modelrig 0.11.0 → 0.12.0

package/CHANGELOG.md

@@ -15,6 +15,88 @@ in `docs/ModelRig/development-rules.md`; enforced by `release-parity.test.ts`).
 
 ## [Unreleased]
 
+## [0.12.0] — 2026-10-25
+
+ARC lane: conversations as runs, delayed outcomes, run limits, the messages
+lane (chat turns and tool loops) with turn replay, generated-media units and
+unit pricing, OTLP-in v1 with framework recipes — plus run failure reasons and
+the ingest fixes. Ships with `modelrig-oracle` 0.4.0 (`record_outcome`,
+`query_conversations`, `query_registry` `kind`).
+
+**Upgrade notes.** Outcomes need control-plane migration **00089** on the
+server first (apply 00089 → deploy the server → upgrade the SDK). Direct-mode
+(service-role) sinks must apply control-plane migrations through **00062**
+before upgrading. `RawRunInput.userPrompt` is now optional (see below).
+
+**Added**
+
+- **Messages lane (ARC-β): chat turns and tool loops, unflattened.** `rig.run(route, { input, messages })` on a route that declares `input: { kind: messages }`, and `runRaw({ ..., messages })` (`userPrompt` optional alongside it), take a `HistoryMessage[]` conversation (user / assistant with `toolCalls` / tool) that every adapter maps to its native shape — Anthropic `tool_use`/`tool_result` blocks, OpenAI `tool_calls`/`role:"tool"` (chat, stream and Responses), Gemini `functionCall`/`functionResponse` contents, and the OpenAI-compatible hosts. The system prompt stays the route template / `systemPrompt`; a `system` message in the history, an orphan tool result or an empty history throws `MessagesInputError` pre-dispatch (`historyProblems` lists the same). Without `messages` every request is byte-identical (pinned by `one-turn-parity.test.ts`). The hosted `/v1/chat/completions` passes a multi-turn transcript through as `messages` on the raw lane and to a messages route on the slug lane; single-turn requests are unchanged. Capture stores the history as JSON (`renderForCapture`); the prefix fingerprint stays stable across a growing conversation (`canonicalizeHistory` is prefix-stable). A messages route cannot declare `mask:`/`screen:` yet, and managed serving fails closed on it. See `docs/messages-lane.md`.
+- **Messages lane — refusals and fallbacks.** A messages route that declares `repair:` (route or variant) is refused at load and at run time (the repair rung cannot see the conversation). Validation now requires the tool results right after an assistant turn with `toolCalls` to answer exactly that turn's calls. Provider rules fail closed before dispatch: Anthropic with tool turns in the history and no declared `tools`, and Gemini 3 models with function calls in the history (no placeholder thought signatures are invented). On `/v1/chat/completions` (raw lane) a tool transcript sent to `anthropic/*` without `tools`, or to `gemini/*`, keeps today's flattening, as does a transcript whose tool results do not pair. In turn replay, a case a candidate cannot take natively is reported as incomparable for that model (`VariantMetrics.incomparable`, "n cases incomparable for <model>: reason") and excluded from its score and regret, never scored as a failure.
+- **Messages lane — incompatible candidates are skipped, not fatal.** `rig.run` now skips a candidate that cannot take the history natively before the ladder (no dispatch, no row) and serves the next compatible one; skips are tagged `history_skipped` on the call's rows, and only an all-incompatible route fails closed (`invariant_violation` listing each reason). Capture records the call's `tools` / `toolChoice` (only when declared) and turn replay re-sends them, so Anthropic tool-loop turns replay as comparable. Tool-call ids are scoped to one turn (`call_0` reused across turns is valid). The Gemini 3 check normalizes `models/` and case.
+- **Turn replay attributes fallback answers.** When a multi-candidate arm's first candidate is skipped for a turn's history and a later one answers, each replay sample records `servedBy` / `historySkipped` (messages routes only), the arm gains `VariantMetrics.servedByFallback` and `armLabel` ("<first> → <fallback> (fallback served n/N)"), and the matrix, headline and reviewer arm meta use that label; the report prints "n cases served by <model> (<skipped model> skipped: <reason>)". Arms with no skip, and every variables route, are unchanged.
+- **Type change: `RawRunInput.userPrompt` is now optional** (`userPrompt?: string`) because `messages` can replace it. Code that reads `input.userPrompt` as a `string` (a custom `runRaw` dependency or wrapper) needs `?? ""` after upgrading.
+- **Turn replay (ARC-β).** A bake-off over a messages route replays each captured turn's history and the candidate answers that final turn only (graded by the route's grader, regret bound unchanged; never a simulated next user turn); eval cases carry the turn (`turnReplayCase`). Runs only where capture is explicitly on, otherwise `TurnReplayRefusedError` (HTTP 422).
+- **Runs carry why they failed.** `run.end("failed", { failureReason })` and `rig.artifacts.run.setFailureReason(reason)` (for `run.scope`) write an identifier reason to `runs.meta.failure_reason` (`RunMetaV1.failure_reason`, sanitized like every envelope field); the hosted `run-outcome@v1` verdict labels a failed run with it instead of a stray step `failure_class`. Additive; a succeeded run ignores the reason.
+- **Conversations as runs (ARC-1).** `rig.conversation(id, { agent, user?, tenant?, limits? })` opens (or re-enters) ONE run for a chat or agent task loop — `pipeline = agent`, `episode_key = id`, `RunMeta.trigger = "conversation"`; `conversation.turn(fn)` records every call inside it as step `turn:<n>.<k>`; `conversation.end()`. Re-entry continues the open run in-process or from the local store. The hosted server closes idle conversations (`conversation-idle-close`, default 30 min) and offers them to auto-grading under `conversation:<agent>`. Console: **Conversations** (list + transcript). Starter guide: `templates/support-conversation.json`.
+- **Outcomes (ARC-3).** `rig.outcome({ key, name, score?, value?, scale?, at?, idempotencyKey? })` and `POST /v1/outcomes` record a delayed business outcome (resolved, booked, CTR) as a `kind: "outcome"` grade (`grader = outcome:<name>`, raw `value`); a conversation key joins lazily, so it may arrive before its run. Outcome agreement on a grader's page is a held-out receipt, never a calibration gate. Needs control-plane migration **00089** — deploy order: apply 00089 → deploy the server → upgrade the SDK. Until the server sees it (lazy re-probe, ~1 min, no restart), `POST /v1/outcomes` answers 503 and outcome rows on `/v1/ingest` are dropped and counted (`outcome_schema_unavailable`) without holding back the rest of the batch. `idempotencyKey` ids are UUIDv5(org, `name:key`), identical on both doors; `rig.grade({ kind: "outcome" })` is refused (use `rig.outcome`). With an idempotency key, the id `rig.outcome` returns is local; the server stores the row under the org-scoped id. Workspace note: a local telemetry SQLite file created by an unpublished pre-fix workspace build of this branch (its grades table at `user_version` 17 without the `idempotency_key` column / transactional rebuild) must be deleted before using this build; published-SDK databases (≤ v16) migrate in place. MCP: `record_outcome`, `query_conversations`.
+- **Run limits (ARC-5).** `run.start({ limits: { maxUsd, maxCalls, maxWallMs, maxRepeatToolCall } })` (also on `rig.conversation`) refuses the next `rig.run` / `rig.runRaw` in-process once a limit is reached, throwing `RunLimitExceeded` (new terminal failure class `limit_exceeded`) and ending the run failed with `failure_reason: limit:<which>`; `rig.recordCall` only accumulates. Repeat-tool detection keeps a hash of the arguments only.
+- **Generated media: units and unit pricing (ARC-4).** `rig.recordCall({ ..., units: { images?, seconds?, characters?, steps? } })` records an image / video / text-to-speech / transcription call on `meta.units` (`AttemptMetaV1.units`, `UNIT_KEYS`; zero DDL). `estimateUnitCostUsd(provider, model, units)` prices one billed unit per model from the registry's declared generation entry (`pricing_v2.modality`, with provenance), else the vendored pricing map's per-mode rate, else `pricingMissing` — never from token rates; token and unit cost add. `ModalityRates` gains `video_per_second`, `characters_per_million` and `provenance`. Registry entries gain an optional `kind` (`llm` | `image_gen` | `video_gen` | `tts` | `stt`) and 12 declared-only generation models, unprobed and never ranked; the oracle's `query_registry` takes `kind`. Media artifacts (`image/*`, `video/*`, `audio/*`) are stored flagged unscreened (`scrub.unscreened = "media"`); `POST /v1/grade` accepts `artifactId`; the console shows approval rate per $ on Costs and a Media approval tab in the Grade queue. Starter guide: `templates/creative-approval.json`.
+- **OTLP-in v1 (ARC-6).** Attested steps keep `parent_span_id` and `operation_class`; a trace with `gen_ai.conversation.id` and `invoke_agent` spans becomes a conversation run (`trigger: conversation`) with generations numbered `turn:<n>.<k>` and tool calls counted per turn. Content extraction exists behind the server gate `MODELRIG_OTLP_CONTENT_EXTRACTION` (OFF; also needs a managed posture) pending ruling M14. Framework recipes: Vercel AI SDK, OpenAI Agents SDK, LangGraph / LangChain JS, Mastra, LiveKit Agents, single-sourced through the new `OTLP_EXPORTER` and `OTLP_TURN` setup blocks.
+
+**Fixed**
+
+- **ingest: inference rows are written with a uniform key set — a web-search surcharge row (`meta: null`) no longer makes the whole batch 422 (PGRST102).** `stampBatch` gives every `inferences` row the same columns (absent optional fields written as their default: `null`, or `0`/`false`/`{}` for the NOT NULL ones); `steps` / `artifacts` fill an optional column with `null` when any row in the write carries it; `tasks` / `runs` (merge upserts) are split into one write per key set so a `null` never overwrites a stored value. Before this, one such row poisoned the whole bulk insert and the exporter dropped the batch after its self-heal window. Applies to both transports (`stampBatch` is shared; the hosted server picks it up on deploy).
+- **direct mode: apply control-plane migrations through 00062 before upgrading (migrate first).** Because every `inferences` row now carries the full column set, a service-role (direct-mode) sink writes `call_id`, `attempt_seq`, `candidate_index` (00054), `meta` (00056) and `mask` (00062) on every batch; a database missing those columns rejects every batch (`PGRST204`). The hosted server is already current.
+- **Rejected writes log the PostgREST code.** The server logs the table, status and error code (`PGRST102`, `23514`, …) of a rejected control-plane write, with no host, header or row content, and the `/v1/ingest` 422 body adds `reason_code` (the code alone). The SDK carries it as `SinkSendError.reasonCode` and names it on the exporter's `POISON-SKIP` line.
+
+## [0.11.1] — 2026-10-21
+
+Setup correctness (SURF-1, the 2026-10-20 setup audit). Patch: no new public
+API, no DDL, no new environment variable.
+
+**Fixed**
+
+- **The `modelrig observe` seam no longer crashes your app on import.** The
+  generated `modelrig/observe/record-step.ts` builds its rig with
+  `createRig({ ...loadConfigFromEnv(), routesDir: null })`, so importing it in a
+  repo with no `./modelrig/routes` cannot throw `RouteConfigError`.
+- **`createRig` with an absent *default* routes directory warns and builds a
+  raw-only rig** (`rig.runRaw`, `rig.runRawStream`, `rig.recordCall` work;
+  `rig.run` refuses with an `invariant_violation` that names the missing
+  directory). "Default" is tracked by where the value came from:
+  `loadConfigFromEnv()` filled it in (no `MODELRIG_ROUTES_DIR`, no `routesDir`
+  override) and it was left as loaded. A routes directory you set (env, a
+  `loadConfigFromEnv({ routesDir })` override, or a hand-built config, even
+  `"./modelrig/routes"`) that is missing still throws, as before. Pass
+  `routesDir: null` to mean raw-only and silence the warning.
+- **The observe wiring reports cost.** `recordStep(stepKey, { …, call })`
+  reports the provider call through `rig.recordCall`, so observed steps carry
+  tokens and cost and appear on /inferences; the seam exports `closeRig()`.
+  Gemini wiring counts thinking tokens (`thoughtsTokenCount`) as output.
+  OBSERVE.md's run wrap passes `environment`, and the dead
+  `MODELRIG_ARTIFACT_STEPS` line is gone (the SDK never read it).
+- **`modelrig init` writes the starter route's two files**
+  (`schemas/support_summarize.schema.json`, `prompts/support_summarize.system.md`
+  under the routes dir, never overwriting yours), so the
+  `example.support_summarize` route every on-ramp shows validates as pasted.
+- **`bakeoff --from-eval-cases` runs with no routes directory** (an empty or
+  unrehydratable eval set reports honestly; a case that replays still needs the
+  task's route bundle for its variants), and is listed in `--help`.
+- **`modelrig doctor` prints the org slug**: `→ org «Your Org» (your-org) ✓`.
+  The slug is what `--expect-org` and `createRig({ expectOrg })` compare.
+
+**Docs** (modelrig.dev, shipped in the tarball where listed in `files`)
+
+- Shared setup blocks for the raw rig, `rig.recordCall`, run grades, bake-offs
+  and `--expect-org`, one text on every surface. The TypeScript shared blocks
+  and the generated observe seam are compiled strict +
+  `exactOptionalPropertyTypes` + `noUncheckedIndexedAccess` against the built
+  types in CI; other page snippets are not.
+- A run grade's subject id is `run.runId` (the run's id; the grade shows on
+  the run page and on /projects); the old `run_id` examples did not compile. The A/B `arm` example spreads `arm` only when drawn.
+- The modelrig.dev header is a five-link top bar plus a grouped sidebar
+  (collapsible on mobile) instead of ~40 flat links.
+
 ## [0.11.0] — 2026-10-20
 
 **`rig.recordCall()` — record a provider call ModelRig did not dispatch**

package/dist/adapters/anthropic.d.ts

@@ -12,6 +12,7 @@
 import type { CapabilityFlag, ProviderAdapter } from "../types/provider";
 import type { ClassifiedFailure, ShapingInput } from "./gemini";
 import type { AnthropicWireToolChoice } from "./tool-mapping";
+import type { AnthropicWireMessage } from "./history-mapping";
 export interface AnthropicAdapterConfig {
     readonly apiKey: string;
 }
@@ -47,10 +48,9 @@ export interface AnthropicShapedParams {
     /** Plain string by default; the block-array form carries a `cache_control`
      * breakpoint when the request opts into caching (WS-C). */
     readonly system?: string | readonly AnthropicSystemBlock[];
-    readonly messages: ReadonlyArray<{
-        readonly role: "user";
-        readonly content: string;
-    }>;
+    /** Single-turn: one user message (byte-identical to pre-ARC-β). Messages lane:
+     * the mapped history (adapters/history-mapping.ts). */
+    readonly messages: readonly AnthropicWireMessage[];
     readonly stream: true;
     readonly tools?: readonly AnthropicTool[];
     readonly tool_choice?: {

package/dist/adapters/anthropic.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"anthropic.d.ts","sourceRoot":"","sources":["../../src/adapters/anthropic.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAEhB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAEhE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,gBAAgB,CAAC;AAK9D,MAAM,WAAW,sBAAsB;IACrC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AASD;yDACyD;AACzD,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAEhE;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAclF;AAED,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAwB3E;AAED,UAAU,aAAa;IACrB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,sDAAsD;IACtD,QAAQ,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC;CACxB;AAED;;mEAEmE;AACnE,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,aAAa,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;KAAE,CAAC;CACzD;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B;wFACoF;IACpF,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;+DAC2D;IAC3D,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,oBAAoB,EAAE,CAAC;IAC3D,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAC;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtF,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,aAAa,EAAE,CAAC;IAC1C,QAAQ,CAAC,WAAW,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,uBAAuB,CAAC;CACnG;AAED,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,EACb,GAAG,EAAE,YAAY,EACjB,MAAM,UAAQ,GACb,qBAAqB,CAuDvB;AAED;iEACiE;AACjE,MAAM,WAAW,sBAAsB;IACrC,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IAC1C,QAAQ,CAAC,2BAA2B,CAAC,EAAE,MAAM,CAAC;IAC9C,QAAQ,CAAC,cAAc,CAAC,EAAE;QACxB,QAAQ,CAAC,yBAAyB,CAAC,EAAE,MAAM,CAAC;QAC5C,QAAQ,CAAC,yBAAyB,CAAC,EAAE,MAAM,CAAC;KAC7C,CAAC;CACH;AAED,0EAA0E;AAC1E,MAAM,WAAW,6BAA6B;IAC5C,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;CACrC;AAED;;;;;;GAMG;AACH,wBAAgB,4BAA4B,CAC1C,CAAC,EAAE,sBAAsB,GACxB,6BAA6B,CAa/B;AAED,wBAAgB,sBAAsB,CAAC,MAAM,EAAE,sBAAsB,GAAG,eAAe,CAsOtF"}
+{"version":3,"file":"anthropic.d.ts","sourceRoot":"","sources":["../../src/adapters/anthropic.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAEhB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAEhE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,gBAAgB,CAAC;AAG9D,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAK9D,MAAM,WAAW,sBAAsB;IACrC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AASD;yDACyD;AACzD,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAEhE;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAclF;AAED,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAwB3E;AAED,UAAU,aAAa;IACrB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,sDAAsD;IACtD,QAAQ,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC;CACxB;AAED;;mEAEmE;AACnE,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,aAAa,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;KAAE,CAAC;CACzD;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B;wFACoF;IACpF,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;+DAC2D;IAC3D,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,oBAAoB,EAAE,CAAC;IAC3D;2DACuD;IACvD,QAAQ,CAAC,QAAQ,EAAE,SAAS,oBAAoB,EAAE,CAAC;IACnD,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,aAAa,EAAE,CAAC;IAC1C,QAAQ,CAAC,WAAW,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;KAAE,GAAG,uBAAuB,CAAC;CACnG;AAED,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,EACb,GAAG,EAAE,YAAY,EACjB,MAAM,UAAQ,GACb,qBAAqB,CAwDvB;AAED;iEACiE;AACjE,MAAM,WAAW,sBAAsB;IACrC,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IAC1C,QAAQ,CAAC,2BAA2B,CAAC,EAAE,MAAM,CAAC;IAC9C,QAAQ,CAAC,cAAc,CAAC,EAAE;QACxB,QAAQ,CAAC,yBAAyB,CAAC,EAAE,MAAM,CAAC;QAC5C,QAAQ,CAAC,yBAAyB,CAAC,EAAE,MAAM,CAAC;KAC7C,CAAC;CACH;AAED,0EAA0E;AAC1E,MAAM,WAAW,6BAA6B;IAC5C,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;CACrC;AAED;;;;;;GAMG;AACH,wBAAgB,4BAA4B,CAC1C,CAAC,EAAE,sBAAsB,GACxB,6BAA6B,CAa/B;AAED,wBAAgB,sBAAsB,CAAC,MAAM,EAAE,sBAAsB,GAAG,eAAe,CAyOtF"}

package/dist/adapters/anthropic.js

@@ -22,6 +22,8 @@ exports.normalizeAnthropicInputUsage = normalizeAnthropicInputUsage;
 exports.createAnthropicAdapter = createAnthropicAdapter;
 const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
 const tool_mapping_1 = require("./tool-mapping");
+const history_mapping_1 = require("./history-mapping");
+const messages_1 = require("../run/messages");
 const attempt_meta_map_1 = require("./attempt-meta-map");
 const sampling_1 = require("./sampling");
 const DEFAULT_MAX_TOKENS = 32000;
@@ -127,7 +129,8 @@ function buildAnthropicParams(model, req, strict = false) {
         max_tokens: (0, sampling_1.effectiveMaxOutputTokens)(req, DEFAULT_MAX_TOKENS),
         ...(0, sampling_1.openAISamplingFields)(req),
         ...systemField,
-        messages: [{ role: "user", content: req.userPrompt }],
+        // ARC-β: history present ⇒ mapped conversation; absent ⇒ today's single user turn.
+        messages: req.history !== undefined ? (0, history_mapping_1.toAnthropicMessages)(req.history, req.userPrompt) : [{ role: "user", content: req.userPrompt }],
         stream: true,
         ...toolBlock,
     };
@@ -175,6 +178,10 @@ function createAnthropicAdapter(config) {
                 ...(usage !== undefined ? { usage } : {}),
                 ...(attemptMeta !== undefined ? { meta: attemptMeta } : {}),
             });
+            // ARC-β ruling M3(b): a history Anthropic would reject fails closed BEFORE dispatch.
+            const incompatible = (0, messages_1.historyIncompatibility)("anthropic", model, req.history, req.tools);
+            if (incompatible !== null)
+                return fail("invariant_violation", incompatible, "declare the tool loop's `tools` on this request");
             const wantJson = req.outputSchema !== null;
             try {
                 const client = new sdk_1.default({ apiKey: config.apiKey, timeout: req.timeoutMs });

package/dist/adapters/deepseek.d.ts

@@ -15,15 +15,13 @@
 import type { CapabilityFlag, ProviderAdapter } from "../types/provider";
 import type { ClassifiedFailure, ShapingInput } from "./gemini";
 import type { OpenAIWireTool, OpenAIWireToolChoice } from "./tool-mapping";
+import type { OpenAIChatWireMessage } from "./history-mapping";
 export interface DeepSeekAdapterConfig {
     readonly apiKey: string;
 }
 export declare function deepseekSupports(feature: CapabilityFlag, model: string): boolean;
 export declare function classifyDeepSeekFailure(message: string): ClassifiedFailure;
-interface ChatMessage {
-    readonly role: "system" | "user";
-    readonly content: string;
-}
+type ChatMessage = OpenAIChatWireMessage;
 export interface DeepSeekShapedParams {
     readonly model: string;
     readonly max_tokens: number;

package/dist/adapters/deepseek.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"deepseek.d.ts","sourceRoot":"","sources":["../../src/adapters/deepseek.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAChB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAMhE,OAAO,KAAK,EAAuB,cAAc,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAQhG,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAKD,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAMhF;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAiB1E;AAED,UAAU,WAAW;IACnB,QAAQ,CAAC,IAAI,EAAE,QAAQ,GAAG,MAAM,CAAC;IACjC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,CAAC;IAC1C,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,aAAa,EAAE,IAAI,CAAA;KAAE,CAAC;IAC1D,8EAA8E;IAC9E,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,eAAe,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;KAAE,CAAC;IAC5D,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC3C,QAAQ,CAAC,WAAW,CAAC,EAAE,oBAAoB,CAAC;CAC7C;AAOD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,oBAAoB,CAkC1F;AAED,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,qBAAqB,GAAG,eAAe,CAoLpF"}
+{"version":3,"file":"deepseek.d.ts","sourceRoot":"","sources":["../../src/adapters/deepseek.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAChB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAMhE,OAAO,KAAK,EAAuB,cAAc,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAQhG,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,mBAAmB,CAAC;AAE/D,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAKD,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAMhF;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAiB1E;AAED,KAAK,WAAW,GAAG,qBAAqB,CAAC;AAEzC,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,CAAC;IAC1C,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,aAAa,EAAE,IAAI,CAAA;KAAE,CAAC;IAC1D,8EAA8E;IAC9E,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,eAAe,CAAC,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;KAAE,CAAC;IAC5D,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC3C,QAAQ,CAAC,WAAW,CAAC,EAAE,oBAAoB,CAAC;CAC7C;AAOD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,oBAAoB,CAqC1F;AAED,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,qBAAqB,GAAG,eAAe,CAoLpF"}

package/dist/adapters/deepseek.js

@@ -26,6 +26,7 @@ const tool_mapping_1 = require("./tool-mapping");
 const openai_stream_1 = require("./openai-stream");
 const attempt_meta_map_1 = require("./attempt-meta-map");
 const sampling_1 = require("./sampling");
+const history_mapping_1 = require("./history-mapping");
 const DEEPSEEK_BASE_URL = "https://api.deepseek.com";
 const DEFAULT_MAX_TOKENS = 64000; // R1/V3 max — full capacity for long reports (ported)
 function deepseekSupports(feature, model) {
@@ -71,13 +72,16 @@ function buildDeepSeekParams(model, req) {
     // (spec R4), not silently intercepted here.
     const jsonObject = wantJson || req.responseFormat === "json_object";
     let systemPrompt = req.systemPrompt;
-    if (wantJson && !req.systemPrompt.includes("JSON") && !req.userPrompt.includes("JSON")) {
+    if (wantJson && !req.systemPrompt.includes("JSON") && !(0, history_mapping_1.userSideMentions)(req, "JSON")) {
         systemPrompt = systemPrompt === "" ? JSON_GUARD : `${systemPrompt}\n\n${JSON_GUARD}`;
     }
-    const messages = [
-        ...(systemPrompt !== "" ? [{ role: "system", content: systemPrompt }] : []),
-        { role: "user", content: req.userPrompt },
-    ];
+    // ARC-β: history present ⇒ mapped conversation; absent ⇒ today's literal (byte-identical).
+    const messages = req.history !== undefined
+        ? (0, history_mapping_1.toOpenAIChatMessages)(systemPrompt, req.history, req.userPrompt, { flattenParts: true })
+        : [
+            ...(systemPrompt !== "" ? [{ role: "system", content: systemPrompt }] : []),
+            { role: "user", content: req.userPrompt },
+        ];
     return {
         model,
         max_tokens: (0, sampling_1.effectiveMaxOutputTokens)(req, DEFAULT_MAX_TOKENS),

package/dist/adapters/gemini.d.ts

@@ -13,6 +13,7 @@
  * Copyright (c) 2023 Berri AI, MIT License — retain this notice if LiteLLM's
  * gemini transformation logic is consulted during the port.
  */
+import type { GeminiWireContent } from "./history-mapping";
 import type { AdapterRequest, CapabilityFlag, ProviderAdapter, ServiceTier } from "../types/provider";
 import type { FailureClass } from "../types/failures";
 import type { PromptDelivery, ServedTierSource } from "../telemetry/attempt-meta";
@@ -52,7 +53,9 @@ export type ShapingInput = Omit<AdapterRequest, "candidate">;
 export declare function geminiReasoningUnsupported(model: string, req: ShapingInput): ClassifiedFailure | null;
 export interface GeminiShapedRequest {
     readonly model: string;
-    readonly contents: string;
+    /** Single-turn: the user turn as a string (byte-identical to pre-ARC-β).
+     * Messages lane: the mapped `contents` array (adapters/history-mapping.ts). */
+    readonly contents: string | readonly GeminiWireContent[];
     readonly config: Record<string, unknown>;
     /** Issue #5 (R1): how the rendered prompt was delivered — stamped on the
      * attempt's `meta.prompt_delivery` so the control plane can prove it. */

package/dist/adapters/gemini.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"gemini.d.ts","sourceRoot":"","sources":["../../src/adapters/gemini.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAGH,OAAO,KAAK,EACV,cAAc,EAGd,cAAc,EACd,eAAe,EACf,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EAAE,YAAY,EAAc,MAAM,mBAAmB,CAAC;AAGlE,OAAO,KAAK,EAAiB,cAAc,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAKjG,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED;;;;mCAImC;AACnC,eAAO,MAAM,8BAA8B,EAAE,WAAW,CAAC,MAAM,CAI7D,CAAC;AA0BH,wBAAgB,cAAc,CAAC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAG9E;AAED,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,GAAG,EAAE,YAAY,CAAC;IAC3B,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED;;;;;GAKG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAyCxE;AAED;6EAC6E;AAC7E,MAAM,MAAM,YAAY,GAAG,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AAE7D;;;;;;;;GAQG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,iBAAiB,GAAG,IAAI,CAQrG;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACzC;6EACyE;IACzE,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC;CACzC;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAE/E;AAED,iEAAiE;AACjE,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,mBAAmB,CA4DxF;AAUD;;;;;gEAKgE;AAChE,wBAAgB,uBAAuB,CACrC,cAAc,EAAE,MAAM,GAAG,SAAS,EAClC,WAAW,EAAE,MAAM,GAAG,SAAS,EAC/B,SAAS,EAAE,WAAW,GACrB;IAAE,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,gBAAgB,CAAA;CAAE,CAOnE;AAED;;gEAEgE;AAChE,KAAK,mBAAmB,GAAG;IACzB,QAAQ,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC;IAChC,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC;CAC5B,CAAC;AACF,KAAK,oBAAoB,GAAG;IAAE,QAAQ,CAAC,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,mBAAmB,EAAE,CAAA;KAAE,CAAA;CAAE,CAAC;AAEvG;;;;;;;;;GASG;AACH,wBAAgB,0BAA0B,CAAC,SAAS,EAAE,oBAAoB,GAAG,SAAS,GAAG,MAAM,CAM9F;AAED;;qEAEqE;AACrE,wBAAgB,8BAA8B,CAAC,SAAS,EAAE,oBAAoB,GAAG,SAAS,GAAG,OAAO,CAInG;AAED;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAC/B,SAAS,EAAE,oBAAoB,GAAG,SAAS,EAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,QAAQ,EAAE,OAAO,GAChB,MAAM,CAGR;AAED,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,mBAAmB,GAAG,eAAe,CA8MhF"}
+{"version":3,"file":"gemini.d.ts","sourceRoot":"","sources":["../../src/adapters/gemini.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAGH,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAG3D,OAAO,KAAK,EACV,cAAc,EAGd,cAAc,EACd,eAAe,EACf,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EAAE,YAAY,EAAc,MAAM,mBAAmB,CAAC;AAGlE,OAAO,KAAK,EAAiB,cAAc,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAKjG,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED;;;;mCAImC;AACnC,eAAO,MAAM,8BAA8B,EAAE,WAAW,CAAC,MAAM,CAI7D,CAAC;AA0BH,wBAAgB,cAAc,CAAC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAG9E;AAED,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,GAAG,EAAE,YAAY,CAAC;IAC3B,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED;;;;;GAKG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAyCxE;AAED;6EAC6E;AAC7E,MAAM,MAAM,YAAY,GAAG,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AAE7D;;;;;;;;GAQG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,iBAAiB,GAAG,IAAI,CAQrG;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB;mFAC+E;IAC/E,QAAQ,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,iBAAiB,EAAE,CAAC;IACzD,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACzC;6EACyE;IACzE,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC;CACzC;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAE/E;AAED,iEAAiE;AACjE,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,mBAAmB,CAiExF;AAUD;;;;;gEAKgE;AAChE,wBAAgB,uBAAuB,CACrC,cAAc,EAAE,MAAM,GAAG,SAAS,EAClC,WAAW,EAAE,MAAM,GAAG,SAAS,EAC/B,SAAS,EAAE,WAAW,GACrB;IAAE,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,gBAAgB,CAAA;CAAE,CAOnE;AAED;;gEAEgE;AAChE,KAAK,mBAAmB,GAAG;IACzB,QAAQ,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC;IAChC,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC;CAC5B,CAAC;AACF,KAAK,oBAAoB,GAAG;IAAE,QAAQ,CAAC,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,mBAAmB,EAAE,CAAA;KAAE,CAAA;CAAE,CAAC;AAEvG;;;;;;;;;GASG;AACH,wBAAgB,0BAA0B,CAAC,SAAS,EAAE,oBAAoB,GAAG,SAAS,GAAG,MAAM,CAM9F;AAED;;qEAEqE;AACrE,wBAAgB,8BAA8B,CAAC,SAAS,EAAE,oBAAoB,GAAG,SAAS,GAAG,OAAO,CAInG;AAED;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAC/B,SAAS,EAAE,oBAAoB,GAAG,SAAS,EAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,QAAQ,EAAE,OAAO,GAChB,MAAM,CAGR;AAED,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,mBAAmB,GAAG,eAAe,CAqNhF"}

package/dist/adapters/gemini.js

@@ -26,6 +26,8 @@ exports.extractGeminiCandidateText = extractGeminiCandidateText;
 exports.geminiCandidateHasFunctionCall = geminiCandidateHasFunctionCall;
 exports.resolveGeminiText = resolveGeminiText;
 exports.createGeminiAdapter = createGeminiAdapter;
+const history_mapping_1 = require("./history-mapping");
+const messages_1 = require("../run/messages");
 const genai_1 = require("@google/genai");
 const tool_mapping_1 = require("./tool-mapping");
 const attempt_meta_map_1 = require("./attempt-meta-map");
@@ -180,7 +182,10 @@ function buildGeminiRequest(model, req) {
     let promptDelivery;
     if (usingCache) {
         config.cachedContent = req.cache?.key;
-        contents = cachedUserTurn(req.systemPrompt, req.userPrompt);
+        // ARC-β: with history, the system prompt folds into the first user turn (same rule).
+        contents = req.history !== undefined
+            ? (0, history_mapping_1.toGeminiContents)(req.history, req.userPrompt, req.systemPrompt)
+            : cachedUserTurn(req.systemPrompt, req.userPrompt);
         promptDelivery = "user+cache";
     }
     else {
@@ -203,6 +208,9 @@ function buildGeminiRequest(model, req) {
     if (req.requestedTier === "flex") {
         config.serviceTier = "flex";
     }
+    // ARC-β: history present (no cache handle) ⇒ mapped `contents`; system stays systemInstruction.
+    if (req.history !== undefined && !usingCache)
+        contents = (0, history_mapping_1.toGeminiContents)(req.history, req.userPrompt);
     return { model, contents, config, promptDelivery };
 }
 /** Best-effort served-tier extraction (ported): response.serviceTier, then
@@ -291,6 +299,13 @@ function createGeminiAdapter(config) {
                 clearTimeout(timeoutId);
                 return fail(reasoningUnsupported.cls, `reasoning control declared on the non-thinking Gemini model "${model}"`, reasoningUnsupported.fixHint);
             }
+            // ARC-β ruling M3(b): Gemini 3 + function calls in history fails closed BEFORE dispatch
+            // (no placeholder thought signatures are invented).
+            const incompatible = (0, messages_1.historyIncompatibility)("gemini", model, req.history, req.tools);
+            if (incompatible !== null) {
+                clearTimeout(timeoutId);
+                return fail("invariant_violation", incompatible, "serve tool-loop turns with a non-Gemini-3 model");
+            }
             try {
                 const shaped = buildGeminiRequest(model, req);
                 const dispatchedAt = Date.now();

package/dist/adapters/grok.d.ts

@@ -13,15 +13,13 @@
 import type { CapabilityFlag, ProviderAdapter } from "../types/provider";
 import type { ClassifiedFailure, ShapingInput } from "./gemini";
 import type { OpenAIWireTool, OpenAIWireToolChoice } from "./tool-mapping";
+import type { OpenAIChatWireMessage } from "./history-mapping";
 export interface GrokAdapterConfig {
     readonly apiKey: string;
 }
 export declare function grokSupports(feature: CapabilityFlag, _model: string): boolean;
 export declare function classifyGrokFailure(message: string): ClassifiedFailure;
-interface ChatMessage {
-    readonly role: "system" | "user";
-    readonly content: string;
-}
+type ChatMessage = OpenAIChatWireMessage;
 export interface GrokShapedParams {
     readonly model: string;
     readonly max_tokens: number;

package/dist/adapters/grok.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"grok.d.ts","sourceRoot":"","sources":["../../src/adapters/grok.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAChB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAMhE,OAAO,KAAK,EAAuB,cAAc,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAUhG,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAKD,wBAAgB,YAAY,CAAC,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAc7E;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAkDtE;AAED,UAAU,WAAW;IACnB,QAAQ,CAAC,IAAI,EAAE,QAAQ,GAAG,MAAM,CAAC;IACjC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,CAAC;IAC1C,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,aAAa,EAAE,IAAI,CAAA;KAAE,CAAC;IAC1D,8EAA8E;IAC9E,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;8EAC0E;IAC1E,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC;;;4DAGwD;IACxD,QAAQ,CAAC,gBAAgB,CAAC,EAAE,KAAK,GAAG,MAAM,CAAC;IAC3C,QAAQ,CAAC,eAAe,CAAC,EACrB;QAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;KAAE,GAChC;QACE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;QAC7B,QAAQ,CAAC,WAAW,EAAE;YAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;YAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;YAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAA;SAAE,CAAC;KACjG,CAAC;IACN,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC3C,QAAQ,CAAC,WAAW,CAAC,EAAE,oBAAoB,CAAC;CAC7C;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,gBAAgB,CAmDlF;AAED,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,iBAAiB,GAAG,eAAe,CAqN5E"}
+{"version":3,"file":"grok.d.ts","sourceRoot":"","sources":["../../src/adapters/grok.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAIV,cAAc,EACd,eAAe,EAChB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAMhE,OAAO,KAAK,EAAuB,cAAc,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAQhG,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,mBAAmB,CAAC;AAI/D,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAKD,wBAAgB,YAAY,CAAC,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAc7E;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,MAAM,GAAG,iBAAiB,CAkDtE;AAED,KAAK,WAAW,GAAG,qBAAqB,CAAC;AAEzC,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,EAAE,SAAS,WAAW,EAAE,CAAC;IAC1C,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;IACtB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,aAAa,EAAE,IAAI,CAAA;KAAE,CAAC;IAC1D,8EAA8E;IAC9E,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;8EAC0E;IAC1E,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC;;;4DAGwD;IACxD,QAAQ,CAAC,gBAAgB,CAAC,EAAE,KAAK,GAAG,MAAM,CAAC;IAC3C,QAAQ,CAAC,eAAe,CAAC,EACrB;QAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;KAAE,GAChC;QACE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;QAC7B,QAAQ,CAAC,WAAW,EAAE;YAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;YAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;YAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAA;SAAE,CAAC;KACjG,CAAC;IACN,QAAQ,CAAC,KAAK,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC3C,QAAQ,CAAC,WAAW,CAAC,EAAE,oBAAoB,CAAC;CAC7C;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,YAAY,GAAG,gBAAgB,CAsDlF;AAED,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,iBAAiB,GAAG,eAAe,CAqN5E"}

package/dist/adapters/grok.js

@@ -24,6 +24,7 @@ const tool_mapping_1 = require("./tool-mapping");
 const openai_stream_1 = require("./openai-stream");
 const attempt_meta_map_1 = require("./attempt-meta-map");
 const sampling_1 = require("./sampling");
+const history_mapping_1 = require("./history-mapping");
 const raw_support_1 = require("../run/raw-support");
 const responses_api_1 = require("./responses-api");
 const GROK_BASE_URL = "https://api.x.ai/v1";
@@ -96,10 +97,13 @@ function classifyGrokFailure(message) {
     return { cls: "network" };
 }
 function buildGrokParams(model, req) {
-    const messages = [
-        ...(req.systemPrompt !== "" ? [{ role: "system", content: req.systemPrompt }] : []),
-        { role: "user", content: req.userPrompt },
-    ];
+    // ARC-β: history present ⇒ mapped conversation; absent ⇒ today's literal (byte-identical).
+    const messages = req.history !== undefined
+        ? (0, history_mapping_1.toOpenAIChatMessages)(req.systemPrompt, req.history, req.userPrompt, { flattenParts: false })
+        : [
+            ...(req.systemPrompt !== "" ? [{ role: "system", content: req.systemPrompt }] : []),
+            { role: "user", content: req.userPrompt },
+        ];
     return {
         model,
         max_tokens: (0, sampling_1.effectiveMaxOutputTokens)(req, DEFAULT_MAX_TOKENS),

package/dist/adapters/history-mapping.d.ts

@@ -0,0 +1,126 @@
+/**
+ * Messages lane — provider-neutral history → each provider family's native
+ * conversation shape (ARC-β, builder-prompt-02 §1.2). The single place every
+ * dialect is translated, so each adapter's diff is one branch:
+ * `req.history !== undefined ? <family mapper> : <today's literal>`.
+ *
+ * SEMANTICS (identical for every family, `AdapterRequest.history`):
+ *   provider messages = [system(systemPrompt)] + map(history)
+ *                       + (userPrompt !== "" ? [user(userPrompt)] : [])
+ * The system prompt keeps its family-native slot (OpenAI `system` message,
+ * Anthropic top-level `system`, Gemini `systemInstruction`, Responses `system`
+ * input item) exactly as the single-turn path delivers it.
+ *
+ * Pure functions, no I/O. History is assumed VALID (run/messages.ts validates
+ * at every SDK entry): tool results answer a prior call, arguments are JSON.
+ */
+import type { HistoryMessage } from "../run/messages";
+/** True when `word` appears in the user side of a request — the single-turn
+ * `userPrompt`, or (messages lane) any history text. The DeepSeek / compat
+ * JSON-guard keys on this so a one-turn history is guarded exactly like the
+ * legacy `userPrompt` it replaces. */
+export declare function userSideMentions(req: {
+    readonly userPrompt: string;
+    readonly history?: readonly HistoryMessage[];
+}, word: string): boolean;
+export type OpenAIChatTextPart = {
+    readonly type: "text";
+    readonly text: string;
+};
+export type OpenAIChatWireMessage = {
+    readonly role: "system";
+    readonly content: string;
+} | {
+    readonly role: "user";
+    readonly content: string | readonly OpenAIChatTextPart[];
+} | {
+    readonly role: "assistant";
+    readonly content: string | readonly OpenAIChatTextPart[] | null;
+    readonly tool_calls?: ReadonlyArray<{
+        readonly id: string;
+        readonly type: "function";
+        readonly function: {
+            readonly name: string;
+            readonly arguments: string;
+        };
+    }>;
+} | {
+    readonly role: "tool";
+    readonly tool_call_id: string;
+    readonly content: string;
+};
+export interface OpenAIChatMappingOptions {
+    /** Collapse multi-part text content to one string — for hosts whose
+     * OpenAI-compatibility does not accept content-part arrays (DeepSeek and the
+     * generic compat hosts). Native OpenAI and xAI keep the part array. */
+    readonly flattenParts: boolean;
+}
+/** History → OpenAI Chat Completions `messages` (tool_calls / role:"tool"). */
+export declare function toOpenAIChatMessages(systemPrompt: string, history: readonly HistoryMessage[], userPrompt: string, opts: OpenAIChatMappingOptions): OpenAIChatWireMessage[];
+export type ResponsesInputItem = {
+    readonly role: "system" | "user" | "assistant";
+    readonly content: string;
+} | {
+    readonly role: "user";
+    readonly content: ReadonlyArray<{
+        readonly type: "input_text";
+        readonly text: string;
+    }>;
+} | {
+    readonly type: "function_call";
+    readonly call_id: string;
+    readonly name: string;
+    readonly arguments: string;
+} | {
+    readonly type: "function_call_output";
+    readonly call_id: string;
+    readonly output: string;
+};
+/** History → Responses API `input` (messages + function_call /
+ * function_call_output items). */
+export declare function toResponsesInput(systemPrompt: string, history: readonly HistoryMessage[], userPrompt: string): ResponsesInputItem[];
+export type AnthropicWireBlock = {
+    readonly type: "text";
+    readonly text: string;
+} | {
+    readonly type: "tool_use";
+    readonly id: string;
+    readonly name: string;
+    readonly input: Record<string, unknown>;
+} | {
+    readonly type: "tool_result";
+    readonly tool_use_id: string;
+    readonly content: string;
+};
+export interface AnthropicWireMessage {
+    readonly role: "user" | "assistant";
+    readonly content: string | readonly AnthropicWireBlock[];
+}
+/** History → Anthropic Messages `messages` (the system prompt stays the
+ * top-level `system`, built by the adapter exactly as today). */
+export declare function toAnthropicMessages(history: readonly HistoryMessage[], userPrompt: string): AnthropicWireMessage[];
+export type GeminiWirePart = {
+    readonly text: string;
+} | {
+    readonly functionCall: {
+        readonly name: string;
+        readonly args: Record<string, unknown>;
+    };
+} | {
+    readonly functionResponse: {
+        readonly name: string;
+        readonly response: Record<string, unknown>;
+    };
+};
+export interface GeminiWireContent {
+    readonly role: "user" | "model";
+    readonly parts: readonly GeminiWirePart[];
+}
+/**
+ * History → Gemini `contents`. `cachedPrefix` (non-empty) is the system prompt
+ * when a `cachedContent` handle forbids `systemInstruction` (issue #5 R1): it is
+ * delivered in the user turn exactly like the single-turn `cachedUserTurn` —
+ * folded into the first user message's first text part as `prefix\n\ntext`.
+ */
+export declare function toGeminiContents(history: readonly HistoryMessage[], userPrompt: string, cachedPrefix?: string): GeminiWireContent[];
+//# sourceMappingURL=history-mapping.d.ts.map

package/dist/adapters/history-mapping.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"history-mapping.d.ts","sourceRoot":"","sources":["../../src/adapters/history-mapping.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,KAAK,EAAkB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAKtE;;;sCAGsC;AACtC,wBAAgB,gBAAgB,CAC9B,GAAG,EAAE;IAAE,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,SAAS,cAAc,EAAE,CAAA;CAAE,EAClF,IAAI,EAAE,MAAM,GACX,OAAO,CAIT;AAgBD,MAAM,MAAM,kBAAkB,GAAG;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAElF,MAAM,MAAM,qBAAqB,GAC7B;IAAE,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,kBAAkB,EAAE,CAAA;CAAE,GACnF;IACE,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;IAC3B,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,kBAAkB,EAAE,GAAG,IAAI,CAAC;IAChE,QAAQ,CAAC,UAAU,CAAC,EAAE,aAAa,CAAC;QAClC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;QACpB,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;QAC1B,QAAQ,CAAC,QAAQ,EAAE;YAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;YAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;SAAE,CAAC;KAC1E,CAAC,CAAC;CACJ,GACD;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvF,MAAM,WAAW,wBAAwB;IACvC;;2EAEuE;IACvE,QAAQ,CAAC,YAAY,EAAE,OAAO,CAAC;CAChC;AAOD,+EAA+E;AAC/E,wBAAgB,oBAAoB,CAClC,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,SAAS,cAAc,EAAE,EAClC,UAAU,EAAE,MAAM,EAClB,IAAI,EAAE,wBAAwB,GAC7B,qBAAqB,EAAE,CA4BzB;AAID,MAAM,MAAM,kBAAkB,GAC1B;IAAE,QAAQ,CAAC,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC5E;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,aAAa,CAAC;QAAE,QAAQ,CAAC,IAAI,EAAE,YAAY,CAAC;QAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CAAE,GAClH;IAAE,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GAC/G;IAAE,QAAQ,CAAC,IAAI,EAAE,sBAAsB,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAEjG;kCACkC;AAClC,wBAAgB,gBAAgB,CAC9B,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,SAAS,cAAc,EAAE,EAClC,UAAU,EAAE,MAAM,GACjB,kBAAkB,EAAE,CAqBtB;AAID,MAAM,MAAM,kBAAkB,GAC1B;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,GAClH;IAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;IAAC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAE7F,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC;IACpC,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,kBAAkB,EAAE,CAAC;CAC1D;AAiBD;iEACiE;AACjE,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,SAAS,cAAc,EAAE,EAAE,UAAU,EAAE,MAAM,GAAG,oBAAoB,EAAE,CA+BlH;AAID,MAAM,MAAM,cAAc,GACtB;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACzB;IAAE,QAAQ,CAAC,YAAY,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAA;CAAE,GAC5F;IAAE,QAAQ,CAAC,gBAAgB,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAA;CAAE,CAAC;AAEzG,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC;IAChC,QAAQ,CAAC,KAAK,EAAE,SAAS,cAAc,EAAE,CAAC;CAC3C;AA0BD;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAC9B,OAAO,EAAE,SAAS,cAAc,EAAE,EAClC,UAAU,EAAE,MAAM,EAClB,YAAY,SAAK,GAChB,iBAAiB,EAAE,CAmCrB"}