@stll/ai-catalog 0.7.1 → 0.8.1

package/dist/{src-CyBTM4lw.js → src-CC7RZ0HV.js}

@@ -1,3 +1,4 @@
+import { MODEL_DEFAULTS_LEDGER, MODEL_TASK_OVERRIDE_GROUP, resolveDefaultModel } from "./model-defaults-ledger.js";
 import * as v from "valibot";
 //#region src/capabilities.gen.ts
 /**
@@ -733,7 +734,7 @@ const MODEL_OUTPUT_TOKEN_LIMITS = {
 	"openai/gpt-5.6-sol": 128e3,
 	"openai/gpt-5.6-terra": 128e3,
 	"openai/gpt-5.6-luna": 128e3,
-	"google/gemini-4-argon": 65536,
+	"google/gemini-4-argon": 943718,
 	"google/gemini-3.8-flash": 65536,
 	"google/gemini-3.7-flash": 65536,
 	"google/gemini-3.6-flash": 65536,
@@ -2061,21 +2062,6 @@ const DOCUMENT_INPUT_OVERRIDES = {
 * ID here against live provider/aggregator listings so a retired model
 * fails CI instead of 400-ing a user at runtime.
 */
-/**
-* Logical model roles. Call sites declare *what* they need, not
-* *which* model to use.
-*
-* - fast: classification, extraction, short generation
-* - chat: conversational with tool use and streaming
-* - reasoning: complex multi-step legal analysis
-* - pdf: native PDF/image understanding
-*/
-const MODEL_ROLES = [
-	"fast",
-	"chat",
-	"reasoning",
-	"pdf"
-];
 const TEMPERATURE_POLICIES = ["emit", "omit"];
 const IMAGE_INPUT_CAPABILITIES = [
 	"supported",
@@ -2173,176 +2159,48 @@ const GPT_56_MODEL_IDS = {
 };
 const GPT_56_OPENAI_MODEL_IDS = Object.values(GPT_56_MODEL_IDS).map(({ openai }) => openai);
 const GPT_56_OPENROUTER_MODEL_IDS = Object.values(GPT_56_MODEL_IDS).map(({ openrouter }) => openrouter);
+const defaultPresentation = (entry) => ({
+	kind: "default",
+	modelId: entry.candidate.modelId,
+	rationaleKey: entry.rationale.rationaleKey
+});
+/** Default labels in the override editor derive from the task ledger. */
 const BYOK_DEFAULT_MODELS = {
 	google: {
-		fast: {
-			kind: "default",
-			modelId: "gemini-3.8-flash",
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: "gemini-3.8-flash",
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: "gemini-3.8-flash",
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: {
-			kind: "default",
-			modelId: "gemini-3.8-flash",
-			rationaleKey: "organization.aiConfig.defaultRationale.pdf"
-		}
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].google),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.google),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.google),
+		pdf: defaultPresentation(MODEL_DEFAULTS_LEDGER.pdf.google)
 	},
 	openrouter: {
-		fast: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.luna.openrouter,
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openrouter,
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openrouter,
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openrouter,
-			rationaleKey: "organization.aiConfig.defaultRationale.pdf"
-		}
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].openrouter),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.openrouter),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.openrouter),
+		pdf: defaultPresentation(MODEL_DEFAULTS_LEDGER.pdf.openrouter)
 	},
 	openai: {
-		fast: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.luna.openai,
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openai,
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openai,
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: {
-			kind: "default",
-			modelId: GPT_6_MODEL_IDS.sol.openai,
-			rationaleKey: "organization.aiConfig.defaultRationale.pdf"
-		}
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].openai),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.openai),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.openai),
+		pdf: defaultPresentation(MODEL_DEFAULTS_LEDGER.pdf.openai)
 	},
 	anthropic: {
-		fast: {
-			kind: "default",
-			modelId: "claude-opus-5-5",
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: "claude-opus-5-5",
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: "claude-opus-5-5",
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: {
-			kind: "default",
-			modelId: "claude-opus-5-5",
-			rationaleKey: "organization.aiConfig.defaultRationale.pdf"
-		}
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].anthropic),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.anthropic),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.anthropic),
+		pdf: defaultPresentation(MODEL_DEFAULTS_LEDGER.pdf.anthropic)
 	},
 	bedrock: {
-		fast: {
-			kind: "default",
-			modelId: "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: {
-			kind: "default",
-			modelId: "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
-			rationaleKey: "organization.aiConfig.defaultRationale.pdf"
-		}
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].bedrock),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.bedrock),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.bedrock),
+		pdf: defaultPresentation(MODEL_DEFAULTS_LEDGER.pdf.bedrock)
 	},
 	mistral: {
-		fast: {
-			kind: "default",
-			modelId: "mistral-small-latest",
-			rationaleKey: "organization.aiConfig.defaultRationale.fast"
-		},
-		chat: {
-			kind: "default",
-			modelId: "mistral-medium-latest",
-			rationaleKey: "organization.aiConfig.defaultRationale.chat"
-		},
-		reasoning: {
-			kind: "default",
-			modelId: "mistral-medium-latest",
-			rationaleKey: "organization.aiConfig.defaultRationale.reasoning"
-		},
-		pdf: { kind: "unsupported" }
-	}
-};
-const defaultModelIds = (defaults) => ({
-	fast: defaults.fast.modelId,
-	chat: defaults.chat.modelId,
-	reasoning: defaults.reasoning.modelId,
-	pdf: defaults.pdf.modelId
-});
-/**
-* Instance-level default model IDs per provider. Extends the BYOK
-* defaults with the providers that only the instance path uses
-* (custom deployments and OpenAI-compatible endpoints).
-*/
-const DEFAULT_MODELS = {
-	google: defaultModelIds(BYOK_DEFAULT_MODELS.google),
-	openrouter: defaultModelIds(BYOK_DEFAULT_MODELS.openrouter),
-	openai: defaultModelIds(BYOK_DEFAULT_MODELS.openai),
-	anthropic: defaultModelIds(BYOK_DEFAULT_MODELS.anthropic),
-	bedrock: defaultModelIds(BYOK_DEFAULT_MODELS.bedrock),
-	mistral: {
-		fast: BYOK_DEFAULT_MODELS.mistral.fast.modelId,
-		chat: BYOK_DEFAULT_MODELS.mistral.chat.modelId,
-		reasoning: BYOK_DEFAULT_MODELS.mistral.reasoning.modelId,
-		pdf: null
-	},
-	azure_foundry: {
-		fast: GPT_56_MODEL_IDS.luna.openai,
-		chat: GPT_56_MODEL_IDS.terra.openai,
-		reasoning: GPT_56_MODEL_IDS.terra.openai,
-		pdf: GPT_56_MODEL_IDS.terra.openai
-	},
-	openai_compatible: {
-		fast: "default",
-		chat: "default",
-		reasoning: "default",
-		pdf: "default"
-	},
-	huggingface: {
-		fast: "speakleash/Bielik-1.5B-v3.0-Instruct",
-		chat: "speakleash/Bielik-11B-v2.3-Instruct",
-		reasoning: "speakleash/Bielik-11B-v2.3-Instruct",
-		pdf: "speakleash/Bielik-11B-v2.3-Instruct"
+		fast: defaultPresentation(MODEL_DEFAULTS_LEDGER["short-text"].mistral),
+		chat: defaultPresentation(MODEL_DEFAULTS_LEDGER.chat.mistral),
+		reasoning: defaultPresentation(MODEL_DEFAULTS_LEDGER.reasoning.mistral),
+		pdf: { kind: MODEL_DEFAULTS_LEDGER.pdf.mistral.kind }
 	}
 };
 /**
@@ -2801,34 +2659,18 @@ const isOfferedBYOKModelForRole = ({ provider, modelId, role }) => {
 		role
 	});
 };
-/**
-* Resolve a model id that will actually work for this provider+role,
-* keeping the SAME provider. Returns the caller's model unchanged when
-* it is still offered; otherwise falls back to the provider's per-role
-* default (`BYOK_DEFAULT_MODELS`). Returns `null` only when the
-* provider has no valid model for the role at all — the sole case
-* today is `mistral` + `pdf`, because the TanStack Mistral adapter
-* exposes no `document` input modality, so no Mistral model (not even
-* the default) can serve the PDF role.
-*
-* Used to auto-heal org AI configs whose pinned model was removed by a
-* catalog bump, so generation resolves to a supported model instead of
-* 400-ing (or forwarding a retired id to the provider).
-*/
-const resolveWorkingBYOKModelForRole = ({ provider, modelId, role }) => {
+/** Keep a supported override, otherwise use that provider's task default. */
+const resolveWorkingBYOKModelForTask = ({ provider, modelId, task }) => {
+	const role = MODEL_TASK_OVERRIDE_GROUP[task];
 	if (isOfferedBYOKModelForRole({
 		provider,
 		modelId,
 		role
 	})) return modelId;
-	const entry = BYOK_DEFAULT_MODELS[provider][role];
-	if (entry.kind === "unsupported") return null;
-	const fallback = entry.modelId;
-	return isOfferedBYOKModelForRole({
-		provider,
-		modelId: fallback,
-		role
-	}) ? fallback : null;
+	return resolveDefaultModel({
+		task,
+		orgProviders: [provider]
+	})?.modelId ?? null;
 };
 /** Native Anthropic adaptive-thinking models, derived from replay capabilities. */
 const ANTHROPIC_ADAPTIVE_THINKING_MODELS = BYOK_MODEL_OPTIONS.anthropic.filter((model) => MODEL_REASONING_CAPABILITIES[model].anthropicThinking === "adaptive");
@@ -2849,24 +2691,6 @@ const REASONING_EFFORTS = [
 ];
 const resolvedReasoningEffortSchema = v.pipe(v.picklist(REASONING_EFFORTS), v.brand("ResolvedReasoningEffort"));
 /**
-* Per-provider model served when a user's included budget is exhausted
-* and no explicit selection overrides it. Falls back to each
-* provider's fast-role default; providers without a first-class
-* adapter path cannot serve the lane. Total over providers so a new
-* provider lands with an explicit decision.
-*/
-const FALLBACK_CHAT_MODEL_BY_PROVIDER = {
-	google: DEFAULT_MODELS.google.fast,
-	openrouter: DEFAULT_MODELS.openrouter.fast,
-	openai: DEFAULT_MODELS.openai.fast,
-	anthropic: DEFAULT_MODELS.anthropic.fast,
-	bedrock: DEFAULT_MODELS.bedrock.fast,
-	mistral: DEFAULT_MODELS.mistral.fast,
-	azure_foundry: null,
-	openai_compatible: null,
-	huggingface: null
-};
-/**
 * Provider-native IDs that are exact aliases of an offered catalog ID.
 * Every metadata lookup normalizes here, so instance/dev overrides cannot
 * bypass rates or capabilities and duplicated alias rows cannot drift apart.
@@ -3054,7 +2878,7 @@ const getModelRate = (modelId) => {
 *
 * Keys are the canonical model IDs stella resolves for the provider
 * adapters (the same forms used in `MODEL_RATES`, `BYOK_MODEL_OPTIONS`,
-* and `DEFAULT_MODELS`, including OpenRouter provider-prefixed slugs and
+* and the task defaults ledger, including OpenRouter provider-prefixed slugs and
 * AWS Bedrock IDs). Values are the providers' publicly documented input
 * windows; where a provider offers a larger beta window we intentionally
 * take the conservative default (e.g. Claude's standard 200K rather than
@@ -3172,4 +2996,4 @@ const getModelReasoningCapabilities = (modelId) => {
 	return Object.hasOwn(MODEL_REASONING_CAPABILITIES_BY_ID, normalized) ? MODEL_REASONING_CAPABILITIES_BY_ID[normalized] ?? null : null;
 };
 //#endregion
-export { MODEL_RATES as $, getModelReasoningCapabilities as A, supportsStreamingToolUse as B, TANSTACK_AI_PROVIDERS as C, getModelDisplayMetadata as D, getModelDefaultReasoningEffort as E, isChatPdfAttachmentModelSupported as F, MODELS_DEV_RATE_PROVIDER_BY_CATALOG_PROVIDER as G, CAPABILITY_OVERRIDES as H, normalizeModelCatalogId as I, MODEL_RATE_UNITS_PER_USD as J, MODELS_DEV_RATE_SOURCE_ALIASES as K, resolveReasoningEffort as L, getOutputTokenLimit as M, isBYOKModelRoleSupported as N, getModelImageInputCapability as O, isBYOKProviderRoleSupported as P, REASONING_REPLAY_FORMATS as Q, resolveWorkingBYOKModelForRole as R, STREAMING_TOOL_USE_SUPPORTS as S, getContextWindowTokens as T, IMAGE_INPUT_OVERRIDES as U, DOCUMENT_INPUT_OVERRIDES as V, MODELS_DEV_RATE_CORRECTIONS as W, resolveModelRate as X, getStandardModelRate as Y, MODEL_REASONING_CAPABILITIES as Z, MODEL_DISPLAY_METADATA as _, BYOK_MODEL_OPTIONS as a, REASONING_EFFORTS as b, DECISION_MODEL_CATALOG as c, DEFAULT_MODELS as d, MODEL_DEFAULT_REASONING_EFFORTS as et, FALLBACK_CHAT_MODEL_BY_PROVIDER as f, MODEL_CATALOG_PROVIDER_KIND as g, MODEL_CATALOG_ID_ALIASES as h, BYOK_DOCUMENT_INPUT_MODEL_OPTIONS as i, MODEL_TEMPERATURE_POLICIES as it, getModelReasoningEfforts as j, getModelRate as k, DECISION_MODEL_PROVIDERS as l, IMAGE_INPUT_CAPABILITIES as m, ANTHROPIC_ADAPTIVE_THINKING_MODELS as n, MODEL_OUTPUT_TOKEN_LIMITS as nt, CHAT_PDF_ATTACHMENT_MODEL_OPTIONS as o, FIRST_PARTY_MODEL_PROVIDERS as p, RETAINED_MODELS_DEV_RATE_ENTRIES as q, BYOK_DEFAULT_MODELS as r, MODEL_REASONING_EFFORTS as rt, CONTEXT_WINDOW_TOKENS as s, AI_PROVIDERS as t, MODEL_IMAGE_INPUT_CAPABILITIES as tt, DEFAULT_CONTEXT_WINDOW_TOKENS as u, MODEL_ROLES as v, TEMPERATURE_POLICIES as w, RECOMMENDED_CHAT_MODELS as x, MODEL_STREAMING_TOOL_USE as y, shouldEmitTemperature as z };
+export { MODEL_OUTPUT_TOKEN_LIMITS as $, isBYOKModelRoleSupported as A, IMAGE_INPUT_OVERRIDES as B, getModelDefaultReasoningEffort as C, getModelReasoningCapabilities as D, getModelRate as E, resolveWorkingBYOKModelForTask as F, MODEL_RATE_UNITS_PER_USD as G, MODELS_DEV_RATE_PROVIDER_BY_CATALOG_PROVIDER as H, shouldEmitTemperature as I, MODEL_REASONING_CAPABILITIES as J, getStandardModelRate as K, supportsStreamingToolUse as L, isChatPdfAttachmentModelSupported as M, normalizeModelCatalogId as N, getModelReasoningEfforts as O, resolveReasoningEffort as P, MODEL_IMAGE_INPUT_CAPABILITIES as Q, DOCUMENT_INPUT_OVERRIDES as R, getContextWindowTokens as S, getModelImageInputCapability as T, MODELS_DEV_RATE_SOURCE_ALIASES as U, MODELS_DEV_RATE_CORRECTIONS as V, RETAINED_MODELS_DEV_RATE_ENTRIES as W, MODEL_RATES as X, REASONING_REPLAY_FORMATS as Y, MODEL_DEFAULT_REASONING_EFFORTS as Z, REASONING_EFFORTS as _, BYOK_MODEL_OPTIONS as a, TANSTACK_AI_PROVIDERS as b, DECISION_MODEL_CATALOG as c, FIRST_PARTY_MODEL_PROVIDERS as d, MODEL_REASONING_EFFORTS as et, IMAGE_INPUT_CAPABILITIES as f, MODEL_STREAMING_TOOL_USE as g, MODEL_DISPLAY_METADATA as h, BYOK_DOCUMENT_INPUT_MODEL_OPTIONS as i, isBYOKProviderRoleSupported as j, getOutputTokenLimit as k, DECISION_MODEL_PROVIDERS as l, MODEL_CATALOG_PROVIDER_KIND as m, ANTHROPIC_ADAPTIVE_THINKING_MODELS as n, CHAT_PDF_ATTACHMENT_MODEL_OPTIONS as o, MODEL_CATALOG_ID_ALIASES as p, resolveModelRate as q, BYOK_DEFAULT_MODELS as r, CONTEXT_WINDOW_TOKENS as s, AI_PROVIDERS as t, MODEL_TEMPERATURE_POLICIES as tt, DEFAULT_CONTEXT_WINDOW_TOKENS as u, RECOMMENDED_CHAT_MODELS as v, getModelDisplayMetadata as w, TEMPERATURE_POLICIES as x, STREAMING_TOOL_USE_SUPPORTS as y, CAPABILITY_OVERRIDES as z };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stll/ai-catalog",
-  "version": "0.7.1",
+  "version": "0.8.1",
   "description": "Provider-neutral AI model roles, capabilities, and curated model metadata.",
   "keywords": [
     "ai",
@@ -36,6 +36,10 @@
     "./benchmarks": {
       "types": "./dist/benchmarks.d.ts",
       "import": "./dist/benchmarks.js"
+    },
+    "./model-defaults-ledger": {
+      "types": "./dist/model-defaults-ledger.d.ts",
+      "import": "./dist/model-defaults-ledger.js"
     }
   },
   "publishConfig": {
@@ -47,9 +51,11 @@
     "gen:benchmarks": "bun ../scripts/src/model-catalog-benchmarks-gen.ts",
     "gen:capabilities": "bun --preload ../../scripts/offline-network-preload.ts ../scripts/src/model-catalog-capabilities-gen.ts",
     "check:generations": "bun --preload ../../scripts/offline-network-preload.ts ../scripts/src/model-catalog-generation-check.ts",
+    "check:defaults": "bun ../../scripts/check-model-defaults-ledger.ts",
     "gen:rates": "bun --preload ../../scripts/offline-network-preload.ts ../scripts/src/model-catalog-rates-gen.ts",
     "pack:dry-run": "bun pm pack --dry-run",
     "test": "bun test src",
+    "test:defaults": "bun test src/model-defaults-ledger.test.ts",
     "test:property": "bun test --preload @stll/property-testing/preload $(grep -rlE 'fc\\.assert|assertProperty' src --include='*.test.ts')",
     "typecheck": "bun check --no-pretty --all --project=tsconfig.json",
     "lint": "cd ../.. && bun --bun oxlint -c oxlint.config.ts --report-unused-disable-directives-severity=error --type-aware --type-check packages/ai-catalog",