feat(network): standard CQF synthesis baseline (REQ-TSN-SYNTH-CQF-BASE-001)

[avrabe]

What

Standard 802.1Qch CQF (Cyclic Queuing and Forwarding) synthesis — the second TSN synthesizer, complementing the Qbv GCL baseline (#290). synthesize_cqf takes a set of reserved flows (each with a per-cycle reservation in bits, a path, and an end-to-end deadline) and produces a CqfSchedule: a global cycle time T plus per-link admission.

This implements REQ-TSN-SYNTH-CQF-BASE-001 (standard two-buffer, single-cycle-T baseline), a deliberate precursor split from REQ-TSN-SYNTH-CQF-001 — mirroring the REQ-TSN-SYNTH-QBV-BASE-001 → REQ-TSN-SYNTH-QBV-001 split.

Why CQF is the clean next synthesis target

CQF's worst-case end-to-end delay is structurally decoupled from per-link load: a frame received in cycle c is forwarded in c+1, so for H hops at cycle time T:

(H − 1)·T  ≤  D  ≤  (H + 1)·T

independent of topology and load, given no cycle is oversubscribed (IETF draft-eckert-detnet-tcqf-05, IEEE 802.1Qch). So feasibility factors into two small, local checks — no network-calculus curve composition:

1. structural delay — (H+1)·T ≤ deadline for every flow;
2. per-cycle admission — every link's aggregate reservation ≤ csize = T·link_rate.

That is exactly why CQF routes around the network-wide NC-composition bridge that an MILP frame-scheduler needs to be both necessary and checkable. Synthesis: pick the largest whole-ns T meeting every deadline (looser T → more bandwidth, so the deadline-limited T maximizes admission headroom), then admit iff every link fits csize.

The oracle is anchored to external ground truth

Unlike the Qbv baseline (whose oracle was self-consistent against its own checker), the CQF checker's delay formula is pinned to a published worked example in draft-eckert-detnet-tcqf-05:

*"if the number of hops is 24, when cycle interval is set to 10us, the end-to-end latency bound can be around (24+1)10 = 250 us."

delay_bound_matches_published_tcqf_example asserts cqf_delay_max_ps(24, 10µs) == 250µs and cqf_delay_min_ps == 230µs (jitter exactly 2T). So if the delay model were subtly wrong (e.g. H·T instead of (H+1)·T), the pin fails against the literature — something a self-consistent synth↔checker loop would miss. The synthesizer self-certifies against this same checker before returning.

Falsification statement

If synthesize_cqf returns a CqfSchedule with cycle time T, then (a) for every input flow, (hops+1)·T ≤ deadline_ps, and (b) for every link, the aggregate per-cycle reservation ≤ T·link_rate — where the delay formula matches the IETF TCQF published bound. Otherwise it returns a typed CqfSynthError. A returned schedule in which any flow misses its deadline, or any link exceeds csize, or whose delay formula disagrees with the 250µs/230µs TCQF example, falsifies this.

Scope notes (honest to the primitive)

• Standard two-buffer CQF, one global cycle time, homogeneous link rate, reservation-based admission. Multi-CQF/TCQF (3–7 tagged buffers, heterogeneous cycles, per-flow injection-time planning across a hyperperiod, metaheuristic scale numbers) is REQ-TSN-SYNTH-CQF-001, deferred to v0.21.0.
• synthesize_cqf is a library primitive with a unit oracle; there is no AADL→synth→export path yet (the DEC-TSN-OSS-001 wedge is demonstrated once the bridge exists).
• No solver, no new dependency — pure integer arithmetic; builds with --no-default-features.

Gates (local)

• cargo test -p spar-network — 134 lib pass (7 new CQF) + integration green
• cargo clippy -p spar-network --all-targets -- -D warnings — clean
• cargo fmt -p spar-network -- --check — clean
• cargo build -p spar-network --no-default-features — OK

🤖 Generated with Claude Code

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}

[codecov]

Codecov Report

❌ Patch coverage is 87.73585% with 26 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/spar-network/src/cqf.rs	87.73%	26 Missing ⚠️

📢 Thoughts on this report? Let us know!