feat(tdm): ACM / SIAM / AMS as tdm-ieee siblings — blocked on #458, or they ship as three more sources that do nothing

[sotashimozono]

ADR-0039 names IEEE, ACM, SIAM and AMS together: all four answer a scripted client with 202 and
an empty body, and per-publisher TDM credentials are the supported route for all four. #430 did
IEEE. These three are the siblings.

The template is settled — tdm_ieee.rs is four touch points plus docs, and ADR-0042 is the
precedent for shipping against an inferred contract when the programme gates the real one.

Blocked on #458, deliberately

#458 shows that a Tier-3 source is consulted only when Crossref failed, and is metadata-only.
Crossref resolves ACM, SIAM and AMS DOIs as readily as it resolves IEEE's. So adding these three
today ships three more sources that:

• are never consulted for the DOIs they exist to serve, and
• could not return a PDF if they were.

That is the #442 defect class multiplied by three, entered into deliberately this time. Doing it
before #458 lands would be the same mistake #442 warned about when it said "adding IEEE before
this lands would ship a fourth unreachable source" — and that warning was about wiring, which
turned out to be the smaller of the two problems.

Also blocked on information, as IEEE was

Per publisher, the same four unknowns: endpoint, auth shape, response shape, rate limits and
terms. Public starting points:

• ACM — ACM Digital Library TDM; ACM has an OPEN access programme and a documented TDM
  contact route.
• SIAM — TDM enquiries go through SIAM directly; no public developer portal found.
• AMS — MathSciNet has an API; whether the journal full text is reachable by the same
  credential is unknown. Relevant to A 429 on the content leg ends the run: no fall-through to the other enabled sources, and the 0.8.8 resolution trace is not emitted on this path #445, whose reported failure was an AMS 429.

ADR-0042's posture applies unchanged: implement what is public, mark it (unverified), fail
loudly on shape mismatch, and provide a DOIGET_<PUBLISHER>_BASE from the start.

Prefixes, for when this is unblocked

Verified against the Crossref registrant registry:

publisher	prefix
ACM	10.1145
AMS	10.1090
SIAM	10.1137

Order

One PR per publisher, after #458. Not a tdm-all umbrella (SCOPE.md non-goal 12).

Refs #430, #442, #458, ADR-0039, ADR-0041, ADR-0042.

[sotashimozono]

Status check while working through the open issues.

The #458 blocker is lifted

#458 closed on 2026-08-26 and ADR-0044 records the decision: Tier-3 sources are consulted on a blocked content leg, not on a Crossref miss. resolve_tdm_chain runs before the Tier-2 chain in fetch_paper_doi, prefix-scoped per ADR-0041, and orchestrator.rs::tdm_chain_tests drives it — including tdm_content_leg_rejects_a_non_pdf_body_and_keeps_the_original_block, which is the case a WAF holding response would have slipped through.

So the argument in this issue's "Blocked on #458, deliberately" section no longer applies. ACM / SIAM / AMS sources added today would be reached for the DOIs they exist to serve.

The second blocker has not moved, and I cannot move it

This issue names it too: "Also blocked on information, as IEEE was." Per publisher, four unknowns — endpoint, auth shape, response shape, rate limits and terms.

• ACM — the DL has a TDM contact route; no public developer portal documenting an API contract.
• SIAM — enquiries go through SIAM directly; no public portal found.
• AMS — MathSciNet has an API; whether journal full text is reachable by the same credential is unknown.

ADR-0042 let tdm-ieee ship against an inferred contract, and that was defensible because IEEE publishes a developer portal to infer from. For these three there is no equivalent public document, so an implementation would be inferred from nothing — which is the failure mode ADR-0046 and ADR-0047 were both written about, and #460 is what it costs.

I am not shipping three sources against guessed contracts. Getting the contracts is a maintainer action: a TDM enquiry to each publisher.

Suggested state

Keep open, with the blocker relabelled. The dependency is no longer #458; it is three unanswered emails. #462 (the end-to-end suite) is still worth landing first, for the reason #474 gives — each new source then arrives with a one-line proof that the production path reaches it.

The prefix table in this issue (10.1145 ACM, 10.1090 AMS, 10.1137 SIAM) is verified against the Crossref registrant registry and is the part that needs no further work.