id	Cisco	Ruckus	Why it is missing
switch_temperature	Y	—	Pure name mismatch. The map's key is `temperature`; the producer emits `switch_temperature`. One string away from working.
bytes_rx	Y	—	schema_notes.md assigns `RX_BYTES = 108`. Never added to the map.
bytes_tx	Y	—	`TX_BYTES = 109`. Same.
speed	Y	—	`IF_SPEED = 87`. Same.
connection_status	Y	—	`IF_OPER_STATUS = 83`. Same. (Producer name differs from the notes' `status`.)
radio_channel_list	Y	Y	Map entry exists but is commented out: `#'radio_channel_list' => 125`. Cisco sends it as `svalue`.
ap_to_client_total_transfer	Y	Y	Map has a placeholder `#'…' => ?` — protobuf ID never assigned.
client_to_ap_total_transfer	Y	Y	Same placeholder.
ap_to_client_total_throughput	Y	Y	Same placeholder. Computed as transfer × 8 ÷ window.
client_to_ap_total_throughput	Y	Y	Same placeholder.
radio_oper_state	Y	—	No protobuf ID proposed anywhere. Cisco-only, sent as `svalue`.
device_uptime	Y	—	Intentional. schema_notes.md: "internal use only in rXg, not mapped to a protobuf ID".

The four *_total_transfer / *_total_throughput elements are the volumetric ones — they are also the metrics that the missing SUM() in the continuous aggregates would affect. Assigning them protobuf IDs and fixing the rollup are the same piece of work viewed from two ends.

3 Reserved — mapped, no producer

40 keys in the map that no code emits. These are commitments, not capabilities. schema_notes.md is candid about it, marking both large blocks "not yet implemented in the rXg repository".

Gateway, 3000–3020 20 IDs

Whole-rXg metrics: no per-device source_id. schema_notes.md proposes source_subtype 14 (GATEWAY) but flags it uncertain. 3002 is unassigned. att-templates!20 (ldd/gateway_metrics) is the intended producer, and it targets hbh/combined-new-schema rather than master.

Group	IDs	Keys
DHCP	3000–3001	gateway_dhcp_succ_num, gateway_dhcp_fail_num
Social-media tonnage	3003–3004	gateway_soc_media_tonnage_ds, …_us
Total tonnage	3005–3006	gateway_total_tonnage_dl, …_ul
Passpoint	3007–3009	gateway_passpt_tonnage_dl, …_ul, gateway_passpt_clients
Non-Passpoint	3010–3012	gateway_non_passpoint_tonnage_dl, …_ul, gateway_non_passpoint_clients
AAA (internal)	3013–3016	gateway_aaa_auth_time_avg, …_time_max, …_succ_num, …_fail_num
AAA (external)	3017–3020	gateway_external_aaa_auth_time_avg, …_time_max, …_succ_num, …_fail_num

The notes' original names were period-scoped (PASSPT_TONNAGE_24HR / 30DAY / 365DAY); the map replaced those with direction-scoped names (_dl / _ul) and no period. Whoever implements the producer has to settle what window a gateway tonnage value covers — the webhook asserts collection_interval: 300 for everything it sends.

WLAN sensor, 2000–2016 17 IDs

The absorbed phase-1 radio IDs. schema_notes.md lists PACKET_RETRANSMIT_AVG 118, LATENCY_AVG 119, JITTER_AVG 120 and PACKET_LOSS_AVG 121 as radio measurements, but none of 118–121 appear in the map — they were re-scoped into this sensor block instead. Worth confirming with AT&T that 118–121 are genuinely retired. We do have the underlying data: sensor ping jitter/loss/RTT already exist in the rXg as SensorMetric.

Group	IDs	Keys
Link quality	2000–2004	wlan_sensor_packet_retransmit_avg, …_latency_avg, …_jitter_avg, …_packet_loss_avg, …_mos_score
Onboarding	2005–2007	wlan_sensor_dhcp_time, …_capport_time_avg, …_capport_time_max
Services	2008–2010	wlan_sensor_dns_average, …_avg_download, …_avg_upload
Client attributes	2011–2016	wlan_sensor_ssid, …_rssi, …_client_to_ap_mcs, …_primary_channel, …_channel_width, …_phyrate

No AAA block here, although the notes ask for five WLAN_SENSOR_AAA_* measurements. Those look to have become the gateway 3013–3020 range.

Three orphans

ID	Key	Status
4001	temperature	Name mismatch with the producer's `switch_temperature` and an int collision with `switch_poe_budget`. Two bugs in one entry.
4004	switch_description	Notes say "intended but not currently implemented in rXg".
5010	port_description	A `descriptions` KPI exists on the Cisco class but is excluded from `PORT_KPIS_FOR_REPORT`, so nothing emits it.

4 Identity fields

How AT&T tells one measurement apart from another. Set once per element by the producer, then stamped by the webhook.

Field	Set by	Value
measurement_id	producer	Snake-case name; the webhook rewrites underscores to spaces before sending.
measurement_type	webhook	The protobuf int from the map.
source_id	producer	AP ethernet MAC, or switch primary MAC. `source_id_type 1` = MAC.
extra	producer	Radio slot ID (AP) or port BIA MAC (switch).
source_subtype	producer	`0` AP, `1` switch. The proto enum also defines CONTROLLER 2, UPS 3, ONT 4, CELL 5, SERVER 6, CLIENT 7, DB 8, APPLIANCE 9, LB 10, GSLB 11, SAAS 12, HA 13, GATEWAY 14 — none of which we emit yet.
source_subcategory	producer	Band: `0` 2.4G, `1` 5G, `2` 6G. Radio metrics only; `nil` elsewhere.
reporter_id	webhook	The rXg's own FQDN, from `DeviceOption.active.domain_name`. Sent with `reporter_id_type 1` (MAC) even though it is a domain name.
customer_id / venue_id	webhook	From the `att_venue_info` CustomDataSet. If that record is absent, both silently become `""`.
collection_interval	webhook	Hard-coded `300` for every measurement, with the comment "time in seconds?". The Ruckus report window is actually 15 minutes.
timestamp	either	Producer's value if present, else `Time.now.to_i` at send time.

5 Caveats to carry into any analysis

Ruckus SNR and RSSI come from crossed vendor fields. EXTRACT_SNR reads client["rssi"] and EXTRACT_RSSI reads client["receiveSignalStrength"], annotated in the source as "ruckus messed this up on their end". IDs 115 and 116 are therefore not comparable between a Cisco venue and a Ruckus venue without knowing this.
mcs_avg is inferred, not measured. Both vendors back-solve an MCS index from rate, PHY, NSS and channel width. Any gap in the PHY or width tables yields no value at all rather than a wrong one — which is the safer failure, but means missing data is expected for new PHYs.
Ruckus client-derived metrics vanish when a radio is idle. to_kpi_report only appends SNR, RSSI, MCS and the throughput family when client_count.nonzero?. An idle radio still reports noise, utilisation, channel and power — so a missing 115 means "no clients", not "no data".
Radio metrics are per radio; the ID says "per AP". client_count_per_ap (111) is emitted once per radio with the slot in extra. Summing it across an AP is the caller's job.
Nothing here comes from OpenWiFi. Our own APs feed ClientMetric / RadioMetric / ThroughputMetric and never reach this webhook. A venue on our own hardware reports nothing to AT&T today.
Ruckus values are bounded by a 20-minute cache TTL on the payload body, which is the only place the raw JSON exists. See the project context page.

Generated 2026-09-17 from the SHAs named at the top. Re-run the extraction before trusting the counts after either branch moves. ← Project context