Split the GitHub mirror so Gitea creates its runs and no delete is dropped

AGENTS.md

@@ -32,23 +32,16 @@ one wins. They do not override the hard rules below.
    about a message the application sent reaches it as a report rather than as an inbound message, and
    says whether it is final, so nothing has to read the PDU to tell those apart. An option retunes a
    default or opts out of it; an option does not switch on the thing the caller obviously wanted.
-6. **Configurable and extendable, never at the defaults' expense.** Where an application needs other
-   than the default and cannot build it from what is exported — a rate limit counted per PDU, an
-   alphabet, a receipt format — it gets an option or a hook rather than a fork. A call that passes no
-   options stays exactly as easy and as safe, and a hook is a seam the library calls, never a way into
-   its internals.
-7. **A small, stable public surface over reshapeable internals.** Only what `src/index.ts` exports is
+6. **A small, stable public surface over reshapeable internals.** Only what `src/index.ts` exports is
    published. A new option has to beat "the application can do this itself", and has to keep a
    promise this library can verify. The low-level surface is a passthrough: policy binds what the
    library composes, never what the caller wrote.
-8. **State wider than one session goes through one store.** A pool of sessions, a limit shared
-   between processes, and what has to survive a restart — receipts still awaited, a message half
-   reassembled — are held through a store interface and never beside it. Without a store the
-   application supplies, that state is in memory and ends with the process, and the defaults need
-   none. The interface carries the library's own versioned records, never an internal shape handed
-   to the application to persist. Coordinating processes any other way is declined without a fresh
-   argument each time.
-9. **It builds, tests and runs the same everywhere.** Container-only toolchain, no runtime
+7. **Nothing that needs state wider than one session.** No throughput throttling, no persistence
+   across a restart, no coordination between processes — and no seam handing the application state to
+   persist for one of those either, which commits to the same scope through the back door and
+   publishes an internal shape to do it. This is the scope floor, and it is why an otherwise
+   reasonable feature is declined without a fresh argument each time.
+8. **It builds, tests and runs the same everywhere.** Container-only toolchain, no runtime
    dependencies, the Node 18 floor verified in CI rather than asserted, every README example executed
    by the suite.
 
@@ -301,7 +294,7 @@ Grouped by what each one constrains.
 - **`PduRefusedError` is exported, and `sessionError` names it in the event's type.** Maintainer's
   call, 2026-09-05, from a product review: one event carries both a PDU the peer malformed and the
   session's own failure, and `instanceof` is the only way to separate them that hard rule 4 allows —
-  without the class as a value an application is left string-matching `err.message`. Goal 7 is paid by
+  without the class as a value an application is left string-matching `err.message`. Goal 6 is paid by
   exporting the discriminant and the struct it carries and nothing else: `PduHeader` is named because
   an application that logs or forwards a header wants a name for it, `PduRefusalReason` is not
   because `reason` is compared against string literals, and an accessor
@@ -328,7 +321,7 @@ Grouped by what each one constrains.
   runtime: `Object.hasOwn(encodings, x)` is the only test the published surface offered and it
   narrows nothing, so `isEncodingName()` is exported beside `isCommandName()` and `isErrorName()`,
   which serve their own tables that way. Rejected: a `Result` signature on all three, which costs
-  every typed consumer a narrow forever — goal 7, and the tag is the last cheap chance to spend it —
+  every typed consumer a narrow forever — goal 6, and the tag is the last cheap chance to spend it —
   to guard a state the compiler refuses. Where the domain really is open the check is already there:
   `sendSms()` takes its options as `unknown` and refuses `encoding` by name, which is what a caller
   without types gets. `smppTime.encode()` is where that reasoning lands the other way and is recorded
@@ -433,13 +426,13 @@ Grouped by what each one constrains.
   the suite ran took the 0x40 this library sends on a concatenated segment, but Route Mobile and
   Kaleyra both document `esm_class` 0x43 for one, and a caller facing either had to hand-build every
   segment through `send()` — giving up the split, the per-segment ids, the send window and the
-  receipt merge, which is what goal 7 means by beating "the application can do this itself". The four
+  receipt merge, which is what goal 6 means by beating "the application can do this itself". The four
   modes of SMPP 3.4 5.2.12 are a `MESSAGING_MODE` constant group and the option takes one of their
   names, so 0x43 is a composition this library makes rather than a value a caller states, and the UDH
   indicator a segment carrying a header needs cannot be cleared by anything the option can express.
   It takes three of those four: 2.10.3 carries transaction mode on `data_sm` alone, and none goes out
   of here, so `FORWARD` stays in the group that mirrors the spec table and `sendSms()` refuses it by
-  that reason rather than as an unknown name — a mode this library cannot deliver is a promise goal 7
+  that reason rather than as an unknown name — a mode this library cannot deliver is a promise goal 6
   will not let it make. `DATAGRAM` with `dlr: true` is refused on the same footing: 2.10.2 defines the
   report away, so arming `DlrMerger` for one is goal 2's wrong answer, where the mode alone and a
   report under any other mode both go out untouched. Those three names left `ESM_CLASS`, where they
@@ -496,7 +489,7 @@ Grouped by what each one constrains.
   `encodingByDataCoding()` reads the alphabet off that same test rather than repeating the group
   masks beside it. It is exported for the reason `concatOf()` is — an application that needs a class
   other than 0 would otherwise rewrite the read this fixed. Rejected: a `messageClass` field on the
-  `sms` event, which pays goal 7 for three classes nothing here acts on, where the boolean the
+  `sms` event, which pays goal 6 for three classes nothing here acts on, where the boolean the
   application already had covers the one it does. Compressed text is out of scope and stays out —
   nothing here implements 3GPP TS 23.042, so a compressed body reaches the application as whatever
   its declared alphabet makes of it — but bit 5 does not move the class bits, so 0x30 is read as
@@ -678,14 +671,14 @@ Grouped by what each one constrains.
   returned `42 44 46` — `"BDF"` — reported as built, while a string `message_payload` was cut to its
   low octets whatever `data_coding` said. Goal 2 owns it, as it owns the `sendSms()` guard above.
   The line falls at the string: a `Buffer` is octets the caller already chose and goes out as given
-  under any `data_coding`, which is what keeps goal 7's escape hatch open — the raw UDH, 8-bit binary
+  under any `data_coding`, which is what keeps goal 6's escape hatch open — the raw UDH, 8-bit binary
   and deliberately malformed bodies `interop-tests/` builds are all still buildable — and a string
   with no `data_coding` is untouched, detection carrying every character it was picked for. The
   guard is `unencodable()` again rather than a second reading, and `unencodableText()` is the
   character, its code point and its index said once for both refusals — unexported where
   `unencodable()` is published, since wording `{ char, index }` into a sentence rewrites no read a
   caller would get wrong, where asking the codec is, and publishing it would freeze this library's
-  error prose as API for an application whose own refusal should read like itself. Goal 7, from the
+  error prose as API for an application whose own refusal should read like itself. Goal 6, from the
   architecture review of [#99](https://github.com/larvit/larvitsmpp/pull/99), 2026-09-09. It is
   reached through
   `encodeBody()` in `message.ts`, which is where the `data_coding`-to-text pair already lives:
@@ -840,7 +833,7 @@ Grouped by what each one constrains.
   dropped with the link — is traffic the peer will not send again, so each one reaches `sessionError`
   as well as the log. Rejected there: an exported `MessageLostError` carrying the group, on the
   `PduRefusedError` pattern — no `sms` ever fired for that group, so there is nothing in it the
-  application could act on, and goal 7 does not buy a second exported class to make a count
+  application could act on, and goal 6 does not buy a second exported class to make a count
   distinguishable. Accepted: a completing segment whose own answer the socket would not carry still
   reaches the application, because the message is whole and correct and the failed answer is on
   `sessionError` — a peer that re-sends after the drop is the smaller risk than dropping a message
@@ -852,7 +845,7 @@ Grouped by what each one constrains.
   of the multipart change: `server()` filled the session's only `onRequest` slot, so the escape hatch
   the error above names was reachable only by hand-wiring a `Session` over a raw socket, giving up
   bind acceptance, `authenticate`, the session set and the drain `close()` runs over it — which is
-  what goal 7 means by beating "the application can do this itself". What the library verifies is the
+  what goal 6 means by beating "the application can do this itself". What the library verifies is the
   ordering rather than the hook's honesty about answering: the hook is consulted only for a non-bind
   request on a session already bound, so no bind — a second one on a live session included — and
   nothing a peer sends before one can be intercepted however the hook is written. One
@@ -1028,10 +1021,3 @@ Grouped by what each one constrains.
   their own. The four files that are not tests are the exception the rule needs stated:
   `dummy-smsc.ts`, `raw-pdus.ts`, `reference-smpp.d.ts` and `teardown.ts` answer no question and are
   named for what they hold.
-
-- **CI tests on Linux only; `src/` keeps off what is known to break on macOS or Windows.** Maintainer's
-  call, 2026-09-14. Nothing verifies either platform, so the code avoids what is known to differ there:
-  shelling out, a path joined by hand, a signal Windows does not deliver, a Unix socket or a file mode.
-  That binds what `dist/` runs; the container tooling, `interop-tests/` and the `package.json` scripts
-  run on Linux by goal 9. Rejected: macOS and Windows runners, on GitHub's mirror or as Gitea
-  host-mode runners on a Windows VM and a Mac.

todo.md

@@ -9,8 +9,7 @@ govern it, and nothing here is a source anything else may cite.
 ## Status
 
 The rewrite is **feature complete and green**: the suite, lint and typecheck are clean on Node 18
-to 26, and 0.5.0 is on npm. What is left is housekeeping around the release, a few things worth
-adding, and the gaps a comparison with other SMPP libraries found.
+to 26. What is left is release work and a few things worth adding before or after 0.5.0.
 
 ## The agreed API
 
@@ -98,7 +97,7 @@ place issues are filed. Maintainer's calls, 2026-09-13 and 2026-09-14.
 
 - [x] 0.5.0 rather than 1.0.0, while usage is this low. Maintainer's call, 2026-09-14.
 - [x] `NPM_TOKEN`, which `.gitea/workflows/release.yaml` needs, is a Gitea organization secret.
-- [x] Tag `v0.5.0` on Gitea to publish. The first publish creates `@larvit/smpp` on npm, provided the
+- [ ] Tag `v0.5.0` on Gitea to publish. The first publish creates `@larvit/smpp` on npm, provided the
       token can publish under `@larvit`.
 - [ ] `npm deprecate larvitsmpp` pointing at `@larvit/smpp`. Maintainer's call to run it; not
       something CI should do.
@@ -246,6 +245,9 @@ the rewrite, for a dependency added later. Maintainer's call, 2026-09-14.
       end to end. The interop suite is the natural place.
 - [ ] **Move to TypeScript 7** once `typescript-eslint` supports it; `renovate.json` pins TypeScript
       below 6.1 for exactly that reason.
+- [ ] **Coverage reporting.** `node --test --experimental-test-coverage` works today; nothing
+      publishes the numbers.
+
 - [ ] **An `onReceipt` hook.** Receipt text is only loosely specified and operators disagree on it,
       but `dlrFromPdu()` is wired into `IncomingRequests` with no seam of its own: an application
       facing a format we do not parse has to take the whole PDU on `onRequest` and reimplement the
@@ -253,186 +255,18 @@ the rewrite, for a dependency added later. Maintainer's call, 2026-09-14.
       Mirror the `onRequest` seam — return a `Dlr` to own the receipt, `undefined` to fall through
       to the built-in parser.
 
-## Gaps against other SMPP libraries
-
-From comparing 0.5.0 with `smpp`, `@semyonf/smpp`, `@leissner/node-red-smpp`, `node-smpp-next`,
-`smpp-js-sdk`, `smppjs`, cloudhopper-smpp, jsmpp, go-smpp, Kannel, Jasmin and php-smpp, 2026-09-14.
-Each lands under AGENTS.md goal 6: an option or a hook, with the call that passes none unchanged.
-
-### Sending
-
-- [ ] **A limiter hook, with a messages-per-second cap built on it.** Maintainer's call, 2026-09-14,
-      reversing the earlier decline: Kannel, Jasmin, go-smpp and `smpp-js-sdk` all limit throughput.
-      Count PDUs, not `sendSms()` calls — a long message is one `submit_sm` per segment and the
-      operator counts those, which an application wrapping `sendSms()` cannot see. The hook takes a
-      limiter the application already runs; the built-in cap counts per session until the store at the
-      bottom lets it span sessions and processes. The default stays uncapped. Open: the hook's shape (a
-      wait that resolves when a PDU may go, cut short by the send's `signal`), which requests it gates
-      — messages, never `enquire_link`, `unbind` or a response — and whether its wait counts against
-      `responseTimeout`.
-
-- [ ] **Back off and resend on `ESME_RTHROTTLED`.** The SMSC refused the PDU, so resending cannot
-      duplicate it and goal 2 holds, and the retry needs nothing wider than the session. Needs a
-      decision: on by default with a bounded budget, as goal 5 suggests, and whether `ESME_RMSGQFUL`
-      counts too. A throttled answer is also what a limiter hook wants to hear about.
-
-- [ ] **`sendSms()` takes the rest of `submit_sm`.** `service_type`, `priority_flag`, `protocol_id`,
-      `replace_if_present_flag` and TLVs on every segment, and `registered_delivery` beyond final
-      receipts: on failure only, and intermediate notifications. Today each needs `send()`, which gives
-      up splitting, the alphabet checks and receipt merging. TLVs are the common case: India's DLT
-      rules put `PE_ID` (0x1400) and `TEMPLATE_ID` (0x1401) on every `submit_sm`, and USSD rides on
-      `ussd_service_op`. Refuse a TLV the send composes itself (`sar_*`, `message_payload`). Open: one
-      spelling for receipts, since `dlr: true` and a raw `registered_delivery` could disagree, and
-      whether goal 4's rule on optional parameters binds a TLV the caller named.
-
-- [ ] **Choose how a long message is spelled on the wire.** Only an 8-bit UDH reference goes out
-      (`message.ts`), though the reader takes a 16-bit UDH, `sar_*` and `message_payload` alike. Some
-      SMSCs take only `sar_*` or `message_payload`; php-smpp offers all three. A 16-bit reference also
-      makes a collision rarer: the 8-bit one wraps every 255 sends on a session.
-
-- [ ] **Failover across SMSC hosts.** Maintainer's call, 2026-09-14. `client()` takes one `host` and
-      `port`; Kannel, Jasmin and php-smpp take several. A list the reconnect loop walks holds only
-      which host the one session is on, so it needs no store. Two options, both maintainer's calls:
-      - **Order**, `fixed` or round robin, default `fixed`. Fixed starts every reconnect at the first
-        host; round robin at the host after the one the link was last on.
-      - **Starting over**, a boolean, default on: once the last host has been tried, go back to the
-        first. Off ends the session once the last host fails, as a drop does under `reconnect: false`.
-      Open: how the list and today's `host` and `port` share one spelling; whether the backoff grows
-      per host or per pass; and whether round robin with starting over off still tries the hosts
-      before the one it started at.
-
-### The server
-
-- [ ] **`sendSms()` on a `server()` session sends `submit_sm` toward the ESME.** Kannel answers
-      `ESME_RINVCMDID` (`interop-tests/kannel.test.ts`, "MO to Kannel"), and goal 1 says that PDU never
-      goes out. A server has no other way to send an MO message either: `sendMo()` in that test builds
-      one from `submitSmParams()` and `ConcatReference`, neither exported. Choosing `deliver_sm` by
-      `linkEnd` gives MO messages the splitting and checks, keeps one method for one goal, and refuses
-      the options 3.4 has `deliver_sm` leave empty (`scheduleDeliveryTime`, `validityPeriod`).
-
-- [ ] **Error TLVs on a response this library builds.** `buildBody()` in `pdu.ts` writes no body for
-      any non-zero status, so a server cannot answer a `data_sm` with `delivery_failure_reason`,
-      `network_error_code` or `additional_status_info_text`, and a 5.0 peer gets none of its error TLVs.
-      3.4 omits the body on error for `submit_sm_resp` by name; read each response's section before
-      widening it. Reading needs nothing: an error response carrying a body already parses.
-
-- [ ] **PROXY protocol on `server()`.** Behind HAProxy or an AWS NLB every session's remote address is
-      the balancer's, so `authenticate` cannot allow-list by IP and logs name the wrong peer. v1 is
-      text; v2 is binary and the only one an NLB sends. `smpp` accepts v1 from anyone; accept either
-      only from addresses the option names.
-
-- [ ] **`outbind`.** In the command table, handled nowhere: a client cannot take an SMSC's `outbind`
-      and bind back, and `server()` cannot send one. Rare; take it on with a peer that uses it.
-
-- [ ] **Register vendor-specific commands.** 3.4 reserves `command_id` `0x00010200`–`0x000102FF` for
-      SMSC vendors; today one arrives as a `PduRefusedError`. `smpp` has `addCommand()`. The same
-      shape question as registering an encoding.
-
-### Encodings
-
-- [ ] **Register a custom encoding.** Maintainer's ask, 2026-09-14. `EncodingName` is a closed union
-      of three (`defs/encodings.ts`). An entry needs a name, a `data_coding`, `encode`, `decode`,
-      `match`, whether `detect()` may pick it, and enough for `splitMessage()` to budget a segment
-      without halving a character. Take encodings as a client or server option rather than mutating a
-      module table as `smpp` does, so two sessions in one process cannot disagree about a name. A taken
-      name is an `err`. Settle `consts.ENCODING`'s names first. An entry may claim a `data_coding` a
-      built-in already uses, maintainer's call, 2026-09-14: an SMSC whose default alphabet, 0x00, is
-      Latin-1 needs Latin-1 written and read under it
-      ([#23](https://github.com/larvit/larvitsmpp/issues/23)). The entry then owns that coding on its
-      session both ways: `encodingByDataCoding()` resolves to it, `detect()` tries it in the
-      built-in's place, and naming the displaced built-in is an `err` naming the entry. Two entries
-      on one coding are an `err`. Settle whether a claim on 0x00 reaches the class groups
-      `messageClassEncoding()` reads GSM 7-bit from, which `flash` writes under.
-
-- [ ] **The alphabets SMPP 3.4 names that no encoding carries.** `consts.ENCODING` lists the
-      `data_coding` ids (5.2.19); only `ASCII`, `LATIN1` and `UCS2` can be sent. Those with a published
-      definition, and what each costs:
-      - 0x01 IA5 (ITU-T T.50, ASCII in practice): trivial.
-      - 0x06 ISO-8859-5 (Cyrillic) and 0x07 ISO-8859-8 (Hebrew): 96-entry tables.
-      - 0x05 JIS X 0208, 0x0D JIS X 0212, 0x0A ISO-2022-JP and 0x0E KS C 5601: two-octet sets.
-        `TextDecoder` reads them through ICU — EUC-JP and EUC-KR once each octet's high bit is set,
-        `iso-2022-jp` as is; checked for JIS X 0208 and KS C 5601 on Node 24.18.0. Nothing built in
-        encodes them, so ship tables or build the reverse map on first use by decoding the 94×94 grid.
-        A Node without full ICU throws from `new TextDecoder()`, which hard rule 1 wraps into an `err`.
-      - 0x09 pictogram has no published definition; leave it out.
-
-- [ ] **GSM 7-bit national language shift tables.** 3GPP TS 23.038 defines them for Turkish, Spanish
-      (single shift only), Portuguese and ten Indian languages — Bengali, Gujarati, Hindi, Kannada,
-      Malayalam, Oriya, Punjabi, Tamil, Telugu and Urdu — selected per message by UDH elements 0x25
-      (locking) and 0x24 (single). They keep that text near GSM's segment size instead of UCS2's 67
-      characters. Reading means honouring those elements in `decodeMessage()`; sending means `detect()`
-      picking a table, with each element's 3 octets off the segment budget. `smpp` has Turkish, Spanish
-      and Portuguese, used only when the caller writes the UDH.
-
-- [ ] **Packed GSM 7-bit, opt-in.** Everything goes out unpacked, SMPP's convention (AGENTS.md, "GSM
-      7-bit is sent unpacked"); go-smpp carries a packed codec for SMSCs that want septets. Find an SMSC
-      that needs it before building it.
-
-- [ ] **`consts.ENCODING` spells five alphabets twice.** `CYRILLIC`/`ISO_8859_5`,
-      `HEBREW`/`ISO_8859_8`, `JIS`/`X_0208_1990`, `EXTENDED_KANJI_JIS`/`X_0212_1990` and
-      `LATIN1`/`ISO_8859_1`; `FLASH` is a message class, not an alphabet. One name each before
-      registration starts taking names. A breaking change to an export.
-
-### Observability
-
-- [ ] **Metrics.** Inbound traffic has `data`, `incomingPdu` and `incomingPduObj`; outbound has no
-      event, and nothing counts requests in flight, queued for a window slot, waiting for a link, or
-      unanswered. `smpp` and `@semyonf/smpp` emit `metrics`; cloudhopper keeps per-session counters. An
-      `outgoingPdu`/`outgoingPduObj` pair mirrors the inbound events; the counters can be one read-only
-      snapshot, read from the owner of each count rather than a second tally that can drift.
-
-### Packaging, tests and CI
-
-- [ ] **Ship `src`, so the source maps lead somewhere.** Maintainer's call, 2026-09-14. `sourceMap`
-      and `declarationMap` write maps whose `sources` are `../src/*.ts`, but `files` publishes only
-      `dist`, so 82 of the package's 167 files, 225 KB of its 555 KB unpacked, point at nothing. Adding
-      `src` (41 files, 209 KB) makes go to definition land in the TypeScript, and lets a debugger or
-      `--enable-source-maps` show it.
-
-- [ ] **A coverage report and a floor in the gate.** `node --test --experimental-test-coverage
-      --test-coverage-include='src/**' test/*.test.ts` on Node 24.18.0, 2026-09-14: 98.77% lines,
-      93.85% branches, 98.28% functions. Gate at 98, 93 and 98 with `--test-coverage-lines`,
-      `--test-coverage-branches` and `--test-coverage-functions`, which Node 22 and later take — a job
-      of its own on 24, since the matrix runs compiled JavaScript — and add it to `main`'s required
-      checks. Raise the floor as coverage rises; never lower it.
-
-- [ ] **Mutation testing.** `@stryker-mutator/tap-runner` runs `node:test` suites and measures whether
-      a test notices a change, which coverage cannot; `@semyonf/smpp` runs Stryker in CI. The session
-      suites are timer-heavy, so start with the codec and the encodings.
-
-- [ ] **A Node-RED node, as a package of its own.** `@leissner/node-red-smpp` is the only SMPP node in
-      the Node-RED library, and by a read of its source it never parses a receipt and never answers the
-      SMSC's `enquire_link`. Its UI is a fair list of what operators set. It builds on this package,
-      never inside it.
-
 ## Declined
 
-- **CommonJS.** ESM only, maintainer's call reaffirmed 2026-09-14, though `node-smpp-next` ships both.
-  `require()` of an ES module works unflagged from Node 20.19 and 22.12.
+- **Merge state surviving a process restart.** Declined by AGENTS.md goal 7, maintainer's call,
+  2026-09-02. A restart loses every incomplete receipt group and a peer has no reason to resend one it
+  already had answered, so the loss is real — but surviving it means handing the application the merge
+  state to persist, which the scope floor covers as squarely as holding the state here would, and
+  which publishes the shape of `DlrMerger`'s groups against goal 6. Nothing is foreclosed: the seam
+  can still be added after 0.5.0 as a minor.
 
-## An optional store
-
-- [ ] **Pooling, and state that survives a restart, through an optional store.** Maintainer's call,
-      2026-09-14. It replaces two declines — merge state surviving a restart, and a pool of sessions —
-      and AGENTS.md goal 8 was rewritten for it. Big: design before code.
-      - **What it holds.** Receipts still awaited and the groups `DlrMerger` collects. Segments of a
-        message already answered but not yet whole, which the peer will not send again (goal 2). The
-        concatenation reference, so a restart does not reuse one. For a pool, the ids every session
-        sent, since an SMSC may deliver a receipt on any bind of the account, and the
-        messages-per-second budget the sessions share.
-      - **What it cannot hold.** A response belongs to the link its request arrived on, so a message
-        left unanswered at a restart stays unanswerable; the peer's own timeout settles it.
-      - **Pooling.** Several sessions, in one process or many, behind one send: a message goes to a
-        bound session with a free window slot, all its segments on that one. In one process the
-        in-memory store is enough; across processes the application supplies one.
-      - **The interface.** Narrow, with keys and records of the library's own making, versioned, with
-        expiry: a record from an older version is read or refused, never misread, and `DlrMerger`'s
-        group shape is never published (goal 7). What processes share needs an atomic operation —
-        compare-and-set or increment — since get-then-set races.
-      - **Adapters live elsewhere.** Redis, Postgres or SQLite stores are packages of their own; this
-        one ships the interface and the in-memory store, and no runtime dependency (goal 9).
-      - **When the store fails.** Open: a send whose awaited receipt cannot be recorded is refused, or
-        sent and reported as undetermined (goal 2); a pool whose store is down stops, or falls back to
-        memory. Either way, an application that supplied no store never waits on one.
-      - **One spelling.** A store-backed cap and the limiter hook both reach a limit shared between
-        processes; settle which owns that case before building the second.
+- **Throughput throttling — a TPS cap, and backing off on `ESME_RTHROTTLED`.** Declined by AGENTS.md
+  goal 7: an operator's rate limit is scoped to the account, while the widest thing this library owns
+  is a session, so a bucket here cannot see a second process binding the same account and is wrong in
+  exactly the case it exists for. `sendSms()` surfaces `ESME_RTHROTTLED` to the caller instead, and
+  `maxOutstanding` stays — a window slot frees on the peer's next response, which is self-limiting in
+  a way a rate ceiling is not.