fix(catalog): synthesize incomplete combo members with context fallback (#1163)

docs-site/src/content/docs/reference/configuration/routing.md

@@ -184,13 +184,16 @@ Per-request route-decision traces are recorded when a policy profile executes.
 A combo remains directly routable even when it cannot be listed. `ocx sync`, `/v1/models`, and the
 Codex picker list it only when every target exposes capabilities that can be intersected:
 
-- a positive `contextWindow`, from live metadata, registry hints, or provider
-  `modelContextWindows` / `contextWindow`; and
+- a positive `contextWindow`, from live metadata, registry hints, provider
+  `modelContextWindows` / `contextWindow`, a known positive `maxInputTokens` on the member row,
+  or — when the provider is known and enabled but every source still omits a window — a
+  conservative 128,000-token fallback (clamped by `providerContextCaps` when set); and
 - a non-empty `inputModalities` intersection, treating an omitted member value as `["text"]`.
 
-A bare relay id with no context metadata or targets with disjoint modalities removes the combo from
-the catalog. Sync emits a summary warning and the dashboard marks it **Needs attention**. Add context
-metadata, align modalities, or target models with discoverable compatible capabilities.
+A target on a disabled provider (even with a complete discovery row), on an unknown provider with
+no discovery row, or targets with disjoint modalities, removes the combo from the catalog. Sync
+emits a summary warning and the dashboard marks it **Needs attention**. Add context metadata,
+align modalities, or target models with discoverable compatible capabilities.
 
 ## Request history and routing analytics
 

docs-site/src/content/docs/zh-cn/reference/configuration/routing.md

@@ -82,11 +82,12 @@ selector 校验、冲突规则和隐私说明见[提供方配置](/reference/con
 
 即使某个 combo 不能被列出，它仍然可以直接路由。只有当所有目标都暴露出可以交集的能力时，`ocx sync`、`/v1/models` 和 Codex 选择器才会列出它：
 
-- 一个正的 `contextWindow`，来源可以是实时元数据、注册表提示，或提供方的
-  `modelContextWindows` / `contextWindow`；以及
+- 一个正的 `contextWindow`，来源可以是实时元数据、注册表提示、提供方的
+  `modelContextWindows` / `contextWindow`、成员行上已知的正 `maxInputTokens`，或者——当提供方已知且启用但所有来源仍未给出窗口时——
+  保守的 128,000 token 回退（若配置了 `providerContextCaps` 则会按上限夹紧）；以及
 - 非空的 `inputModalities` 交集，其中省略的成员值按 `["text"]` 处理。
 
-如果是一个没有上下文元数据的裸 relay id，或者目标之间的模态互不相交，combo 就会从
+目标位于已禁用提供方（即使有完整 discovery 行）、未知且无 discovery 行的提供方，或目标之间的模态互不相交时，combo 会从
 目录中移除。同步时会输出一条汇总警告，仪表板会将其标记为 **Needs attention**。
 补充上下文元数据、对齐模态，或者把目标模型切换为可发现且兼容的能力。
 

src/codex/catalog.ts

@@ -5,7 +5,7 @@ export type { CatalogModel, MultiAgentMode } from "./catalog/parsing";
 export { NATIVE_OPENAI_MODELS, nativeOpenAiContextWindow, disabledNativeSlugs, visibleNativeSlugs, desktopVisibleNativeSlugs, nativeModelRows, applyNativeVisibility, upstreamNativeEntry, nativeOpenAiSlugs, listCatalogNativeSlugs, nativeReasoningEfforts, nativeDefaultReasoningEffort, shouldIncludeAccountBoundNativeOpenAi, shouldIncludeNativeOpenAi } from "./catalog/metadata";
 export { isSpawnableCodexCandidate, codexExecInvocation, loadBundledCodexCatalog, materializeBundledCodexCatalog, loadCatalogTemplate } from "./catalog/bundled";
 export { nativeEffortClamp, shouldApplyNativeEffortClamp, catalogModelEfforts, codexSupportedReasoningEfforts, clampedDefaultEffort, clampEntryToCodexSupportedEfforts, clampCatalogModelsToCodexSupport } from "./catalog/effort";
-export { applyProviderConfigHints, isDatedVariantId, filterCatalogVisibleModels, gatherRoutedModels, clearGatherRoutedModelsInflight, augmentRoutedModelsWithRegistryOpenAiApiRows, augmentRoutedModelsWithMetadata } from "./catalog/provider-fetch";
+export { applyProviderConfigHints, isDatedVariantId, filterCatalogVisibleModels, gatherRoutedModels, clearGatherRoutedModelsInflight, augmentRoutedModelsWithRegistryOpenAiApiRows, augmentRoutedModelsWithMetadata, resolveComboCatalogMember } from "./catalog/provider-fetch";
 export { deriveComboCatalogModel, exactComboCatalogSlugs, getLastComboCatalogOmissions, resetOpenAiApiCatalogWarningStateForTests, uniqueCatalogModelsForPublicList, uniqueCatalogModelsForRawPublicList, buildComboCatalogOmission, comboCatalogOmissionReason, summarizeComboCatalogOmissions } from "./catalog/aggregation";
 export type { ComboCatalogOmission, ComboCatalogOmissionReason } from "./catalog/aggregation";
 export { MAX_SPAWN_AGENT_MODEL_OVERRIDES, CANONICAL_NATIVE_CATALOG_CONTENT_POLICY, effectiveSubagentRoster, buildCatalogEntries, mergeCatalogEntriesFromObservedState, resetCatalogRuntimeStateForTests, orderForSubagents, mergeCatalogEntriesForSync, syncCatalogModels, restoreCodexCatalog, invalidateCodexModelsCache } from "./catalog/sync";

src/codex/catalog/aggregation.ts

@@ -131,9 +131,14 @@ export function deriveComboCatalogModel(
   const inputModalities = intersectStrings(
     members.map(member => member.inputModalities ?? ["text"]),
   );
-  const reasoningEfforts = intersectStrings(
-    members.map(member => member.reasoningEfforts ?? []),
-  );
+  // Unknown ladders (`undefined`) are wildcards for catalog derivation — same
+  // boundary as the GUI picker. An explicit empty ladder still constrains.
+  const advertisedLadders = members
+    .map(member => member.reasoningEfforts)
+    .filter((ladder): ladder is string[] => ladder !== undefined);
+  const reasoningEfforts = advertisedLadders.length === 0
+    ? []
+    : intersectStrings(advertisedLadders);
   const contextWindow = Math.min(...members.map(member => member.contextWindow!));
   const maxInputTokens = Math.min(
     ...members.map(member => member.maxInputTokens ?? member.contextWindow!),

src/codex/catalog/provider-fetch.ts

@@ -620,6 +620,104 @@ export function applyConfigHintsToCachedModels(name: string, prov: OcxProviderCo
   return models.map(model => applyProviderConfigHints(name, prov, model, contextCap));
 }
 
+
+/**
+ * Last-resort context window for combo member synthesis when discovery and
+ * provider config both omit one. Matches the catalog entry default in
+ * `normalizeRoutedCatalogEntry` so incomplete live rows still catalog.
+ */
+const COMBO_MEMBER_CONTEXT_FALLBACK = 128_000;
+
+/**
+ * Resolve a combo target to a catalog member for derivation.
+ * Prefer discovery metadata; when the target is missing from the gather map or
+ * lacks a positive contextWindow, synthesize from the (registry-enriched)
+ * provider config so combos remain catalogued when targets are configured but
+ * discovery metadata is incomplete. Disabled providers stay unresolved.
+ * When hints still omit contextWindow, prefer known maxInputTokens, else
+ * COMBO_MEMBER_CONTEXT_FALLBACK so a live row without ctx does not drop the
+ * whole combo from the public catalog.
+ */
+export function resolveComboCatalogMember(
+  target: { provider: string; model: string },
+  memberByKey: ReadonlyMap<string, CatalogModel>,
+  providers: ReadonlyMap<string, OcxProviderConfig>,
+  contextCap?: number,
+): CatalogModel | undefined {
+  const existing = memberByKey.get(targetKey(target));
+  const prov = providers.get(target.provider);
+  // Disabled providers never contribute members — even a complete discovery row
+  // is unusable for catalog derivation while the provider is off.
+  if (prov?.disabled === true) return undefined;
+
+  // Complete live/configured rows still honor providerContextCaps so a high
+  // discovery window cannot outrun an operator-configured cap.
+  if (
+    existing
+    && typeof existing.contextWindow === "number"
+    && existing.contextWindow > 0
+  ) {
+    const capped = applyProviderContextCap(existing.contextWindow, contextCap);
+    if (capped === undefined || capped === existing.contextWindow) return existing;
+    const maxInput = typeof existing.maxInputTokens === "number" && existing.maxInputTokens > 0
+      ? Math.min(existing.maxInputTokens, capped)
+      : capped;
+    return {
+      ...existing,
+      contextWindow: capped,
+      maxInputTokens: maxInput,
+      contextCap,
+      contextCapped: true as const,
+    };
+  }
+
+  const base: CatalogModel = existing ?? {
+    id: target.model,
+    provider: target.provider,
+  };
+  const hinted = prov
+    ? applyProviderConfigHints(target.provider, prov, base, contextCap)
+    : base;
+  const hintedContext = typeof hinted.contextWindow === "number" && hinted.contextWindow > 0
+    ? hinted.contextWindow
+    : undefined;
+  // Prefer a known positive maxInputTokens over inventing 128k when discovery
+  // advertised an input limit but no context window (common thin /models rows).
+  const knownMaxInput = typeof hinted.maxInputTokens === "number" && hinted.maxInputTokens > 0
+    ? hinted.maxInputTokens
+    : (typeof base.maxInputTokens === "number" && base.maxInputTokens > 0
+      ? base.maxInputTokens
+      : undefined);
+  const uncappedContext = hintedContext
+    ?? knownMaxInput
+    ?? (existing || prov ? COMBO_MEMBER_CONTEXT_FALLBACK : undefined);
+  if (uncappedContext === undefined) return undefined;
+  const usedFallback = hintedContext === undefined;
+  const cappedContext = applyProviderContextCap(uncappedContext, contextCap);
+  const contextWindow = cappedContext ?? uncappedContext;
+  const fallbackCapped = usedFallback
+    && contextCap !== undefined
+    && cappedContext !== undefined
+    && cappedContext !== uncappedContext;
+
+  const inputModalities = hinted.inputModalities ?? base.inputModalities ?? ["text"];
+  const reasoningEfforts = hinted.reasoningEfforts
+    ?? (prov ? configuredReasoningEfforts(prov, target.model) : undefined)
+    ?? base.reasoningEfforts;
+  const maxInputTokens = knownMaxInput !== undefined
+    ? Math.min(knownMaxInput, contextWindow)
+    : contextWindow;
+
+  return {
+    ...hinted,
+    inputModalities,
+    ...(reasoningEfforts !== undefined ? { reasoningEfforts } : {}),
+    contextWindow,
+    maxInputTokens,
+    ...(fallbackCapped ? { contextCap, contextCapped: true as const } : {}),
+  };
+}
+
 export function isDatedVariantId(liveId: string, configuredId: string): boolean {
   if (!liveId.startsWith(`${configuredId}-`)) return false;
   return /^\d{8}$/.test(liveId.slice(configuredId.length + 1));
@@ -1319,21 +1417,26 @@ async function gatherRoutedModelsUncached(
       if (!memberByKey.has(key)) memberByKey.set(key, synthetic);
     }
   }
+  // Enriched (registry-hydrated) provider clones — shared by combo member synthesis and
+  // custom-model vision-sidecar inheritance so both see the same merged registry view.
+  const enrichedByName = new Map(activeProviders.map(provider => [provider.name, provider.provider]));
   for (const id of listComboIds(config)) {
     const combo = getCombo(config, id);
     if (!combo) continue;
     const members = combo.targets
-      .map(target => memberByKey.get(targetKey(target)))
+      .map(target => resolveComboCatalogMember(
+        target,
+        memberByKey,
+        enrichedByName,
+        providerContextCap(config, target.provider),
+      ))
       .filter((member): member is CatalogModel => member !== undefined);
     const derived = deriveComboCatalogModel(id, combo, members);
     if (derived) all.push(derived);
     else warnUncataloguedComboOnce(id, combo, members, localOmissions);
   }
   replaceLastComboCatalogOmissions(localOmissions);
   all.sort((a, b) => (a.provider === b.provider ? a.id.localeCompare(b.id) : a.provider.localeCompare(b.provider)));
-  // Enriched (registry-hydrated) provider clones, keyed by name — the same view used above so
-  // custom rows get the same noVisionModels / inputModalities treatment as discovered rows.
-  const enrichedByName = new Map(activeProviders.map(provider => [provider.name, provider.provider]));
   // Provider-derived rows keyed by their Codex-facing slug: a custom override replaces the row
   // with the same slug below, so that row's provider capability metadata is the inheritance source.
   const replacedByRoutedSlug = new Map(all.map(model => [routedSlug(model.provider, model.id), model]));