@plurnk/plurnk-providers 1.26.0 → 1.28.0

package/.env.defaults

@@ -21,6 +21,10 @@ PLURNK_PROVIDERS_MAX_CONCURRENCY=-1
 # Reserved output INCLUDING reasoning: positive tokens or context percentage in (0,100).
 # Known model limits cap it; input capacity is derived. Exact-counting calls may grant unused context.
 PLURNK_PROVIDERS_OUTPUT_BUDGET=35%
+# The least response room any request keeps, in the same units; capped by the output budget.
+# The context window less this floor is the input wall: a packet past the reservation is still sent with
+# its grant flexed down to the remainder; only a packet past the wall cannot be sent.
+PLURNK_PROVIDERS_OUTPUT_FLOOR=10%
 # Optional reasoning subset of output, using the same units; must be smaller than total output.
 # Unset/empty = provider-adaptive budget, not reasoning off.
 # PLURNK_PROVIDERS_REASONING_BUDGET=8192
@@ -99,21 +103,14 @@ PLURNK_PROVIDERS_CACHE_WRITE_POLICY=stable-system
 # Supported keys: input, output, reasoning, cacheRead, cacheWrite. This is estimation, not billing truth.
 # PLURNK_PROVIDERS_COST=input=0.22,output=0.66,reasoning=0.66,cacheRead=0.007
 
-# --- Connectivity (milliseconds; each deadline accepts 0 = off) ---
+# --- Connectivity (milliseconds; 0 = off) ---
 # Logical generation deadline, including admission queues, physical attempts and retry delays.
 PLURNK_PROVIDERS_OPERATION_TIMEOUT=2700000
-# Deadline per physical generation attempt; a streamed attempt is bounded only until content flows.
+# Discovery and tokenizer HTTP requests only; never a generation deadline.
 PLURNK_PROVIDERS_FETCH_TIMEOUT=600000
-# Wait from dispatch (after admission, before response headers) to first semantic content;
-# metadata/empty deltas do not count.
-PLURNK_PROVIDERS_FIRST_CONTENT_TIMEOUT=180000
-# Maximum silence between semantic content chunks after content begins.
-PLURNK_PROVIDERS_STREAM_IDLE_TIMEOUT=120000
-# A line of 16+ characters streamed this many times stops the response as a repetition; 0 = off.
-PLURNK_PROVIDERS_REPEATED_LINE_LIMIT=32
 # A response billed for this many more output tokens than it streamed characters (text and reasoning)
-# never delivered its output; it fails as output_dropped and is re-issued (a tool-call finish always
-# does). Reasoning billed but not streamed is exempt. 0 = off.
+# never delivered its output; it fails as output_dropped. Native tool calls are retained with a
+# warning, not retried for this reason. Reasoning billed but not streamed is exempt. 0 = off.
 PLURNK_PROVIDERS_DROPPED_OUTPUT_TOKENS=32
 # Retries for 429, Retry-After, or X-Should-Retry:true; 0 = none. Other failures return immediately
 # to Core's separate recovery window. Retry-After takes precedence over backoff.

package/README.md

@@ -12,6 +12,9 @@ Ordinary provider behavior is intentionally not reimplemented here:
 - PLURNK owns aliases, generation envelopes, normalized usage and errors,
   evidence capture, first-party metadata, and local endpoint capabilities.
 
+Native tool calling is explicitly disabled through each protocol's standard
+controls; PLURNK operations travel as text ({§provider-native-tools-disabled}).
+
 See [SPEC.md](SPEC.md) for the contract, [the model reference](docs/models.md)
 for what a model is told about routes and generation, and
 [.env.defaults](.env.defaults) for every knob.
@@ -114,6 +117,11 @@ flowchart TD
 - **Output** accepts positive tokens or a percentage of context; known model
   limits cap it. An explicit reasoning budget must be smaller than total output.
   Leaving the reasoning budget unset does not disable reasoning.
+- **Output floor** (`PLURNK_PROVIDERS_OUTPUT_FLOOR`, shipped `10%`) is the least
+  response room any request keeps, never above the output budget. The window less
+  the floor is the input wall: a packet past the budget is still sent with its
+  grant flexed down to the remainder; only a packet past the wall cannot be sent
+  ({§provider-output-floor}).
 - **Context** is derived from the live endpoint or catalog. An operator cap can
   shrink known capacity or declare unknown capacity, never enlarge known limits.
   Prompt projection and initialization reasoning READ limits are separate Core policy.
@@ -188,9 +196,10 @@ opt-in for endpoints that emit a leading `<think>` envelope, not a reasoning swi
 | Concern | Boundary |
 | --- | --- |
 | Inference concurrency | Optional per-endpoint, process-local admission; queued calls remain cancellable. The default is unrestricted ({§provider-inference-admission}). |
-| Attempt, first-content, idle deadlines | Provider transport; a timeout surfaces a failure, not a fabricated response. |
+| Generation deadline | One whole-call allowance across queueing, requests and retries; caller cancellation still applies. No first-content or inter-chunk cutoff. |
+| Discovery/tokenizer HTTP | Separate non-generation fetch bound; does not limit inference. |
 | Provider-directed waits | Bounded retries honor `Retry-After`; other recoverable failures return to Core. |
-| Loop recovery | Core reissues within its recovery window, then parks for a prompt or wake. |
+| Loop recovery | Core reissues within its recovery window, then parks attended loops or concludes unattended loops on the failure. |
 | Cache affinity | Stable Worker identity through documented provider controls; not a cache-hit guarantee. |
 | Explicit cache writes | Separate policy on supported routes; can affect billing. |
 | Service tier | Explicit route choice that may change price and availability. |

package/SPEC.md

@@ -95,12 +95,12 @@ fails hard.
 
 §provider-capacity-admission `assessRequestCapacity` intersects every known
 physical input constraint: independent `maxInputTokens` and
-`contextWindow - outputBudget`. Its result is `admit`, `reject`, or `defer` and
+`contextWindow - outputFloor`, the input wall ({§provider-output-floor}). Its result is `admit`, `reject`, or `defer` and
 retains the complete limit and measurement evidence. Exact fit admits; exact
-overflow rejects. A proven upper bound admits only when it fits. Unknown limits,
+overflow of the wall rejects; a prompt between the curation reservation and the wall is admitted with its grant flexed down. A proven upper bound admits only when it fits. Unknown limits,
 an upper bound above a limit, estimates, and unavailable measurements defer to
-the upstream provider as capacity oracle. The same stable intersection is
-exposed as `inputCapacity`; `null` means the available limits cannot establish
+the upstream provider as capacity oracle. The curation reservation's intersection, `maxInputTokens` and `contextWindow - outputBudget`, is
+exposed as `inputCapacity`, the line Core packs the room against; the wall's, as `inputWall`; `null` means the available limits cannot establish
 one. A known combined context and output budget must leave positive input
 capacity. Consumers may display or use that fact as policy, but MUST NOT
 substitute their own content heuristic for request-shaped admission.
@@ -167,6 +167,25 @@ message count, grammar presence, worker identity, or another incidental request
 shape. The signal never enters model-facing messages. Generic
 provider callers MAY omit it; Core supplies it for every model call.
 
+§provider-native-tools-disabled Generation requests declare no native tools and
+select the SDK's portable `none` tool choice, for emissions and BARE alike.
+The built-in transports preserve that intent after SDK serialization:
+
+| Protocol | No-tool wire representation |
+| --- | --- |
+| OpenAI-compatible Chat Completions and Responses | `tools: []`, `tool_choice: "none"` |
+| Anthropic Messages | `tools: []`, `tool_choice: {"type":"none"}` |
+| Gemini GenerateContent | `tools: []`, `toolConfig: {"functionCallingConfig":{"mode":"NONE"}}` |
+| Bedrock Converse | No tool configuration; the protocol has no `none` choice. |
+| Installed SDK provider plugins | Portable `none` intent; the plugin owns its wire projection. |
+
+This is a text-generation contract, not provider/model tuning or prompt
+teaching. Callers cannot supply tools or override the choice through request
+extensions. Do not add deprecated function-call fields, dummy tools, or fields
+from another protocol. The opt-out neither constrains arbitrary prose nor
+proves that an endpoint honors it; unexpected native calls retain their full
+evidence and follow {§provider-native-tool-calls}.
+
 A successful return carries the model's raw content and reasoning, normalized
 finish reason, model identity, its ordered {§provider-request-accounting}, the
 request's `ProviderRequestCapacity`, opaque evidence, optional metadata, and
@@ -175,6 +194,12 @@ accounting evidence. The provider transports and observes model
 output; it never retries, discards, or repairs an otherwise completed exchange
 because PLURNK grammar did not accept it.
 
+§provider-generation-completion Stream consumption follows the provider's
+completion boundary, subject to caller cancellation and configured deadlines.
+Repeated content or reasoning does not interrupt a response or request another
+generation. Preserve the complete response, its finish reason, and trailing
+usage and charge evidence before returning it to the consumer.
+
 §provider-request-observer When a consumer supplies the request observer, the
 adapter opens one durable identity through it immediately before each physical
 I/O and settles that identity with the resulting
@@ -184,6 +209,15 @@ observer is a durability sink, not an alternate evidence representation; the
 same ordered records remain on the final generation result or
 error.
 
+The adapter also settles {§provider-request-evidence}. A successful request
+retains its normalized transport response; a failed request retains received
+SDK raw chunks (or the error response body), partial content and reasoning,
+available response identity/headers, usage/charge evidence, and its error and
+cancellation cause. Failure capture is independent of successful raw-body
+capture. These are the SDK's received values, not a claim of byte-for-byte
+HTTP capture. Settlement is durable before returning the failure; a process
+crash before settlement leaves a pending request, not an invented response.
+
 §provider-reasoning-observer When a consumer supplies `observeReasoning`, the
 provider synchronously delivers each exact, ordered, nonempty readable-reasoning
 delta as it becomes available. A transport without incremental reasoning emits
@@ -446,7 +480,7 @@ The universal groups are:
 - reasoning activation and optional explicit budget;
 - explicit reasoning response-content style;
 - decode tuning;
-- operation, physical-attempt, first-content, stream-idle, retry, and probe budgets;
+- whole-call generation, discovery/tokenizer HTTP, retry, and probe budgets;
 - local GBNF and llama-server capability pins;
 - context-window and generation-envelope overrides;
 - provider-documented cache affinity and explicit cache-write policy;
@@ -503,7 +537,8 @@ Provider and model facts resolve independently:
 | Context window | Catalog metadata or local endpoint probe. | `PLURNK_PROVIDERS_CONTEXT_WINDOW`. | Minimum when both exist; sole value otherwise. Cataloged cloud miss fails construction; compatible probe miss remains `null` with one warning. |
 | Maximum input | Catalog `limit.input`; no generic live probe. | None. | Catalog value or `null`; never reconstructed from context and output. |
 | Maximum output | Catalog `limit.output`; no generic live probe. | None. | Minimum of catalog value and effective context, or `null`. |
-| Total output budget | None. | `PLURNK_PROVIDERS_OUTPUT_BUDGET`. | Curation reservation: percentage of effective context or absolute count, capped by known context/output limits; a call may only tighten it. The response grant may expand under {§provider-flexed-allowance}. |
+| Total output budget | None. | `PLURNK_PROVIDERS_OUTPUT_BUDGET`. | Curation reservation: percentage of effective context or absolute count, capped by known context/output limits; a call may only tighten it. The response grant flexes under {§provider-flexed-allowance}. |
+| Output floor | None. | `PLURNK_PROVIDERS_OUTPUT_FLOOR`, shipped as `10%`. | The least response room any request keeps: percentage of effective context or absolute count, capped by the output budget; `window − floor` is the input wall ({§provider-output-floor}). |
 | Effort | Catalog `reasoning_options` intersected with the installed adapter; explicit adapter declaration for uncataloged routes. | `PLURNK_PROVIDERS_EFFORT`, initially; durable worker selection thereafter. | A supported member of {§effort-wire}, projected under {§provider-effort}. The shipped selection is `adaptive`. |
 | Reasoning budget | None. | Optional `PLURNK_PROVIDERS_REASONING_BUDGET`. | Percentage of effective context or absolute count; valid only as a strict subset of total output and effective unless reasoning is `off`. |
 | Cost override | None. | Optional `PLURNK_PROVIDERS_COST`. | {§operator-cost-override} — comma-separated `key=value` per-1M-token USD rates over `input, output, reasoning, cacheRead, cacheWrite`; merges over the Models.dev catalog block (the catalog is the starting point), alias-scoped like every knob. Without catalog rates the override must declare `input` and `output`. The cost estimate's `source` names the override; a provider-reported response cost still outranks any estimate. |
@@ -643,16 +678,6 @@ reasoning enclosure only after preserving grammar evidence. Process-wide
 llama-server flags are fallback server configuration, not part of the PLURNK
 contract and need not be synchronized with an alias.
 
-§repetition-stop **A response that repeats itself is stopped.** A degenerate model can stream one line
-without end; content keeps arriving, so no silence deadline fires, and the call runs to its output
-allowance. The transport counts the complete lines of each streamed channel, text and reasoning, and
-when one line of 16 or more characters has appeared `PLURNK_PROVIDERS_REPEATED_LINE_LIMIT` times it stops
-reading: the stream is cancelled, the physical request settles with whatever usage the chunks carried,
-and the call fails as `repetition` (422, not retryable) carrying the partial attempt and the sentence
-"The response repeated one line N times and was stopped: `<line>`". The consumer reads it as an invalid
-emission, never as a provider failure to recover. Measured: a granite-4.2-8b run repeated one line 358
-times across 45,875 tokens and 9 min 42 s; a limit of 32 stops it 14.6% of the way in (#876).
-
 §llama-reasoning-request The allowance is cumulative across the complete response. Opening a second or
 later reasoning block does not replenish it. Template parsing, the reasoning
 sampler, normalized usage, and the returned reasoning channel MUST agree on that
@@ -730,17 +755,18 @@ request-accounting row. Capacity failures are not connectivity failures and are
 never retried by the provider scheduler; bounded packet recovery belongs to the
 consumer.
 
-§provider-flexed-allowance **Overflow is tolerated, never invited.** The
-configured output budget is the floor curation packs the input against
+§provider-flexed-allowance **The grant is the window's remainder, between the floor and the model's limit.** The
+configured output budget is the reservation curation packs the input against
 (`window − outputBudget`), and it is the only allowance the model is ever
-shown — the packet's disclosed `tokensResponseMax` stays the configured floor
-so the response discipline it teaches never varies. The wire is quietly
-generous: a transport with an exact prompt count (llama-server's input-token
-endpoint) grants `max(outputBudget, window − promptTokens − 256)`, capped by
-the model's own output limit, so a response overflowing the disclosed floor
-completes whenever the window's unclaimed room can hold it. Every other
-measurement kind — and any pool — keeps the floor, because an estimate proves
-nothing about the true remainder. A response exceeding even the grant is cut,
+shown — the packet's disclosed `tokensResponseMax` stays the configured reservation
+so the response discipline it teaches never varies. The wire grants the remainder: with an exact prompt count
+(llama-server's input-token endpoint) `window − promptTokens − 256`, clamped to no less than the
+output floor and no more than the model's own output limit, so a response overflowing the disclosed reservation
+completes whenever the window's unclaimed room can hold it, and a prompt that ate into the reservation still leaves the floor. Every other
+measurement kind — and any pool — grants the reservation, clamped by the same remainder taken from the prompt's estimate so the wire never asks the window for more than it has; an estimate proves
+nothing about the true remainder, so it never rejects.
+
+§provider-output-floor **The floor is the wall's other side.** `PLURNK_PROVIDERS_OUTPUT_FLOOR`, shipped as `10%`, is the least response room any request keeps, capped by the output budget; `window − floor`, intersected with `maxInputTokens`, is the physical input wall exposed as `inputWall`. A request under the curation reservation is the ordinary case; a request between the reservation and the wall is admitted with its grant flexed down to the remainder; an exact prompt over the wall is rejected before I/O; an estimate over the wall defers to the upstream provider as capacity oracle with its grant at the floor. Core consults the wall, never the reservation, to end a loop ({§context-wall} in the core contract). A response exceeding even the grant is cut,
 and the notice names the true per-call grant (`capacity.responseMax` on the
 response) — the tolerance's honest edge, never worse than the fixed allowance
 it forgives.
@@ -772,29 +798,30 @@ unconfigured. Their grammar protection remains at {§provider-grammar-transport}
 §provider-connectivity The provider adapter owns one attempt scheduler around
 the complete generation exchange; SDK-internal retries are disabled.
 `PLURNK_PROVIDERS_RETRY_ATTEMPTS=N` permits at most `N + 1` physical requests.
-The layers are independent and a configured value of zero disables only that
-deadline:
+Generation has one configurable deadline; `0` disables it. Caller cancellation
+and the caller's task allowance remain authoritative. Initial silence, a pause
+between chunks, and buffered output do not acquire separate deadlines.
 
 | Layer | Operator knob | Boundary | Expiry |
 | --- | --- | --- | --- |
 | Operation | `PLURNK_PROVIDERS_OPERATION_TIMEOUT` | Complete logical call, including admission queueing, every attempt and retry delay. | Final `deadline_exceeded` Problem at 504 with `timeoutPhase=operation`; not retried inside this operation. Consumer recovery is separate. Enforced as a race, not only the advisory signal, so a wedged transport that never observes the abort cannot hang the loop past the deadline (#505); a well-behaved transport unwinds within a short grace and settles its own attempt evidence first. |
-| Attempt | `PLURNK_PROVIDERS_FETCH_TIMEOUT` | One physical generation request. A non-streamed request is bounded through response consumption; a streamed one only until its first semantic content. After content begins, stream-idle bounds silence and the operation deadline still bounds the whole call. | Surfaced `network_failure` with `timeoutPhase=attempt`; never transport-retried (#479) — the consumer's recovery owns re-issue. |
-| First content | `PLURNK_PROVIDERS_FIRST_CONTENT_TIMEOUT` | Dispatch through first semantic model content ({§provider-first-content-at-dispatch}); metadata, empty deltas, and transport activity do not satisfy it. | Surfaced `network_failure` with `timeoutPhase=first_content`; never transport-retried (#479). |
-| Stream idle | `PLURNK_PROVIDERS_STREAM_IDLE_TIMEOUT` | Silence between semantic content chunks after content begins. | Surfaced `network_failure` with `timeoutPhase=stream_idle`; never transport-retried (#479). |
-| Repeated line | `PLURNK_PROVIDERS_REPEATED_LINE_LIMIT` | A streamed line of 16 or more characters repeated this many times; `0` disables. | Stopped as `repetition` ({§repetition-stop}); never transport-retried. |
-
-§provider-first-content-at-dispatch The first-content deadline is armed when the
-admitted attempt is dispatched, before response headers, so an endpoint that accepts
-the request and never answers fails at `timeoutPhase=first_content` instead of holding
-the attempt for the whole attempt deadline. It still begins only after admission
-({§provider-inference-admission}).
+| Discovery/tokenizer HTTP | `PLURNK_PROVIDERS_FETCH_TIMEOUT` | Non-generation endpoint discovery and tokenizer requests only. | Ordinary HTTP failure at the owning discovery or measurement boundary. Never interrupts generation. |
 
 Settled calls remove their deadline timers and cancellation subscriptions.
 
 Caller cancellation spans the operation and preserves the caller's reason.
+
+§provider-cancellation-evidence Received usage/charge evidence and available
+response identity survive cancellation and deadline normalization through
+physical-request settlement, independently of optional raw-body capture.
+This includes SDK abort events and non-Error cancellation reasons. Concurrent
+requests keep independent evidence even when cancelled by the same scope.
+Absent evidence remains absent; stopping a request neither invents usage nor
+retries generation to obtain it. Partial output is never a completed response.
+
 A 2xx exchange whose body cannot be processed (a provider invalid-response)
 classifies as the non-retryable 502 on the first failure unless an explicit
-`x-should-retry` directive says otherwise (#479). Inner deadline failures surface on the first failure; when a
+`x-should-retry` directive says otherwise (#479). When a
 directive-driven retry sequence exhausts, `attempts` and `retryExhausted`
 are added and the classification is final. Every admitted physical attempt opens
 and settles exactly one ordered {§provider-request-accounting} record, including
@@ -988,7 +1015,7 @@ alias scoping. Zero, other negative values and non-integers are invalid.
 | Admission | FIFO among live waiters. The lease begins before the physical request observer and ends after the complete response or transport failure settles, including streamed bodies. |
 | Cancellation | A queued abort removes that waiter and preserves the caller's reason. It opens no physical request or accounting row. In-flight cancellation signals the transport; capacity is released when that attempt unwinds. A transport still running despite abort does not authorize exceeding the limit. |
 | Retries | Backoff holds no lease. Each retry rejoins admission as a new physical attempt. |
-| Deadlines | The existing operation deadline includes queueing. Attempt, first-content and stream-idle deadlines begin only after admission; queued work is not a stalled stream. |
+| Deadline | The whole-call operation deadline includes queueing, every physical attempt and retry wait. Transport adds no generation deadline. |
 | Scope | Workers, tools, messages, waits, token measurement and endpoint discovery are not serialized by inference admission. Independent endpoints progress independently. No cross-process or machine-wide capacity guarantee is implied. |
 
 Admission neither changes worker lifecycle nor selects another model or endpoint.

package/dist/AiSdkProvider.d.ts

@@ -28,9 +28,6 @@ export type AiSdkProviderConfig = {
     attributions?: (context: PluginAttributionContext) => PluginAttribution;
     fetchTimeoutMs: number;
     operationTimeoutMs: number;
-    firstContentTimeoutMs: number;
-    streamIdleTimeoutMs?: number;
-    repeatedLineLimit?: number;
     droppedOutputTokens?: number;
     headers?: Record<string, string>;
     fetch?: ProviderFetch;
@@ -39,6 +36,7 @@ export type AiSdkProviderConfig = {
     maxInputTokens?: number | null;
     maxOutputTokens?: number | null;
     outputBudget?: number | null;
+    outputFloor?: number | null;
     reasoningBudget?: number | null;
     supportedEfforts?: readonly Effort[];
     adaptiveEffort?: NativeEffort | "provider-default";
@@ -92,9 +90,11 @@ export default class AiSdkProvider implements Provider {
     get maxInputTokens(): number | null;
     get maxOutputTokens(): number | null;
     get outputBudget(): number | null;
+    get outputFloor(): number | null;
     get reasoningBudget(): number | null;
     get supportedEfforts(): readonly Effort[];
     get inputCapacity(): number | null;
+    get inputWall(): number | null;
     get model(): string;
     get servedModel(): string | undefined;
     get requiresOutputBudget(): boolean | undefined;

package/dist/AiSdkProvider.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"AiSdkProvider.d.ts","sourceRoot":"","sources":["../src/AiSdkProvider.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,WAAW,EAAmB,sBAAsB,EAAE,QAAQ,EAAmB,sBAAsB,EAAE,oBAAoB,EAA6B,uBAAuB,EAA6B,gBAAgB,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,YAAY,CAAC;AACxR,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAE7D,OAAO,KAAK,EAAe,SAAS,EAAE,MAAM,IAAI,CAAC;AACjD,OAAO,KAAK,EAAE,aAAa,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAEtE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAEhD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAMxC,OAAO,KAAK,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAQvF,OAAO,KAAK,aAAa,MAAM,oBAAoB,CAAC;AACpD,OAAO,KAAK,kBAAkB,MAAM,yBAAyB,CAAC;AAE9D,MAAM,MAAM,aAAa,GAAG,OAAO,UAAU,CAAC,KAAK,CAAC;AAwBpD,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,UAAU,CAAC;AAE3D,MAAM,MAAM,YAAY,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAI3E,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,UAAU,CAAC;AAE/C,MAAM,MAAM,aAAa,GACnB;IAAE,QAAQ,CAAC,MAAM,EAAE,QAAQ,GAAG,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GAC7D;IAAE,QAAQ,CAAC,MAAM,EAAE,iBAAiB,CAAC;IAAC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAE/F,MAAM,MAAM,oBAAoB,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,GAAG,SAAS,CAAC,CAAC,CAAC;AAIzF,MAAM,MAAM,mBAAmB,GAAG;IAC9B,kBAAkB,CAAC,EAAE,kBAAkB,CAAC;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,YAAY,CAAC,EAAE,CAAC,OAAO,EAAE,wBAAwB,KAAK,iBAAiB,CAAC;IACxE,cAAc,EAAE,MAAM,CAAC;IACvB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,eAAe,CAAC,EAAE,WAAW,CAAC,aAAa,CAAC,CAAC;IAC7C,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAIrC,cAAc,CAAC,EAAE,YAAY,GAAG,kBAAkB,CAAC;IACnD,6BAA6B,CAAC,EAAE,oBAAoB,CAAC;IAKrD,yBAAyB,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;IACpD,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,sBAAsB,CAAC,EAAE,sBAAsB,CAAC;IAChD,iBAAiB,CAAC,EAAE,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,sBAAsB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;IACzI,YAAY,CAAC,EAAE,CAAC,KAAK,EAAE,aAAa,GAAG,SAAS,KAAK,YAAY,CAAC;IAClE,aAAa,CAAC,EAAE,sBAAsB,CAAC;IACvC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,YAAY,CAAC;IAG5B,aAAa,CAAC,EAAE,aAAa,CAAC;IAG9B,0BAA0B,CAAC,EAAE,oBAAoB,CAAC;IAGlD,gCAAgC,CAAC,EAAE,oBAAoB,CAAC;IAGxD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAI1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAGrB,eAAe,CAAC,EAAE,MAAM,CAAC;IAKzB,WAAW,CAAC,EAAE,MAAM,CAAC;IAIrB,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAG/B,MAAM,EAAE,aAAa,CAAC;IAGtB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAE7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAK1B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAKrB,aAAa,EAAE,MAAM,CAAC;IAItB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAQ1B,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,CAAC,EAAE,OAAO,CAAC;CAOrB,CAAC;AA8CF,MAAM,CAAC,OAAO,OAAO,aAAc,YAAW,QAAQ;;IAuDlD,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC,OAAO,EAAE,wBAAwB,KAAK,iBAAiB,CAAC;IAMjF,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IAE/C,YAAY,MAAM,EAAE,mBAAmB,EA4KtC;IAED,IAAI,aAAa,IAAI,MAAM,GAAG,IAAI,CAAgC;IAClE,IAAI,eAAe,IAAI,WAAW,CAAC,aAAa,CAAC,CAAkC;IACnF,IAAI,cAAc,IAAI,MAAM,GAAG,IAAI,CAAiC;IACpE,IAAI,eAAe,IAAI,MAAM,GAAG,IAAI,CAAkC;IACtE,IAAI,YAAY,IAAI,MAAM,GAAG,IAAI,CAA+B;IAChE,IAAI,eAAe,IAAI,MAAM,GAAG,IAAI,CAAkC;IACtE,IAAI,gBAAgB,IAAI,SAAS,MAAM,EAAE,CAAmC;IAC5E,IAAI,aAAa,IAAI,MAAM,GAAG,IAAI,CAMjC;IACD,IAAI,KAAK,IAAI,MAAM,CAAwB;IAE3C,IAAI,WAAW,IAAI,MAAM,GAAG,SAAS,CAA8B;IAEnE,IAAI,oBAAoB,IAAI,OAAO,GAAG,SAAS,CAAuC;IAItF,IAAI,gBAAgB,IAAI,OAAO,CAA0C;IAEnE,iBAAiB,CACnB,QAAQ,EAAE,SAAS,WAAW,EAAE,EAChC,MAAM,CAAC,EAAE,WAAW,GACrB,OAAO,CAAC,sBAAsB,CAAC,CAqDjC;IAEK,qBAAqB,CACvB,QAAQ,EAAE,SAAS,WAAW,EAAE,EAChC,eAAe,CAAC,EAAE,MAAM,EACxB,MAAM,CAAC,EAAE,WAAW,GACrB,OAAO,CAAC,uBAAuB,CAAC,CAmBlC;IA4BK,QAAQ,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,cAAc,EAAE,gBAAgB,EAAE,QAAQ,EAAE,EAAE,oBAAoB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAygB9K;CAqBJ"}
+{"version":3,"file":"AiSdkProvider.d.ts","sourceRoot":"","sources":["../src/AiSdkProvider.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,WAAW,EAAmB,sBAAsB,EAAE,QAAQ,EAAmB,sBAAsB,EAAE,oBAAoB,EAA6B,uBAAuB,EAA6B,gBAAgB,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,YAAY,CAAC;AACxR,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAE7D,OAAO,KAAK,EAAe,SAAS,EAAE,MAAM,IAAI,CAAC;AACjD,OAAO,KAAK,EAAE,aAAa,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAEtE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAEhD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAMxC,OAAO,KAAK,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAQvF,OAAO,KAAK,aAAa,MAAM,oBAAoB,CAAC;AACpD,OAAO,KAAK,kBAAkB,MAAM,yBAAyB,CAAC;AAE9D,MAAM,MAAM,aAAa,GAAG,OAAO,UAAU,CAAC,KAAK,CAAC;AAwBpD,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,UAAU,CAAC;AAE3D,MAAM,MAAM,YAAY,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAI3E,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,UAAU,CAAC;AAE/C,MAAM,MAAM,aAAa,GACnB;IAAE,QAAQ,CAAC,MAAM,EAAE,QAAQ,GAAG,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GAC7D;IAAE,QAAQ,CAAC,MAAM,EAAE,iBAAiB,CAAC;IAAC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAE/F,MAAM,MAAM,oBAAoB,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,GAAG,SAAS,CAAC,CAAC,CAAC;AAIzF,MAAM,MAAM,mBAAmB,GAAG;IAC9B,kBAAkB,CAAC,EAAE,kBAAkB,CAAC;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,YAAY,CAAC,EAAE,CAAC,OAAO,EAAE,wBAAwB,KAAK,iBAAiB,CAAC;IACxE,cAAc,EAAE,MAAM,CAAC;IACvB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,eAAe,CAAC,EAAE,WAAW,CAAC,aAAa,CAAC,CAAC;IAC7C,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAIrC,cAAc,CAAC,EAAE,YAAY,GAAG,kBAAkB,CAAC;IACnD,6BAA6B,CAAC,EAAE,oBAAoB,CAAC;IAKrD,yBAAyB,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;IACpD,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,sBAAsB,CAAC,EAAE,sBAAsB,CAAC;IAChD,iBAAiB,CAAC,EAAE,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,sBAAsB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;IACzI,YAAY,CAAC,EAAE,CAAC,KAAK,EAAE,aAAa,GAAG,SAAS,KAAK,YAAY,CAAC;IAClE,aAAa,CAAC,EAAE,sBAAsB,CAAC;IACvC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,YAAY,CAAC;IAG5B,aAAa,CAAC,EAAE,aAAa,CAAC;IAG9B,0BAA0B,CAAC,EAAE,oBAAoB,CAAC;IAGlD,gCAAgC,CAAC,EAAE,oBAAoB,CAAC;IAGxD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAI1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAGrB,eAAe,CAAC,EAAE,MAAM,CAAC;IAKzB,WAAW,CAAC,EAAE,MAAM,CAAC;IAIrB,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAG/B,MAAM,EAAE,aAAa,CAAC;IAGtB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAE7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAK1B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAKrB,aAAa,EAAE,MAAM,CAAC;IAItB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAQ1B,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,CAAC,EAAE,OAAO,CAAC;CAOrB,CAAC;AA8CF,MAAM,CAAC,OAAO,OAAO,aAAc,YAAW,QAAQ;;IAqDlD,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC,OAAO,EAAE,wBAAwB,KAAK,iBAAiB,CAAC;IAMjF,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IAE/C,YAAY,MAAM,EAAE,mBAAmB,EA2KtC;IAED,IAAI,aAAa,IAAI,MAAM,GAAG,IAAI,CAAgC;IAClE,IAAI,eAAe,IAAI,WAAW,CAAC,aAAa,CAAC,CAAkC;IACnF,IAAI,cAAc,IAAI,MAAM,GAAG,IAAI,CAAiC;IACpE,IAAI,eAAe,IAAI,MAAM,GAAG,IAAI,CAAkC;IACtE,IAAI,YAAY,IAAI,MAAM,GAAG,IAAI,CAA+B;IAChE,IAAI,WAAW,IAAI,MAAM,GAAG,IAAI,CAA8B;IAC9D,IAAI,eAAe,IAAI,MAAM,GAAG,IAAI,CAAkC;IACtE,IAAI,gBAAgB,IAAI,SAAS,MAAM,EAAE,CAAmC;IAC5E,IAAI,aAAa,IAAI,MAAM,GAAG,IAAI,CAMjC;IAED,IAAI,SAAS,IAAI,MAAM,GAAG,IAAI,CAM7B;IACD,IAAI,KAAK,IAAI,MAAM,CAAwB;IAE3C,IAAI,WAAW,IAAI,MAAM,GAAG,SAAS,CAA8B;IAEnE,IAAI,oBAAoB,IAAI,OAAO,GAAG,SAAS,CAAuC;IAItF,IAAI,gBAAgB,IAAI,OAAO,CAA0C;IAEnE,iBAAiB,CACnB,QAAQ,EAAE,SAAS,WAAW,EAAE,EAChC,MAAM,CAAC,EAAE,WAAW,GACrB,OAAO,CAAC,sBAAsB,CAAC,CAqDjC;IAEK,qBAAqB,CACvB,QAAQ,EAAE,SAAS,WAAW,EAAE,EAChC,eAAe,CAAC,EAAE,MAAM,EACxB,MAAM,CAAC,EAAE,WAAW,GACrB,OAAO,CAAC,uBAAuB,CAAC,CAoBlC;IA4BK,QAAQ,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,cAAc,EAAE,gBAAgB,EAAE,QAAQ,EAAE,EAAE,oBAAoB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAof9K;CAqBJ"}

package/dist/AiSdkProvider.js

@@ -15,7 +15,7 @@ import { emitWarningOnce } from "./warnings.js";
 import { resolveProviderCost } from "./cost.js";
 import { validateProviderRequestAccounting } from "./accounting.js";
 import { validateProviderUsage } from "./usage.js";
-import { assessRequestCapacity, effectiveInputCapacity, effectiveOutputBudget, effectiveReasoningBudget } from "./capacity.js";
+import { assessRequestCapacity, effectiveInputCapacity, effectiveInputWall, effectiveOutputBudget, effectiveOutputFloor, effectiveReasoningBudget } from "./capacity.js";
 import { nativeFixedEffort } from "./reasoning-effort.js";
 import AiSdkRequestBody from "./AiSdkRequestBody.js";
 import LeadingReasoning from "./LeadingReasoning.js";
@@ -89,9 +89,6 @@ export default class AiSdkProvider {
     #languageModel;
     #fetchTimeoutMs;
     #operationTimeoutMs;
-    #firstContentTimeoutMs;
-    #streamIdleTimeoutMs;
-    #repeatedLineLimit;
     #droppedOutputTokens;
     #headers;
     #fetch;
@@ -103,6 +100,7 @@ export default class AiSdkProvider {
     #maxInputTokens;
     #maxOutputTokens;
     #outputBudget;
+    #outputFloor;
     #reasoningBudget;
     #additiveReasoningProvider;
     #effort;
@@ -156,8 +154,6 @@ export default class AiSdkProvider {
         for (const [name, value] of [
             ["fetchTimeoutMs", config.fetchTimeoutMs],
             ["operationTimeoutMs", config.operationTimeoutMs],
-            ["firstContentTimeoutMs", config.firstContentTimeoutMs],
-            ["streamIdleTimeoutMs", config.streamIdleTimeoutMs],
         ]) {
             if (value !== undefined && (!Number.isInteger(value) || value < 0 || value > MAX_TIMER_MS)) {
                 throw new Error(`${config.source ?? "provider"}: ${name} must be an integer from 0 through ${MAX_TIMER_MS} milliseconds`);
@@ -165,9 +161,6 @@ export default class AiSdkProvider {
         }
         this.#fetchTimeoutMs = config.fetchTimeoutMs;
         this.#operationTimeoutMs = config.operationTimeoutMs;
-        this.#firstContentTimeoutMs = config.firstContentTimeoutMs;
-        this.#streamIdleTimeoutMs = config.streamIdleTimeoutMs;
-        this.#repeatedLineLimit = config.repeatedLineLimit ?? 0;
         this.#droppedOutputTokens = config.droppedOutputTokens ?? 0;
         this.#headers = config.headers ?? {};
         this.#fetch = config.fetch ?? ((input, init) => globalThis.fetch(input, init));
@@ -176,6 +169,7 @@ export default class AiSdkProvider {
         this.#maxInputTokens = config.maxInputTokens ?? null;
         this.#maxOutputTokens = config.maxOutputTokens ?? null;
         this.#outputBudget = config.outputBudget ?? null;
+        this.#outputFloor = config.outputFloor ?? null;
         this.#reasoningBudget = config.reasoningBudget ?? null;
         this.#additiveReasoningProvider = config.additiveReasoningProvider;
         this.#effort = config.effort;
@@ -271,12 +265,15 @@ export default class AiSdkProvider {
             ["maxInputTokens", this.#maxInputTokens],
             ["maxOutputTokens", this.#maxOutputTokens],
             ["outputBudget", this.#outputBudget],
+            ["outputFloor", this.#outputFloor],
             ["reasoningBudget", this.#reasoningBudget],
         ]) {
             if (value !== null && (!Number.isSafeInteger(value) || value <= 0)) {
                 throw new Error(`${this.#source}: ${name} must be a positive safe integer or null`);
             }
         }
+        // {§provider-output-floor}: the floor is never above the output budget.
+        this.#outputFloor = effectiveOutputFloor({ configured: this.#outputFloor, outputBudget: this.#outputBudget });
         if (this.#reasoningBudget !== null
             && this.#outputBudget !== null
             && this.#reasoningBudget >= this.#outputBudget) {
@@ -316,6 +313,7 @@ export default class AiSdkProvider {
     get maxInputTokens() { return this.#maxInputTokens; }
     get maxOutputTokens() { return this.#maxOutputTokens; }
     get outputBudget() { return this.#outputBudget; }
+    get outputFloor() { return this.#outputFloor; }
     get reasoningBudget() { return this.#reasoningBudget; }
     get supportedEfforts() { return this.#supportedEfforts; }
     get inputCapacity() {
@@ -325,6 +323,14 @@ export default class AiSdkProvider {
             outputBudget: this.#outputBudget,
         });
     }
+    // {§provider-output-floor}: the physical input wall, the window less the floor.
+    get inputWall() {
+        return effectiveInputWall({
+            contextWindow: this.#contextWindow,
+            maxInputTokens: this.#maxInputTokens,
+            outputFloor: this.#outputFloor,
+        });
+    }
     get model() { return this.#model; }
     // Backend's self-reported served id; undefined when unprobed/unknown.
     get servedModel() { return this.#servedModel; }
@@ -392,6 +398,7 @@ export default class AiSdkProvider {
             maxInputTokens: this.#maxInputTokens,
             maxOutputTokens: this.#maxOutputTokens,
             outputBudget,
+            outputFloor: this.#outputFloor,
             reasoningBudget,
             measurement: await this.countPromptTokens(messages, signal),
         });
@@ -436,10 +443,10 @@ export default class AiSdkProvider {
             if (capacity.prompt.kind !== "exact") {
                 throw new TypeError(`${this.#source}: only an exact prompt measurement may reject capacity`);
             }
-            throw new ProviderError(this.#source, "capacity_exceeded", `The exact provider request uses ${capacity.prompt.tokens} input tokens, exceeding its ${capacity.inputCapacity} token input capacity.`, { capacity, extensions: { capacityStage: "preflight", capacity } });
+            throw new ProviderError(this.#source, "capacity_exceeded", `The exact provider request uses ${capacity.prompt.tokens} input tokens, exceeding its ${capacity.inputWall} token input wall.`, { capacity, extensions: { capacityStage: "preflight", capacity } });
         }
-        // {§provider-flexed-allowance} (#482): the wire grants the flexed
-        // allowance — the floor, or the exactly-measured slack above it.
+        // {§provider-flexed-allowance}: the wire grants the flexed allowance —
+        // the window's remainder, between the floor and the model's limit.
         const effectiveMaxOutputTokens = capacity.responseMax ?? capacity.outputBudget ?? undefined;
         const nativeReasoningBudget = this.#requestBody.nativeReasoningBudget(capacity.outputBudget, capacity.reasoningBudget);
         // Both transports project the same resolved sampling; managed fields still win.
@@ -529,7 +536,7 @@ export default class AiSdkProvider {
             catch (cause) {
                 throw new ProviderRequestObserverError(cause);
             }
-            const settleAccounting = async (outcome, usage, evidence, status) => {
+            const settleAccounting = async (outcome, usage, evidence, forensicEvidence, status) => {
                 let requestAccounting;
                 let normalizationFailure;
                 try {
@@ -558,7 +565,7 @@ export default class AiSdkProvider {
                 }
                 accounting.push(requestAccounting);
                 try {
-                    await settle?.(requestAccounting);
+                    await settle?.(requestAccounting, forensicEvidence);
                 }
                 catch (cause) {
                     throw new ProviderRequestObserverError(cause);
@@ -579,10 +586,6 @@ export default class AiSdkProvider {
                         messages,
                         signal: operationSignal,
                         fetch: this.#fetch,
-                        fetchTimeoutMs: this.#fetchTimeoutMs,
-                        firstContentTimeoutMs: this.#firstContentTimeoutMs,
-                        streamIdleTimeoutMs: this.#streamIdleTimeoutMs,
-                        repeatedLineLimit: this.#repeatedLineLimit,
                         streaming: this.#streaming,
                         captureRawBody: this.#rawBody,
                         ...observers,
@@ -596,10 +599,6 @@ export default class AiSdkProvider {
                         systemProviderOptions: this.#systemCacheProviderOptions,
                         messages,
                         signal: operationSignal,
-                        fetchTimeoutMs: this.#fetchTimeoutMs,
-                        firstContentTimeoutMs: this.#firstContentTimeoutMs,
-                        streamIdleTimeoutMs: this.#streamIdleTimeoutMs,
-                        repeatedLineLimit: this.#repeatedLineLimit,
                         streaming: this.#streaming,
                         captureRawBody: this.#rawBody,
                         ...observers,
@@ -628,13 +627,13 @@ export default class AiSdkProvider {
                 if (transportFailureOutputObserved(error))
                     recoveredAfterOutput = true;
                 const failure = transportFailureEvidence(error);
-                await settleAccounting("error", failure.usage, failure.chargeEvidence, failure.status);
+                await settleAccounting("error", failure.usage, failure.chargeEvidence, failure, failure.status);
                 throw error;
             }
             if (!structuredReasoning && liveProjection !== undefined)
                 observeRequestReasoning(liveProjection.finish());
             successfulReasoningStream = requestReasoningStream;
-            await settleAccounting("response", response.usage, response.chargeEvidence);
+            await settleAccounting("response", response.usage, response.chargeEvidence, response);
             return response;
         };
         const executeRequest = async () => {
@@ -667,9 +666,9 @@ export default class AiSdkProvider {
                 || err instanceof ProviderReasoningObserverError)
                 throw err.cause;
             if (signal?.aborted)
-                throw err;
+                throw signal.reason;
             if (operationTimeout?.aborted) {
-                const timeout = new ProviderTimeoutError("operation", this.#operationTimeoutMs, err);
+                const timeout = new ProviderTimeoutError(this.#operationTimeoutMs, err);
                 throw new ProviderError(this.#source, "deadline_exceeded", timeout.message, {
                     status: 504,
                     cause: timeout,
@@ -828,9 +827,8 @@ export default class AiSdkProvider {
             ...(notices !== undefined ? { notices } : {}),
         };
         // {§provider-flexed-allowance} (#482): conformance judges the GRANT the
-        // wire actually sent, not the configured floor — output between the two
-        // is overflow tolerance working, never a provider fault. Run7 loop-death
-        // was this guard still holding the floor after the flex landed.
+        // wire actually sent, not the configured reservation — output between
+        // the two is overflow tolerance working, never a provider fault.
         const grantedOutput = capacity.responseMax ?? capacity.outputBudget;
         if (grantedOutput !== null
             && usage?.outputTokens !== undefined
@@ -850,16 +848,6 @@ export default class AiSdkProvider {
                 },
             });
         }
-        if (raw.finishReason === "repetition") {
-            // {§repetition-stop}: the stream was stopped at a repeated line; the partial attempt is evidence, and the
-            // consumer reads it as an invalid emission, never as a provider failure to recover.
-            const attempt = {
-                assistant: { ...assistant, finishReason: raw.finishReason },
-                ...evidence,
-            };
-            const { line, count } = raw.repetition;
-            throw new ProviderError(this.#source, "repetition", `The response repeated one line ${count} times and was stopped: \`${line}\``, { attempt, accounting, extensions: { stage: "provider-response", finishReason: "repetition", repeatedLine: line, repeatedCount: count } });
-        }
         if (raw.finishReason === "resource_interrupted") {
             const attempt = {
                 assistant: { ...assistant, finishReason: raw.finishReason },