open-standards/wai/live-conformance/metering

live-conformance/metering — live metering

The corpus of extensions/live-metering.md §10: rooms, and vectors, each a room-session receipt (WMR1) with what it names (the per-hop session records, WMH1, of version 1 or of version 2 with a hop's DVR figures, each with its ledger; the composition descriptors and receipts, WCD1 and WCR1; the take manifests, WTM1; and the logs document of QoE observations), and the outcome a verifier gives it.

Metering equivalence. For each vector, a conforming receipt verifier given its room (by name, from rooms: the descriptor, the grants, the key-log head and the chains, each hexadecimal), its receipt, hops (each record and ledger), descriptors, compositions, takes and logs (the document's text, or null) reaches the outcome the vector states (§5.4):

• refuse: the reason and where, at: ["room"] or ["room", …] with the room verifier's own place (live-rooms §9.1), ["receipt"], ["hop", i] for the receipt's hop entry i, ["take", i], ["compose"] for the list of composition receipts or ["compose", i] for the receipt i in ascending order of digest, ["qoe"], ["units"], ["energy"] or ["ceilings"];
• accept: the receipt's digest; the number of media tracks the room published in the session (media_tracks, the M of §3.7); each hop record's digest, size, size bound (B(M) for version 1, B₂(M) for version 2) and version, and for a version 2 record its six DVR figures by their §3.9 names (dvr_objects_out, dvr_bytes_out, dvr_objects_stored, dvr_bytes_stored, dvr_byte_s, dvr_dropped); the units; the unmetered hop entries; the totals, one per class; the coverage, whole or partial; each grant's consumption entry (its ceilings, its units, its attributed joules by class, whether its share is unmetered, its funds, its two verdicts and its leaf); the consumption_root; the grants past a ceiling with their flags; and the qoe_cell.

A verifier that reaches another outcome, reason, place or fact does not conform.

Provenance. Generated here: wai_live_metering_vectors gen writes vectors.json from sessions authored in code, from fixed seeds, by capture binding's corpus rules: each key is SHA-256 of that corpus's fixed label and the principal's index (the principals it shares with that corpus and with live recording's are the same principals), every time is a constant (the lesson room's t0 is 2026-10-16T00:00:00Z), every chunk's bytes are SHA-256 output under a fixed label, and Ed25519 signing is deterministic (RFC 8032), so the file is the same bytes on every machine. The lesson's per-hop records are aggregated from sessions written in code by the rules a hop counts by (§3), and stand for what a relay, a compositor and an ingest gateway would count; their joule figures are constants and were measured by nothing. The dvr_relay vector's DVR figures are counted, by §3.9's rules, from a DVR window written in code (the composite's 23 frames of 20 000 bytes, each written as it arrives a third of a second into its second and held ten seconds or to the span's end, one dropped a second after it arrived, a viewer joining at 20 s filled from the window): measured by nothing either. The key-log heads are sealed by the reference's key log under a fixed root; the origin is an .example name (RFC 2606). The generator asserts that each vector reaches the outcome it was written to have before it writes it.

Recorded data, one record. relay-record.wmh and its ledger relay-ledger.wml were signed by joule-moq-relay itself, in its test tests/room_metering.rs (two publishers and three subscribers of the relay room over native QUIC and WebTransport, one subscriber slow, each reading its streams concurrently), run once on 2026-10-18 with WAI_METERING_FIXTURE_DIR set; the relay's key is the room's clock key, under the same label. Its counts, its span (3.196 s of room clock from 11:45:35.966 UTC) and its subscriber time are what that run measured on loopback, so a second run gives other bytes: the pair is committed as recorded and read by the generator, which writes the relay room's lifecycle around the record's span. The relay meters no energy, and the record says so with the unmetered marker. The relay room's grants run thirteen days from its t0, so a pair recorded after 2026-10-31 would fall outside them.

Layout

One file, vectors.json, beside the recorded pair:

• principals: the keys, by name (informative).
• rooms: lesson, a lesson room (a director with the record power, a performer's device and an accompanist's capturing two takes, a compositor laying the room out at one frame a second, three viewers, a key log listing the room authority, the store, the relay, the compositor, an ingest gateway and a metering key, and a revoked metering key and relay key); lesson_ceilings, the same with the second viewer's joule ceiling at 1 000 µJ and the third viewer's funds ceiling at 0; lesson_gap, the clock's join of the third viewer not delivered; lesson_bad_chat, the viewer's chat carrying a bad signature; and relay, the reference relay's test room, with no origin.
• vectors: each with name, about, room, receipt, hops, descriptors, compositions, takes, logs, outcome, and as above reason and at, or the facts.

The vectors:

• accepted: lesson (a relay and a compositor read from an on-chip counter, two takes, ten composition receipts, three viewers' logs and a suppressed QoE cell, under a tariff in euros; signed by the metering key), signed_by_authority, mixed_classes (a relay's counter reading, a compositor's model and an ingest gateway's estimate, one total per class), ceilings_past (the second viewer past its joule ceiling, the third past its funds ceiling), large_audience (ten thousand more subscriber sessions with no grant: the relay's record is the lesson's size; sixty logs publish the QoE cell), relay_recorded (the reference relay's record, unmetered, so the energy is partial) and dvr_relay (the lesson's relay keeping a ten-second DVR window of the composite: a version 2 record, 48 bytes longer than the lesson relay's under a bound 48 bytes higher, stating the frames it stored, the byte-seconds it held them, one dropped for not verifying against its receipt, and the ten it filled a late viewer with; every other record of the corpus is version 1);
• refused, one defect each: the receipt (receipt_bad_signature, receipt_truncated, receipt_wrong_room, session_invalid, signer_not_authorised, signer_revoked, total_of_no_class); the hop records (record_missing, record_too_large, record_invalid, energy_invalid, record_wrong_room, record_bad_signature, entry_restated_wrong, hop_not_authorised, hop_class_not_authorised, hop_revoked, span_outside_session, track_unknown, ledger_invalid, ledger_ceiling_mismatch, ledger_root_mismatch, hop_units_mismatch, and for version 2 dvr_out_exceeds and dvr_stored_exceeds, its bytes sent from its store above its out entries' and its objects stored above its in entries' (record_invalid), record_unknown_version, the DVR relay's record with its fifth byte 3 and signed again (unknown_version), and dvr_record_too_large, past B₂(M) by less than one entry (record_too_large)); the takes, compositions and logs (take_missing, take_unknown, composition_missing, composition_refused, composition_outside_session, logs_missing, qoe_invalid, qoe_not_suppressed); the units, energy and ceilings (units_mismatch, recorded_mismatch, classes_summed, total_mismatch, ceiling_mismatch, consumption_root_mismatch); and the room (room_refused, lifecycle_incomplete).

Running

cd ../../wai-rs
cargo run --no-default-features --features live_metering --bin wai_live_metering_vectors -- verify ../live-conformance/metering
cargo run --no-default-features --features live_metering --bin wai_live_metering_vectors -- gen    ../live-conformance/metering   # regenerate
cargo test --no-default-features --features live_metering --test live_metering
cd ../live-conformance
python3 live_metering_verify.py metering

wai_live_metering_vectors verify and tests/live_metering.rs require the file to regenerate byte for byte and wai-rs to give every vector its outcome; the test also requires every reason of §9 to refuse a vector, the accepted vectors to be the ones above, the large audience's relay record to be the lesson's size within the same bound, the DVR relay's record to be version 2 with every figure above 0 and every other accepted record version 1, and §9 to list exactly the reasons the reference reports. live_metering_verify.py is the second implementation, written from the extension's text and sharing no code with wai-rs: it verifies each room with live_rooms_verify.py, each take with capture_take_verify.py and each composition receipt with live_compose_verify.py, recomputes every unit, every total, every grant's consumption and the QoE cell itself, gives every vector the same outcome, requires the same coverage, and then mutates copies of the corpus, one defect each, each of which must fail. Pass = an exit status of 0 from each.

To record the relay's pair again (it changes the corpus), run the relay's test with WAI_METERING_FIXTURE_DIR set to this directory, then gen:

cd ../../../joule-code/joule-code-rs
WAI_METERING_FIXTURE_DIR=$PWD/../../wai/live-conformance/metering cargo test -p joule-moq-relay --test room_metering