Video Quality Operations

Nishi vs the field — every Nishi cell is measured against real organ source at emit time; each gap names the watch contract that will close it.

How production systems MONITOR, SPOT, ADDRESS and PREVENT video quality issues -- Nishi vs Netflix and YouTube and Zoom and Twitch

Layer 1 · Executive

Where we are. Measured 2026-08-19. The four pillars are present on the sovereign stack -- QoE telemetry, perceptual judge, artifact detection, faithfulness probes, recon-hash drift, cross-instance farm, ABR, PLI, fallback, self-upgrade, relay backpressure, negotiation, conformance gates, fuzzing and flag-gated ship -- with the integer-deterministic codec as the exceed. The one symbol gap is the ops dashboard; three watch contracts are admitted here for the rows where Netflix and Zoom lead: percentage canary, a full-reference metric and a real-device matrix.

Where we need to go. See every session, control rollout by percentage with automatic halt, and measure like the majors with a full-reference metric and a real-device matrix -- all on the sovereign loop.

The unit. 1 u = one measured session-leg. Calibration from landed rungs: the mangagen panel compositor went from existing substrate to shipped and live-verified in ONE leg (2026-08-13); the citations rung went from 3 to 55 domains in one leg across seven seats (2026-08-18); a greenfield engine with a bite-proven gate has measured 2 to 4 legs. Estimates recalibrate as rungs land and PR7 actuals write back.

Cost to see sessions: 1.5 u. Through M0.

Cost to control rollout: 2.5 u. Through M1.

Cost to measure like the majors: 6.5 u. Everything below.

14 of 21 capabilities measured|2 of them measured exceeds|7 open|coverage 666/1000|adoption 4 full / 10 partial

Layer 2 · Roadmap

Do this next — computed by the ranker, never chosen by a seat

Order from nx_compare_rank (nx_dr_ocm: (deficit + cost-of-delay + option + enables) x sponsor x self-sufficiency x momentum / cost). FINISH rows are rungs whose symbol is present but whose organ is short of full adoption: the cheapest closures on this board, listed before any new work. Stamp: # asof=1787883506 domain=videoops target_version=0.1 rungs=4 done=0 open=4 finish=0 ranker=nx_dr_ocm

#	Stage	Rung	Priority	Derivation
#1	0.1	Live ops dashboard (R0) vo_dashboard	2800	v=14 m=1 c=5
#2	later	Percentage canary (R1) vo_canary_pct	3000	v=9 m=1 c=3
#3	later	Real-device matrix (R3) vo_device_matrix	1500	v=9 m=1 c=6
#4	later	Full-reference perceptual metric (R2) ns_fr_vmaf	450	v=9 m=1 c=20

Critical path — contract, done-rule, executor, cost

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Live ops dashboard (R0)	vo_dashboard	Per-session QoE beacons (fps, bytes, mode, drift) aggregated on the beat into a served page composing nx_netobs series; a session in trouble is visible before a user reports it	Organ	1.5 u
Percentage canary (R1) after R0	vo_canary_pct	A build ships to a declared percentage of sessions and halts automatically on a drift or fps regression measured by the dashboard; the halt is a recorded event	Organ	1 u
Full-reference perceptual metric (R2)	ns_fr_vmaf	A VMAF-class full-reference score against the source frame computed in integer; correlation with the banked subjective set PRE-DECLARED; published beside the no-reference judge	Organ	2 u
Real-device matrix (R3) after R0	vo_device_matrix	Conformance results from real phones and browsers in the farm as a matrix with every cell a link to its recon-hash verdict	Organ	2 u

Milestones

Milestone	Rungs	Cumulative
M0 · See sessions	R0	1.5 u
M1 · Control rollout	R1	2.5 u
M2 · Measure like the majors	R2,R3	6.5 u

Layer 3 · Engineering

How this is scored. Every Nishi mark is measured: the generator reads the real organ source on disk and requires the implementing symbol to exist (no self-grading). A watching tag names the organ and symbol contracted to close a gap — the mark flips itself on the next compare beat when that workstream ships, and the comparewatch- plane row flips with it. The flip is necessary, not sufficient: it proves the symbol exists, never that the capability is good. The bar is the rung's pre-declared done-rule, proven by its gate — a symbol shipped without the behaviour behind it is a defect, and the flip is exactly what makes that defect visible instead of quiet. Competitor marks record documented capability presence — presence, not depth or scale. Adoption is measured too: every measured row carries where its organ stands on the estate's ladder (source → built → promoted → registered → invoked; libraries by importer reach minus validation importers; gates by the execution surfaces that run them). A row is fully adopted only at the top of its ladder; anything short is tagged partial with the exact remedy, so a build nobody promoted can no longer read as shipped. Census stamps: importers asof 1787849099, gate census asof 1787855507 (unix seconds; -1 = census absent).

Capability matrix — measured against source

◉ leads / measured exceed● present◐ partial○ absent · click any capability for its evidence

Capability	Nishi	Netflix	YouTube	Zoom	Twitch
MONITOR
per-session QoE telemetry Open — no implementing organ is measured for this axis yet. Netflix device telemetry and YouTube playback QoE run at planetary scale; Nishi beacon carries fps/bytes/mode	○	◉	◉	●	●
perceptual quality metric Measured: `ns_mscn_rho` exists in `runtime/nx_natstat.nx`, verified at emit. Netflix invented VMAF [vmaf16]; YouTube UVQ [uvq22]; Nishi nx_natstat 5-axis integer IQA judges output realism Adoption: `LIB-WIRED importers=9 nonval=5` — fully adopted (top of its ladder).	●	◉	●	○	◐
artifact detection (blockiness) Measured: `blki` exists in `runtime/_hdl_build/nx_vcodec_t8noise_gate.nx`, verified at emit. 8-grid blockiness index in the governing codec gate; majors run artifact QC in pipeline Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder).	●	●	●	●	●
live ops dashboard for sessions Open — watching `runtime/nx_video_ops.nx` : `vo_dashboard`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Netflix Atlas and Zoom dashboard class [atlas14]; Nishi has no ops surface yet watching vo_dashboard	○	◉	◉	◉	●
SPOT
decoder-faithfulness probes Open — no implementing organ is measured for this axis yet. Silent-corruption spot checks; Nishi faithdiff proves decoder==reference on live streams	○	●	●	●	●
cross-instance recon-hash drift detection Measured: `vc_drift_verify` exists in `runtime/nx_video_codec_wasm.nx`, verified at emit. SOVEREIGN: hash (vc_recon_hash) + matcher/verdict (vc_drift_) live in the codec wasm -- one drift brain for browser tab, Nishi browser, and native NishiOS app; JS only routes bytes Adoption:* `LIB-WIRED importers=62 nonval=17` — fully adopted (top of its ladder).	●	●	●	●	●
device/browser conformance matrix Measured: `mkInstance` exists in `../sites/nishifamily/video/vc_xinst.js`, verified at emit. Netflix device certification is Best-in-class; Nishi cross-instance farm covers separate engines, 4 tiers, field qp — device-farm depth still theirs Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	◉	◉	●	●
ADDRESS
adaptive bitrate ladder Measured: `rungQ` exists in `../sites/nishifamily/video/app.v2.js`, verified at emit. Netflix per-title encoding is the reference [pertitle15]; Nishi rungQ adapts qp and geometry per network Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	◉	◉	●	●
keyframe request on loss (PLI-class) Measured: `function kreq` exists in `../sites/nishifamily/video/app.v2.js`, verified at emit. RTC bread-and-butter (RFC 4585) [rfc4585]; Netflix is VOD; Nishi kreq recovers in ~1 RTT Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	○	●	◉	●
auto-fallback to safe codec mode Measured: `t8Sup` exists in `../sites/nishifamily/video/app.v2.js`, verified at emit. Zoom degrades gracefully under stress; Nishi auto-drops t8 to rc on receiver-confirmed drift (795) Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	●	●	◉	●
stale-client self-upgrade Open — no implementing organ is measured for this axis yet. Forced client refresh with damper; all majors force-update clients	○	●	●	●	●
relay backpressure -- no head-of-line stall Measured: `s2tx_send` exists in `runtime/nx_sig2_txq.nx`, verified at emit. SFU congestion control; Nishi relay now frame-atomic non-blocking (nx_sig2_bp_gate 4/4) -- one slow receiver drops whole frames, never stalls the room Adoption: `SOURCE-ONLY` — PARTIAL: source exists, never compiled: /api/build it. adoption SOURCE-ONLY	●	◉	◉	◉	●
PREVENT
capability/version negotiation Measured: `p.vcv` exists in `../sites/nishifamily/video/app.v2.js`, verified at emit. Zoom/WebRTC SDP negotiation is Best [rfc3264]; Nishi vcv syntax-generation gate stops mixed-build decode Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	●	●	◉	●
codec conformance gates (real content+noise) Measured: `run_noise` exists in `runtime/_hdl_build/nx_vcodec_t8noise_gate.nx`, verified at emit. Google drives AV1 argon conformance [argon22]; Nishi runs 15+ adversarial gates on real frames plus the cross-instance farm Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder).	●	●	◉	●	●
decoder fuzz/robustness harness Measured: `mkInstance` exists in `../sites/nishifamily/video/vc_fuzz.js`, verified at emit. Google oss-fuzz is Best [ossfuzz]; Nishi fuzzes 349k hostile inputs -- decoder now bounds-hardened (MV clamp in qp_pixel path), traps 558 to 0, GREEN; sovereign fix not a JS belt Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	●	◉	●	●
staged rollout / canary Measured: `t8Flag` exists in `../sites/nishifamily/video/app.v2.js`, verified at emit. Netflix canary analysis is the reference [kayenta18]; Nishi ships flag-gated first but has no percentage rollout Adoption: `ABSENT` — PARTIAL: no source and no binary: this is a build contract, not a partial deploy. adoption ABSENT	●	◉	◉	●	●
Integer-deterministic codec -- bit-exact on every device by construction Measured exceed: `vc_mb_t8_ok` in `runtime/nx_vcodec.nx`, verified at emit. No-float integer pipeline reproduces bit-exact everywhere, making recon-hash conformance possible; float pipelines cannot Adoption: `BUILT-UNPROMOTED` — PARTIAL: compiled, never promoted to the serving root: /api/promote it. adoption BUILT-UNPROMOTED	◉	○	○	○	○
Sovereign full-stack quality loop -- codec, client, relay, metrics one auditable stack Measured exceed: `vc_t8_init` in `runtime/nx_vcodec.nx`, verified at emit. Majors sit on OS/browser/silicon decoders they do not control end to end Adoption: `BUILT-UNPROMOTED` — PARTIAL: compiled, never promoted to the serving root: /api/promote it. adoption BUILT-UNPROMOTED	◉	○	○	○	○
percentage canary rollout Open — watching `runtime/nx_video_ops.nx` : `vo_canary_pct`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Flag-gated ship exists; a percentage rollout with automatic halt on drift is the Netflix canary class watching vo_canary_pct	○	◉	◉	●	●
MONITOR
full-reference VMAF-class metric Open — watching `runtime/nx_natstat.nx` : `ns_fr_vmaf`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. The no-reference judge exists; a full-reference perceptual metric against the source frame is what the majors grade with watching ns_fr_vmaf	○	◉	●	○	◐
SPOT
real-device conformance matrix Open — watching `runtime/nx_video_ops.nx` : `vo_device_matrix`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. The cross-instance farm covers engines and tiers; a device matrix with real phones and browsers is the certification depth watching vo_device_matrix	○	◉	◉	●	●

On these two registers. Rows are declared in the domain's plan file and carry the debt id, which is the join key back to the sovereign debt plane — that plane, not this page, is the authority on state. Reconciling them automatically (the regen reading the plane and refreshing these rows) is a named, owed rung; until it lands, treat an id here as a pointer to look up, not a status to trust.

Honest verdict. The coverage above is capability presence measured against source — not depth, scale, or polish, where mature rivals may lead. Exceeds are claimed only where a mechanism backs them. Every open gap is a watch contract: it names the organ and symbol that closes it, and this page flips the cell itself when that workstream ships.

Person · product · place — not yet measured for this domain

Every compare carries this layer. Declare knowledge/compare/videoops.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain videoops, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).

References

Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the matrix notes jump here. A dash means honestly absent, never assumed.

[vmaf16] Netflix Technology Blog. Toward A Practical Perceptual Video Quality Metric. 6 June 2016 -- the VMAF (Video Multimethod Assessment Fusion) announcement and open-sourcing. publisher · read in our library knowledge/fetched/cmp_videoops_vmaf16.html · pin h996e35bd0155c10ab0bd6cb62d824de1890066171a2daaf641f60f2273419aeb · accessed 2026-08-18 · vendor-docGrounds: "MONITOR: perceptual quality metric" -- the row's "Netflix invented VMAF" and its Best code are this post; VMAF is the full-reference fusion metric our no-reference nx_natstat is deliberately not, which is why the row grades us 3.
[uvq22] Wang, Yang (YouTube / Google Research). UVQ: Measuring YouTube's Perceptual Video Quality. Google Research blog, 23 August 2022 -- the learned no-reference UGC quality model YouTube runs at scale. publisher · read in our library knowledge/fetched/cmp_videoops_uvq22.html · pin haf8c0e8afe58e7723a8eed99aafa3d725c6c230a2f612556667e8bc1f21a586c · accessed 2026-08-18 · vendor-docGrounds: "MONITOR: perceptual quality metric" -- the row's "YouTube UVQ" column read; UVQ is a no-reference learned judge, the same class of instrument as nx_natstat, and the reason YouTube's code is 1 rather than 0.
[pertitle15] Netflix Technology Blog. Per-Title Encode Optimization. December 2015 -- per-title analysis picks the bitrate/resolution ladder from each title's own complexity instead of one fixed ladder. publisher · read in our library knowledge/fetched/cmp_videoops_pertitle15.html · pin h46579568d6c70c16fb6745969b1e85a4db7b65c3c94bcd24eb2ade77e0dabed8 · accessed 2026-08-18 · vendor-docGrounds: "ADDRESS: adaptive bitrate ladder" -- the row's "Netflix per-title encoding is the reference": our rungQ adapts qp and geometry per network at runtime; per-title chooses the ladder itself, the deeper practice the Best code marks.
[atlas14] Netflix Technology Blog. Introducing Atlas: Netflix's Primary Telemetry Platform. December 2014 -- in-memory time-series telemetry ingesting billions of series per day for near-real-time operational insight. publisher · read in our library knowledge/fetched/cmp_videoops_atlas14.html · pin hbab3c58ab587f0e6dc41e00b44b29fb48766951a98a87e87d2def8d2c0f13537 · accessed 2026-08-18 · vendor-docGrounds: "MONITOR: live ops dashboard for sessions" -- the row's "Netflix Atlas ... class" is this platform; the gap row is honest because our fps/bytes/mode beacon has no ops surface consuming it yet.
[kayenta18] Netflix Technology Blog. Automated Canary Analysis at Netflix with Kayenta. April 2018 -- open-source (with Google) statistical canary judgement that promotes or fails a rollout from metric comparisons. publisher · read in our library knowledge/fetched/cmp_videoops_kayenta18.html · pin h04de5f046f009778078c0a35cb6e4badcd50ca30546b0b1c7e68376596fe4fe1 · accessed 2026-08-18 · vendor-docGrounds: "PREVENT: staged rollout / canary" -- the row's "Netflix canary analysis is the reference": our t8Flag ships flag-gated with no percentage rollout and no statistical judge, which is the distance this post measures.
[rfc4585] Ott, Wenger, Sato, Burmeister, Rey. Extended RTP Profile for Real-time Transport Control Protocol (RTCP)-Based Feedback (RTP/AVPF). IETF RFC 4585, July 2006 -- defines the Picture Loss Indication (PLI) and FIR-class feedback. publisher · read in our library knowledge/fetched/cmp_videoops_rfc4585.html · pin h8bb0556e71f1747219389c2ebbe9f3e5ce5f755f2a4cae0613b9d6b66cae4cb3 · accessed 2026-08-18 · published-standardGrounds: "ADDRESS: keyframe request on loss (PLI-class)" -- the row cites RFC 4585 by name; kreq is our PLI analogue recovering in about one RTT, and Zoom's Best code is standard AVPF behaviour.
[rfc3264] Rosenberg, Schulzrinne. An Offer/Answer Model with the Session Description Protocol (SDP). IETF RFC 3264, June 2002 -- the capability-negotiation model WebRTC/JSEP builds on. publisher · read in our library knowledge/fetched/cmp_videoops_rfc3264.html · pin h16eff04c7fbbff8d3e1b737f4c67e698e0d8e4d0620098f3586e587fe6c7ae36 · accessed 2026-08-18 · published-standardGrounds: "PREVENT: capability/version negotiation" -- the row's "Zoom/WebRTC SDP negotiation is Best" is this offer/answer model; our p.vcv syntax-generation gate is a narrower, build-version handshake, hence code 1 for us.
[argon22] Alliance for Open Media. Graphcore Open-Sources Argon Streams AV1 Video Decoder Verification Tool. 6 September 2022 -- formal-verification-derived AV1 conformance bitstreams spanning the valid stream space, freely available via AOMedia. publisher · read in our library knowledge/fetched/cmp_videoops_argon22.html · pin h7b7769b93cadfc141d466fa15642ecf66ace26c215bf9d1f03dd6a941c77ed2b · accessed 2026-08-18 · vendor-docGrounds: "PREVENT: codec conformance gates (real content+noise)" -- the row's "AV1 argon conformance" bar: Argon Streams AV1 is spec-derived full-coverage conformance (Argon Design, open-sourced via AOMedia), the class our 15+ adversarial gates plus cross-instance farm are graded beside; the row's Google attribution is a paraphrase of AOM stewardship.
[ossfuzz] Google. OSS-Fuzz: continuous fuzzing for open source software (GitHub repository and documentation). publisher · read in our library knowledge/fetched/cmp_videoops_ossfuzz.html · pin h63f9fad341be143f505e5dbf2a00ac328490fa5775461166d71787d1ab5d12a5 · accessed 2026-08-18 · vendor-docGrounds: "PREVENT: decoder fuzz/robustness harness" -- the row's "Google oss-fuzz is Best": continuous coverage-guided fuzzing at scale is the bar our 349k-input vc_fuzz harness (traps 558 to 0, MV clamp fixed in the sovereign decoder) is measured against.

generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/videoops.matrix · every Nishi cell verified against organ source at emit time · watch cells re-measured on every compare beat · zero JS, zero trackers