Facts registry

This file is the single registry of published facts: measured numbers, verdicts, phase states, the parity ledger and what has been delivered. The website (vxedge.dev) builds every number and state it shows from this file and nothing else. The GitHub READMEs and RESULTS.md quote it.

Each measured value is copied from a frozen summary in vxperiments/results/. Those summaries record the input manifest's sha256. This file does not replace them; it is the one place they are quoted from. RESULTS.md explains the numbers: what each hypothesis tests and the caveat that matters most.

To change a fact: edit it here first, in the same change as the summary or state that moved. Then run python3 scripts/facts.py check from this repo. The website picks the change up on its next build. Its CI fails if the vendored copy is stale, or if a page shows a number that is not registered here.

Format. Every table whose first column is Key is machine-read by scripts/facts.py. A table with a Value column holds one fact per row. Any other keyed table is a row table: each cell becomes <key>.<column>, where the column name is the first word of its header, lowercased (p99 µs (IQR) → p99). Keys are lowercase and never change meaning once published. A fact that is retired is deleted, not renamed.

State words are fixed. For deliveries: measured (frozen, reproduced results), live (works today, no frozen number), building (partly landed), planned (specified, not built). For the parity ledger: real (the production software or kernel path itself), emulated (real software on modelled hardware), planned, caveat (works, with a stated limit on what it can tell you).

1 · Project

Key	Value	Source
`project.name`	vxedge	—
`project.version`	0.2.0	`CITATION.cff`
`project.released`	2026-10-13	`CITATION.cff`
`project.stage`	pre-release	pre-1.0: the manifest schema and vxAPI may change between minor versions; published results are frozen (`VERSIONING.md`)
`project.status`	Phases 1–2 complete, phase 3 in progress	`ROADMAP.md`
`project.phase`	3	`ROADMAP.md`
`project.licence_code`	Apache-2.0	`LICENSE`
`project.licence_docs`	CC BY 4.0	`LICENSE-DOCS`
`project.hypotheses`	4	`ARCHITECTURE.md` §2
`project.hypotheses_hold`	4	§2 below
`project.reproduced`	4	§2 below: each verdict re-reached by an independent second campaign
`project.repos`	9	`README.md`: `docs` and eight code repositories
`project.mcp_spec`	2026-07-28	`vxmcp` targets this MCP specification revision

Phases

Key	Phase	Gate	State
`phase.1`	Slicing core, API-first	Reproducible H1/H3/H4 numbers	done
`phase.2`	Real slices and workloads	H2: S-NSSAI vs static VLAN	done
`phase.3`	Production SDV alignment	`vxedge vehicle up` boots a full vehicle end to end	in progress
`phase.4`	Hardware in the loop	Jetson / DRIVE / BlueField / V2X radio	specified
`phase.5`	Remote and multi-user	none yet	backlog

Tracks

Key	Track	Status
`track.m`	M — SysML v2 model governance	M0–M3 closed; M4 waits on a GPU host
`track.c`	C — containerization	complete through C3.2
`track.p`	P — `vxparts` unification	complete

2 · Results

The tables under each hypothesis heading (H1–H4) and under Other completed experiments are measurements. The website paints those values, and only those, in its measured-value colour. The method, parameters and verdicts are not measurements and are not painted.

Method and parameters

The same method for every campaign, and the parameters each hypothesis was tested with. These are configuration, not measurements.

Key	Value	Source
`method.reps`	5 reps	per arm or load point
`method.rate`	1000 pps	probe rate
`method.duration`	60 s	per point
`method.probes`	60 000 probes	per point
`method.warmup`	10 s	discarded before probing
`method.campaign_time`	~35 min	full `exp-001` campaign on an idle host, `GETTING-STARTED.md`
`h1.max_load`	600 mbit	H1: highest offered infotainment load
`h1.uplink`	500 mbit	H1: root rate of the shared uplink, `exp-001-ws/main.yaml`
`h1.overload`	1.2×	H1: max load ÷ uplink
`h1.loads`	5 loads	H1, H2: load levels swept
`h1.threshold`	2.0×	H1: pass threshold for p99 at max load ÷ p99 idle
`h2.tolerance`	1.1×	H2: S-NSSAI p99 growth may exceed VLAN growth by at most this factor
`h3.arms`	6 arms	H3: packaging arms compared
`h3.probes`	1.8M probes	H3: probes across all archives
`h4.artifact`	5 MB	H4: artifact size, delivered back to back
`h4.arms`	2 arms	H4: baseline and OTA burst

Verdicts

Key	Hypothesis	Experiment	Verdict	Manifest sha256	Archives
`h.1`	Isolation: VLAN + tc reservations keep ADAS p99 near baseline under infotainment load	`exp-001`	holds	`2666fbf561ae…`	25 archives
`h.2`	Real S-NSSAI slicing vs static VLAN mapping	`exp-010`	holds	`4fa24418b7a1…`	25 archives
`h.3`	Virtual-ECU packaging affects zone-to-compute latency	`exp-002b`	holds	`625840eeeb0c…`	30 archives
`h.4`	Co-located OTA does not compromise the ADAS slice	`exp-003`	holds	`1efebdef615f…`	10 archives

H1 — isolation (`exp-001`)

Key	Value	Source
`h1.p99_max`	81.9 µs	ADAS p99 at the highest load
`h1.p99_idle`	89.6 µs	ADAS p99 at zero load
`h1.ratio`	0.91×	p99 at max load ÷ p99 idle
`h1.loss`	0.0 %	packet loss at every load level

Per load level. Latency in µs with the IQR across repetitions in brackets. Source: exp-001-summary.md.

Key	load (mbit)	p50 µs (IQR)	p95 µs (IQR)	p99 µs (IQR)
`h1.load.0`	0	49.9 (0.8)	71.4 (6.7)	89.6 (6.4)
`h1.load.100`	100	51.1 (3.0)	72.0 (6.1)	87.5 (7.0)
`h1.load.250`	250	49.8 (1.9)	70.1 (3.3)	84.5 (4.7)
`h1.load.400`	400	50.6 (1.3)	72.3 (3.1)	85.2 (4.2)
`h1.load.600`	600	51.4 (1.8)	69.2 (2.8)	81.9 (5.0)

H2 — real slicing vs static mapping (`exp-010`)

Key	Value	Source
`h2.ratio_snssai`	0.95×	S-NSSAI p99 growth, idle → max load
`h2.ratio_vlan`	0.91×	VLAN p99 growth over the same sweep (`exp-001`)
`h2.floor`	317–337 µs	S-NSSAI p99 across all loads: software UPF and GTP-U

Per load level, S-NSSAI path. Source: exp-010-summary.md.

Key	load (mbit)	p50 µs (IQR)	p95 µs (IQR)	p99 µs (IQR)
`h2.load.0`	0	269.3 (7.3)	305.2 (11.5)	335.1 (14.1)
`h2.load.100`	100	254.1 (3.6)	299.5 (2.6)	324.1 (6.9)
`h2.load.250`	250	261.6 (4.0)	313.1 (4.7)	336.6 (12.7)
`h2.load.400`	400	260.9 (1.9)	294.7 (7.4)	318.7 (23.0)
`h2.load.600`	600	260.7 (4.7)	293.4 (5.9)	317.2 (10.8)

H3 — virtualization choice (`exp-002b`)

Key	Value	Source
`h3.transport`	−26.0 µs	Δp50, virtio MMIO → PCI (`microvm-pci` vs `microvm-mmio`)
`h3.transport_iqr`	2.2 µs	IQR of that comparison
`h3.management`	−17.3 µs	Δp50, libvirt vs raw QEMU (`libvirt-q35` vs `q35-pci`)
`h3.management_iqr`	2.0 µs	IQR of that comparison
`h3.vhost`	−9.5 µs	Δp50, vhost-net on vs off (`q35-vhost` vs `q35-pci`)
`h3.vhost_iqr`	2.5 µs	IQR of that comparison
`h3.machine`	−0.3 µs	Δp50, q35 vs microvm (`q35-pci` vs `microvm-pci`): no measurable effect

Per packaging arm. CPU and RSS are means across repetitions. Source: exp-002b-summary.md.

Key	arm	p50 µs (IQR)	p99 µs (IQR)	CPU %	RSS MB
`h3.arm.container`	container	54.6 (1.3)	94.4 (2.8)	4.3	7.3
`h3.arm.libvirt-q35`	libvirt-q35	70.3 (1.6)	117.6 (2.4)	8.2	166.2
`h3.arm.q35-vhost`	q35-vhost	78.1 (2.5)	122.9 (4.4)	9.1	150.6
`h3.arm.q35-pci`	q35-pci	87.6 (2.0)	138.6 (2.6)	11.3	150.9
`h3.arm.microvm-pci`	microvm-pci	87.9 (2.2)	138.2 (3.2)	11.4	150.2
`h3.arm.microvm-mmio`	microvm-mmio	113.9 (2.0)	165.0 (6.3)	15.0	146.4

H4 — OTA co-location (`exp-003`)

Key	Value	Source
`h4.p50_baseline`	52.2 µs	ADAS p50 without OTA
`h4.p50_ota`	51.9 µs	ADAS p50 during the OTA burst
`h4.delta`	−0.3 µs	difference of the two
`h4.iqr`	6.0 µs	baseline IQR, the one-sided threshold
`h4.deliveries`	361 deliveries	signed deliveries per repetition (median)
`h4.failed`	0	failed deliveries
`h4.loss`	0.0 %	packet loss in both arms
`h4.repro_inside`	0 of 6	metric/arm pairs of the same-day second campaign inside the first campaign's IQR; the verdict held

Per arm. Source: exp-003-summary.md.

Key	arm	p50 µs (IQR)	p95 µs (IQR)	p99 µs (IQR)	OTA deliveries ok/fail
`h4.arm.baseline`	baseline	52.2 (6.0)	80.7 (4.9)	97.7 (5.6)	n/a
`h4.arm.ota-burst`	ota-burst	51.9 (1.7)	79.9 (6.3)	93.1 (5.5)	361 / 0

Other completed experiments

Key	Value	Source
`exp005.confirmed`	20 of 20	`exp-005`: cold-start first-packet effect, repetitions confirming it
`exp014.verdict`	holds	`exp-014`: star vs daisy-chained zones, hop-count effect

3 · Deliveries

Parity ledger

How close each part of the virtual vehicle is to production today, and what closes the gap (p3, p4: the phase that does).

Key	Part	Today	State	Closes the gap
`parity.network`	Network path	Linux kernel: OVS or bridge, tc HTB per slice, VXLAN, WireGuard	real	BlueField offload · p4
`parity.core`	5G core	Open5GS with real S-NSSAI PDU sessions	real	—
`parity.radio`	5G radio	UERANSIM gNB and UE in software	emulated	Aerial RAN · p4
`parity.zones`	Zone ECUs	Real guest OS under container, microVM or KVM	emulated	QNX SDP 8 · p3
`parity.harness`	Harness	Per-link rate, delay, jitter, loss; 100BASE-T1 length and weight	emulated	TSN 802.1Qbv · p3
`parity.mcu`	MCU firmware	Unmodified firmware on register-level models	planned	Renode · p3
`parity.sensors`	Sensors	World streams carried as slice traffic	planned	CARLA · p3
`parity.compute`	Central compute	Autoware and ROS 2 in containers	planned	DRIVE AGX · p4
`parity.timing`	Timing	Userspace timestamps on a shared host	caveat	Kernel timestamps · open

Use cases

Key	Use case	State	Evidence
`case.isolation`	Slice isolation	measured	H1, H2
`case.topology`	Zonal topology studies	measured	`exp-014`, `exp-005`
`case.packaging`	ECU packaging and fidelity	measured	H3
`case.ota`	OTA beside safety workloads	measured	H4
`case.agents`	Agent-operated experiments	building	`vxmcp`, `vxmodels`
`case.autonomy`	Autonomy closed loop	planned	`exp-011`
`case.signals`	Signals and diagnostics	building	phase 3
`case.reproduce`	Reproduce anything	live	every archive

Agent path

How a generated experiment reaches the bench.

Key	Step	State
`gate.propose`	Propose a SysML v2 model	live
`gate.typecheck`	Gate 1: model type check	live
`gate.validate`	Gate 2: `vxedge manifest validate --json`	live
`gate.confirm`	Multi-round-trip confirmation before a state change	planned
`gate.run`	Idempotent `up`, streamed as a job, audited	live
`gate.archive`	Evidence archived with digests	live

vxmcp tools

Key	Tool	Service	State
`mcp.status`	`experiment_status`	ExperimentService	live
`mcp.list`	zone / compute list	ZoneService · ComputeService	planned
`mcp.watch`	job watch · events	JobService · EventService	planned
`mcp.results`	results summary	results archive	planned
`mcp.mutate`	up · down · traffic up	behind confirmation	planned

Other deliveries

Key	Value	Source
`retarget.validate`	live	an experiment retargets to another vehicle by id and validates
`vehicle.up`	live	the `vehicle` verbs (up, down, status) on a bare vehicle: gateway and zones (verified 2026-10-13, one-zone vehicle). The phase-3 gate, a full vehicle end to end, is not claimed
`retarget.fleet`	planned	two vehicles at once (fleet layer, `PROVISIONING.md`)
`paper.1`	in preparation	phase-1 dataset (H1/H3/H4) for arXiv, `ROADMAP.md`

Cite this page

Djaafri, F. (2026). vxedge: Facts registry. Documentation, version 0.2.0, commit 33ede87. https://vxedge.dev/docs/FACTS.html

@misc{djaafrivxedgefacts2026,
  author = {Djaafri, Fouzi},
  title  = {vxedge: Facts registry},
  year   = {2026},
  url    = {https://vxedge.dev/docs/FACTS.html},
  note   = {Documentation, version 0.2.0, source commit 33ede87}
}

Keyboard shortcuts