diff --git a/CHANGELOG.md b/CHANGELOG.md
index 502c7826..283eafe5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.154.0
+
+### A promotion may not call a candidate cheaper on dollars nobody measured
+
+`BenchmarkCell` carries `usd` and `usdKnown` as required siblings. `promotionGate` read `usd` from both arms and never read `usdKnown`, while non-inferiority mode promotes on a significant paired cost saving. A candidate whose dollars were a catalog estimate or an unreported floor could therefore be promoted for being cheaper than a receipted incumbent.
+
+`Spend.usdKnown` states the rule: a false value must not be treated as a measurement when enforcing a dollar-denominated comparison or limit. Promotion on cost savings is that comparison.
+
+`PromotionVerdict.reason` gains `'cost-unknown'`, and the verdict carries `costUnknownTasks` naming the tasks that caused the refusal. A caller that exhaustively switches on `reason` must handle the new member; a caller in `superiority` mode is unaffected, because only non-inferiority mode reads dollars.
+
 ## 0.153.2
 
 ### Eval 0.163.2 and Knowledge 10.7.0 reach this package
diff --git a/docs/api/primitive-catalog.md b/docs/api/primitive-catalog.md
index ecf6357d..39457d1f 100644
--- a/docs/api/primitive-catalog.md
+++ b/docs/api/primitive-catalog.md
@@ -7,7 +7,7 @@
 
 # Primitive catalog — the never-stale anti-reinvention inventory
 
-> **GENERATED** from `@tangle-network/agent-runtime@0.153.2` and `@tangle-network/agent-eval@0.163.2` by `scripts/gen-primitive-catalog.mjs`. Do NOT hand-edit — run `pnpm run docs:api`. This is the mechanical companion to the JUDGMENT in `canonical-api.md` (§2 decision table + §1.5 AgentProfile law): that doc says WHICH primitive to reach for and what NOT to build; this catalog proves WHAT exists. Per-symbol signatures + `file:line` live in the per-module pages under `docs/api/`.
+> **GENERATED** from `@tangle-network/agent-runtime@0.154.0` and `@tangle-network/agent-eval@0.163.2` by `scripts/gen-primitive-catalog.mjs`. Do NOT hand-edit — run `pnpm run docs:api`. This is the mechanical companion to the JUDGMENT in `canonical-api.md` (§2 decision table + §1.5 AgentProfile law): that doc says WHICH primitive to reach for and what NOT to build; this catalog proves WHAT exists. Per-symbol signatures + `file:line` live in the per-module pages under `docs/api/`.
 
 ## 1. agent-runtime — own public surface
 
diff --git a/docs/api/runtime.md b/docs/api/runtime.md
index 9f2a4dcf..a8238610 100644
--- a/docs/api/runtime.md
+++ b/docs/api/runtime.md
@@ -5854,7 +5854,7 @@ Fixed by the substrate by default — the same report always yields the same ver
 
 ##### reason
 
-> **reason**: `"identical-champion"` \| `"few-tasks"` \| `"no-margin"` \| `"significant"` \| `"non-inferior-and-cheaper"` \| `"non-inferiority-unproven"` \| `"not-cheaper"`
+> **reason**: `"identical-champion"` \| `"few-tasks"` \| `"no-margin"` \| `"significant"` \| `"non-inferior-and-cheaper"` \| `"non-inferiority-unproven"` \| `"not-cheaper"` \| `"cost-unknown"`
 
 ##### mode
 
@@ -5912,6 +5912,14 @@ non-inferiority mode: paired (incumbent − candidate) cost savings per task (us
 
 > **high**: `number`
 
+##### costUnknownTasks?
+
+> `optional` **costUnknownTasks?**: `string`[]
+
+non-inferiority mode: the tasks whose dollars were not measured on at least one arm.
+ Present only with `reason: 'cost-unknown'`; naming them is what makes the refusal
+ actionable instead of a bare no.
+
 ##### latency?
 
 > `optional` **latency?**: `object`
diff --git a/docs/canonical-api.md b/docs/canonical-api.md
index 1997e1b0..1c43f754 100644
--- a/docs/canonical-api.md
+++ b/docs/canonical-api.md
@@ -4,7 +4,7 @@
 Generated signatures and the complete export list live in docs/api/.
 Run pnpm docs:freshness after editing this file. -->
 
-> **Version 0.153.2.**
+> **Version 0.154.0.**
 > [`docs/api/primitive-catalog.md`](./api/primitive-catalog.md) lists every export and import path.
 > `agent-eval` must satisfy `>=0.163.2 <0.164.0`.
 > `sandbox` must satisfy `>=0.31.0 <0.32.0`.
diff --git a/package.json b/package.json
index 12673f50..5bf05793 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-runtime",
-  "version": "0.153.2",
+  "version": "0.154.0",
   "description": "Shared task-lifecycle skeleton for agents: a recursive loop kernel for chat turns, one-shot tasks, and multi-attempt loops, with trace capture and eval-gated self-improvement. Domain behavior lives in adapters; scoring and ship-gates in @tangle-network/agent-eval.",
   "homepage": "https://github.com/tangle-network/agent-runtime#readme",
   "repository": {
diff --git a/src/runtime/promotion-gate.ts b/src/runtime/promotion-gate.ts
index 4ce2a5fc..8175aaba 100644
--- a/src/runtime/promotion-gate.ts
+++ b/src/runtime/promotion-gate.ts
@@ -45,6 +45,7 @@ export interface PromotionVerdict {
     | 'non-inferior-and-cheaper'
     | 'non-inferiority-unproven'
     | 'not-cheaper'
+    | 'cost-unknown'
   mode: 'superiority' | 'non-inferiority'
   /** Paired tasks that carried both strategies' cells. */
   n: number
@@ -54,6 +55,10 @@ export interface PromotionVerdict {
   /** non-inferiority mode: paired (incumbent − candidate) cost savings per task (usd).
    *  Positive means the candidate is cheaper; `low` and `high` carried the decision. */
   costSavings?: { mean: number; median: number; low: number; high: number }
+  /** non-inferiority mode: the tasks whose dollars were not measured on at least one arm.
+   *  Present only with `reason: 'cost-unknown'`; naming them is what makes the refusal
+   *  actionable instead of a bare no. */
+  costUnknownTasks?: string[]
   /** Paired (candidate − incumbent) wall-clock per task (ms) — negative = the candidate
    *  is FASTER. Informational in every mode (never gates); the latency answer to "what
    *  does this win actually cost the user?". */
@@ -79,6 +84,8 @@ export function promotionGate(opts: PromotionGateOptions): PromotionVerdict {
   const incMs: number[] = []
   const candMs: number[] = []
   const cellIds: string[] = []
+  /** Tasks where either arm's dollars were never measured — see the refusal below. */
+  const costUnknownTasks: string[] = []
   for (const row of opts.report.perTask) {
     const inc = row.cells?.[opts.incumbent]
     const cand = row.cells?.[opts.candidate]
@@ -87,6 +94,7 @@ export function promotionGate(opts: PromotionGateOptions): PromotionVerdict {
     after.push(cand.score)
     incUsd.push(inc.usd)
     candUsd.push(cand.usd)
+    if (inc.usdKnown === false || cand.usdKnown === false) costUnknownTasks.push(row.taskId)
     incMs.push(inc.ms)
     candMs.push(cand.ms)
     cellIds.push(row.taskId)
@@ -147,6 +155,22 @@ export function promotionGate(opts: PromotionGateOptions): PromotionVerdict {
       ...(opts.resamples !== undefined ? { resamples: opts.resamples } : {}),
     },
   )
+  // `BenchmarkCell.usdKnown` false means the dollars are a floor, not a measurement, and
+  // `Spend.usdKnown` states the rule this gate has to obey: an unknown dollar amount "must not be
+  // treated as $0 when enforcing a dollar-denominated comparison or limit". Promotion on cost
+  // savings IS that comparison, so an unmeasured arm cannot be declared cheaper — it can only be
+  // declared unmeasured. Refuse rather than promote on a number no provider ever billed.
+  if (costUnknownTasks.length > 0) {
+    return {
+      promoted: false,
+      reason: 'cost-unknown',
+      mode,
+      n: before.length,
+      lift,
+      costUnknownTasks: [...costUnknownTasks],
+      latency,
+    }
+  }
   const costSig = heldoutSignificance(
     { before: candUsd, after: incUsd, cellIds },
     {
diff --git a/src/testing/fixtures/agent-improvement-proposal.json b/src/testing/fixtures/agent-improvement-proposal.json
index 0227b6e8..9c8a1f69 100644
--- a/src/testing/fixtures/agent-improvement-proposal.json
+++ b/src/testing/fixtures/agent-improvement-proposal.json
@@ -1,6 +1,6 @@
 {
   "changedSurfaces": ["prompt"],
-  "digest": "sha256:6af30ed1418652d00082d07f90766d0ee30681db2d13322885972ea6d732b3c0",
+  "digest": "sha256:f78c77e891758f0bb6d260e19c9eeb2a2171a3a8b000e51b6e934b4b79ffb0b8",
   "evaluation": {
     "decision": {
       "contributingChecks": [
@@ -4882,7 +4882,7 @@
     ],
     "metadata": {
       "fixture": "agent-improvement-proposal",
-      "runtimeVersion": "0.153.2"
+      "runtimeVersion": "0.154.0"
     },
     "objectives": [
       {
@@ -4993,8 +4993,8 @@
       "baselineContentHash": "sha256:5c21ee53e513fc604cb09754e21c392b24a424da0ef37dbf8f1ee4a8a0b08f09",
       "candidateContentHash": "sha256:60fcbb1c728194bd51d7d19cb732d1c3f1881dce7e0a6266b41c8b98cfd65693",
       "kind": "agent-eval-loop",
-      "recordDigest": "sha256:2cbc803968b26ba66887d9c096a5d5383d416b32990997d89c763a1e13503241",
-      "runId": "agent-runtime-0.153.2-proposal-fixture",
+      "recordDigest": "sha256:d60f130db2672f9ccb836ba15bded1d90481c38368810cbd9f64a9e4ef84a2fb",
+      "runId": "agent-runtime-0.154.0-proposal-fixture",
       "schema": "agent-candidate-experiment"
     }
   },
@@ -5021,5 +5021,5 @@
   ],
   "kind": "agent-improvement-proposal",
   "proposedAt": "2026-07-10T01:00:00.000Z",
-  "runId": "agent-runtime-0.153.2-proposal-fixture"
+  "runId": "agent-runtime-0.154.0-proposal-fixture"
 }
diff --git a/src/testing/fixtures/agent-profile-improvement-proposal.json b/src/testing/fixtures/agent-profile-improvement-proposal.json
index 20f4dac1..5b62f57f 100644
--- a/src/testing/fixtures/agent-profile-improvement-proposal.json
+++ b/src/testing/fixtures/agent-profile-improvement-proposal.json
@@ -1,6 +1,6 @@
 {
   "changedSurfaces": ["prompt", "skills"],
-  "digest": "sha256:731dab22aaa07732a81639375387cf9ed15bfe3547601925ee51f2e629da5154",
+  "digest": "sha256:ca08433b17a8a4a7708f2f6da706dfd18d2dd9f8e549b17be0a8876e13303c88",
   "evaluation": {
     "decision": {
       "contributingChecks": [
@@ -1715,7 +1715,7 @@
     ],
     "metadata": {
       "fixture": "agent-profile-improvement-proposal",
-      "runtimeVersion": "0.153.2"
+      "runtimeVersion": "0.154.0"
     },
     "objectives": [
       {
@@ -1826,7 +1826,7 @@
       "baselineContentHash": "sha256:21c495a37c418c10bde64fbaa188beddeed31f1f051ea60a6a6582a9ee0db704",
       "candidateContentHash": "sha256:103f77bc8481601eef1ad5fe6ba84a40dffabc3a44f421f8c8559121edab84e9",
       "kind": "agent-eval-loop",
-      "recordDigest": "sha256:f50ab9d3e539e3a9c7a10964d69250a5ea719eeb9cebe4099086a9cdc7d4505f",
+      "recordDigest": "sha256:42a01d47c8b81447ff30d54dcc7ef7890eac946f9113e4e60045dc6ff64c7b08",
       "runId": "profile-improvement-1",
       "schema": "agent-profile-improvement-experiment"
     }
diff --git a/tests/kernel/strategy-suite.test.ts b/tests/kernel/strategy-suite.test.ts
index bab8079a..2c403467 100644
--- a/tests/kernel/strategy-suite.test.ts
+++ b/tests/kernel/strategy-suite.test.ts
@@ -682,6 +682,7 @@ describe('promotionGate non-inferiority', () => {
       candScore: number
       incUsd: number
       candUsd: number
+      candUsdKnown?: boolean
     }>,
   ): BenchmarkReport => ({
     n: rows.length,
@@ -703,6 +704,7 @@ describe('promotionGate non-inferiority', () => {
           resolved: false,
           progression: [],
           usd: r.candUsd,
+          ...(r.candUsdKnown !== undefined ? { usdKnown: r.candUsdKnown } : {}),
           ms: 0,
           tokens: { input: 0, output: 0 },
         },
@@ -737,6 +739,39 @@ describe('promotionGate non-inferiority', () => {
     expect(v.latency).toBeDefined()
   })
 
+  it('refuses to call a candidate cheaper when its dollars were never measured', () => {
+    // The same inputs that PROMOTE above, with one arm's dollars marked unmeasured. `usdKnown:
+    // false` means `usd` is a floor, and `Spend.usdKnown` states the rule: an unknown amount must
+    // not be treated as a measurement "when enforcing a dollar-denominated comparison or limit".
+    // Promotion on cost savings is exactly that comparison.
+    const scoreDeltas = [-0.018, -0.012, -0.008, -0.004, 0, 0.006, 0.011, -0.006]
+    const costSavings = [0.011, 0.014, 0.016, 0.013, 0.018, 0.015, 0.012, 0.017]
+    const rows = Array.from({ length: 24 }, (_, i) => {
+      const incScore = 0.48 + (i % 7) * 0.055
+      const incUsd = 0.024 + (i % 5) * 0.0014
+      return {
+        id: `t${i}`,
+        incScore,
+        candScore: incScore + scoreDeltas[i % scoreDeltas.length]!,
+        incUsd,
+        candUsd: incUsd - costSavings[i % costSavings.length]!,
+        // Only the first task is unmeasured; one unmeasured pair is enough to void the comparison.
+        ...(i === 0 ? { candUsdKnown: false } : {}),
+      }
+    })
+    const v = promotionGate({
+      report: costReport(rows),
+      incumbent: 'incumbent',
+      candidate: 'candidate',
+      mode: 'non-inferiority',
+    })
+    expect(v.promoted).toBe(false)
+    expect(v.reason).toBe('cost-unknown')
+    expect(v.costUnknownTasks).toEqual(['t0'])
+    // The refusal replaces the cost verdict rather than reporting one built on the same numbers.
+    expect(v.costSavings).toBeUndefined()
+  })
+
   it('cheaper but score-inferior beyond tolerance LOSES', () => {
     const rows = Array.from({ length: 12 }, (_, i) => ({
       id: `t${i}`,