[c-RCP-AUDIT-02] Backfill TC18 citations on ~801 uncited requirements

[SoundMatt]

Background

.fusa-reqs.json has 1023 requirements. As of 2026-08-07, only 222 (~22%) carry a tc18 citation field pointing back to the governing spec ("§X.Y Table N, TC18.txt L1234-1256" format). This corrects an earlier, wrong impression (based on a flawed grep -c count) that this repo was already near-total on citation coverage — it is not.

This is the same class of work already done for c-RCP's SHOULD/MAY clauses (see docs/TC18-NON-NORMATIVE-CLAUSES.md and the MAY-clause citation pass) and already flagged as needed for cpp-RCP/rust-RCP/go-RCP's MUST clauses — c-RCP's own gap is smaller in relative terms but still large in absolute count.

Current per-category breakdown (categories below 100% cited, TC18-relevant ones only)

Categories correctly at 0% because they are NOT TC18 concepts (RELAY-generic sentinels, protocol-bridge adapters, or tooling) — do not add citations to these, it would be citing a spec section that doesn't actually govern that code: ADMIN, AUTH, CLI, CORE, DDS, DOIP, GRPC, MQTT, OBS, PLATFORM, REC, REST, RL, SHMEM, SOMEIP, UDS, ERR (RELAY-generic sentinels, not TC18 Table 27 — that's WIREERR, see issue #163), FI (fault-injection test tooling).

Categories genuinely TC18-relevant and needing backfill (uncited/total):

AVTP 0/20   DL 0/13    FRAG 0/18   L2 0/10     MDNS 0/11
PWR 0/15    TSN 0/8    UDP 0/19    WDG 0/12    CFG 2/13
DISC 1/29   MOCK 1/27  WDG 1/12    SEQ 4/14    TRIG 4/13
CANCEL 4/12 PWM 4/58   ADC 8/36    GPIO 5/37   SPI 5/35
ISELED 5/28 I2C 3/20   LINEP 6/22  MDIO 6/22   SRV 6/21
WAKEUP 6/22 CHAIN 3/12 CMP 10/27   CANEP 11/32 UART 8/37
E2E 16/44   ACF 15/29  LIFECYCLE 18/37   PWRMODE 16/28   RMAP 47/69

(RMAP/LIFECYCLE/PWRMODE/ACF are already the best-covered — finish those first, see Method below.)

Spot-check evidence (don't assume the rest without checking)

AVTP and WDG were individually verified (2026-08-07): both are legitimate, complete, well-tested content with no citation at all — i.e. this is a citation-backfill problem, not a missing-content problem, at least for those two. The other ~48 categories above were not individually re-verified with the same rigor this session. Do not assume they're all "just uncited" without checking each one — if a category turns out to have a genuine content gap during this backfill, that's a different, more serious finding (stop and open a dedicated issue, don't just add a citation to something wrong).

Method (for the agent picking this up — same discipline that caught real bugs in this repo's earlier passes, e.g. c-RCP#151, #153-158)

1. Work module by module, not all at once. Suggested order: finish the already-partially-cited foundational layers first (ACF, AVTP, RMAP, LIFECYCLE, PWRMODE — these are load-bearing for everything else), then per-endpoint-type batches (GPIO, SPI, PWM, ADC, I2C, UART, CAN, LIN, ISELED, MDIO, WAKEUP), then the remaining transport/protocol layers (L2, UDP, DL, FRAG, TSN, MDNS, PWR).
2. For each requirement, read its text field, find the corresponding TC18 section by re-reading the actual spec (re-extract pdftotext -layout from OA_TC18_specification_v_0.5.1_RC.pdf if no working extraction exists — do not commit the PDF itself, confidentiality restriction), and confirm the requirement's stated behavior actually matches what TC18 says before citing it. This is not a mechanical fill — the equivalent pass on other requirements this session found real bugs (mis-stated behavior, wrong field widths) precisely because each citation forced a real re-read against the primary source.
3. Citation format: "§X.Y (\"short paraphrase\"), TC18.txt LNNNN" or "...LNNNN-MMMM" for a range — match this repo's existing convention exactly (see any of the 222 already-cited entries for examples).
4. Batch into one PR per module/category (not one PR per requirement, not one giant PR) — matches this repo's established size discipline.
5. Verify with a freshly-built, CI-pinned cfusa per PR: check 0 errors (citations are additive, should never break trace coverage since they don't change impl/test tags), full test suite green.
6. Update ROADMAP.md with a milestone entry per batch, following the existing convention (see the SHOULD/MAY pass's own milestone entry for the exact style).

Not in scope for this issue

• The WIREERR cluster's behavioral gap (codes defined but not wired to real response paths) — that's issue [c-RCP-AUDIT-01] Wire TC18 Table 27's 16 unmapped error codes to real response paths #163, a different kind of gap entirely (missing wiring, not missing citation).
• tc18_master_id — a separate, new field this repo doesn't have on any entry yet, blocked on a cross-repo master-catalog rebuild that hasn't happened. Don't add it as part of this backfill; track separately.

[SoundMatt]

Batch 1 (AVTP) done: #171 (v0.121.0), merged and verified. 11/20 AVTP requirements cited against TC18 §11.1's NTSCF/TSCF header definitions; the other 9 are genuine implementation-detail requirements with no TC18 clause, correctly left uncited.

~789 requirements remain across the rest of the catalog. Continuing with the foundational modules next (ACF, RMAP, LIFECYCLE, PWRMODE) as originally planned, since everything else depends on those being solid first.

[SoundMatt]

Batch 2 (ACF) done: #172 (v0.122.0), merged and verified. 11/14 remaining uncited ACF requirements cited against TC18 §11.2.1 Figure 7/Table 4. 3 left uncited pending closer investigation (not force-fit) — one genuinely interesting open question: REQ-ACF-012's RCP_ACF_MTV_UNCERTAIN third mtv state doesn't match TC18's binary mtv bit definition, may be conflating it with the AVTP TSCF header's separate tu bit.

ACF now 27/29 cited, AVTP 11/20. Continuing with RMAP (47/69) next.

[SoundMatt]

Batch 3 (RMAP) merged: PR #173, v0.123.0, commit 6138ddb.

RMAP was 47/69 cited going in. Cited 15 of the remaining 22 against four TC18 sources:

• REQ-RMAP-001/002: §12.3.1.1 (EP0 = discovery byte_bus_id 0x00)
• REQ-RMAP-003: §13.3 Table 33 (svr_root_client_index == no root client)
• REQ-RMAP-004-008: §12.9.1.1 (compound-request options-group bundling)
• REQ-RMAP-009-012: §12.3.1.2 (root-client write access via EP0 / per-endpoint owning-stream write access)
• REQ-RMAP-020-022: §13.2 Table 23 (EP_ID_config's required ascending ordering)

Left 7 uncited, deliberately: REQ-RMAP-013 (pin-property bitmask distinctness) and REQ-RMAP-014-019 (string-helper guarantees, config-struct zero-init boilerplate) -- genuine implementation-detail requirements with no TC18 clause.

RMAP now 62/69 cited. Full suite green (64/64 tests, 0 cfusa check errors, 100% traced+tested). Moving on to the LIFECYCLE batch (19/37 uncited) next.