fix(promotion): report the interval used for decisions

.github/workflows/ci.yml

@@ -77,14 +77,14 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           repository: tangle-network/agent-eval
-          ref: 8e4d6c95ca5beb6fc62a7c3c52c9fb581b928795 # v0.135.1
+          ref: 1243ca2fa7ac9f91ada7c85b71b9f216f1b3f4ec # v0.135.2
           path: .cohort/agent-eval
           persist-credentials: false
 
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           repository: tangle-network/agent-knowledge
-          ref: 502ef4954c0dc85eb0f629bfb6c062b3508977ef # v6.1.10
+          ref: 47d046bdafdcd292af8104e84107c48f5a80f505 # v6.1.11
           path: .cohort/agent-knowledge
           persist-credentials: false
 

.github/workflows/publish.yml

@@ -143,15 +143,15 @@ jobs:
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           repository: tangle-network/agent-eval
-          ref: 8e4d6c95ca5beb6fc62a7c3c52c9fb581b928795 # v0.135.1
+          ref: 1243ca2fa7ac9f91ada7c85b71b9f216f1b3f4ec # v0.135.2
           path: .cohort/agent-eval
           persist-credentials: false
 
       - name: Check out exact Knowledge release
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           repository: tangle-network/agent-knowledge
-          ref: 502ef4954c0dc85eb0f629bfb6c062b3508977ef # v6.1.10
+          ref: 47d046bdafdcd292af8104e84107c48f5a80f505 # v6.1.11
           path: .cohort/agent-knowledge
           persist-credentials: false
 

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.109.2
+
+- Align Runtime with Eval 0.135.2 and Knowledge 6.1.11 so every improvement path uses the corrected paired promotion decisions.
+
 ## 0.109.1
 
 - Move `runToolLoop` and `streamToolLoop` to the Worker-safe `@tangle-network/agent-runtime/tool-loop` entrypoint.

bench/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.4.9
+
+- Consume Runtime 0.109.2 with Eval 0.135.2 and Knowledge 6.1.11 so benchmark runs use the corrected paired promotion decisions.
+
 ## 0.4.8
 
 - Consume Runtime 0.109.0 through the canonical `./kernel` entrypoint.

bench/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-bench",
-  "version": "0.4.8",
+  "version": "0.4.9",
   "type": "module",
   "description": "Benchmark adapters and execution for agent-runtime across coding, tool-use, RAG, memory, browser, and terminal tasks.",
   "repository": {

bench/scripts/wait-for-published-dependencies.test.mjs

@@ -10,8 +10,8 @@ test('collects required published Tangle dependencies from a packed manifest', (
   const dependencies = collectRequiredTangleDependencies({
     name: '@tangle-network/agent-bench',
     dependencies: {
-      '@tangle-network/agent-runtime': '0.109.0',
-      '@tangle-network/agent-eval': '0.135.1',
+      '@tangle-network/agent-runtime': '0.109.2',
+      '@tangle-network/agent-eval': '0.135.2',
       undici: '^7.0.0',
     },
     peerDependencies: {
@@ -27,9 +27,9 @@ test('collects required published Tangle dependencies from a packed manifest', (
   })
 
   assert.deepEqual(dependencies, [
-    { name: '@tangle-network/agent-eval', spec: '0.135.1' },
+    { name: '@tangle-network/agent-eval', spec: '0.135.2' },
     { name: '@tangle-network/agent-interface', spec: '>=0.36.0 <0.37.0' },
-    { name: '@tangle-network/agent-runtime', spec: '0.109.0' },
+    { name: '@tangle-network/agent-runtime', spec: '0.109.2' },
   ])
 })
 

docs/api/primitive-catalog.md

@@ -7,7 +7,7 @@
 
 # Primitive catalog — the never-stale anti-reinvention inventory
 
-> **GENERATED** from `@tangle-network/agent-runtime@0.109.1` and `@tangle-network/agent-eval@0.135.1` 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@0.109.2` and `@tangle-network/agent-eval@0.135.2` 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
 
@@ -605,7 +605,7 @@ Import from `@tangle-network/agent-runtime/kernel` — 594 exports.
 | `printBenchmarkReport` | function | Pretty-print a report — the "free optimization" verdict, with the cost vector. |
 | `probeSandboxCapabilities` | function | Probe (and memoize per client) what the loop may rely on. A client without a |
 | `profileRichnessFinding` | function | Turn a {@link ProfileRichness} verdict into a bus-routable `AnalystFinding` (area `profile-quality`). |
-| `promotionGate` | function | Statistical promotion decision over a holdout benchmark: a seeded paired bootstrap (`heldoutSignificance`) whose CI lower bound must clear `deltaThreshold`. |
+| `promotionGate` | function | Statistical promotion decision over a holdout benchmark using the outcome-appropriate interval selected by `heldoutSignificance`. |
 | `providerAsExecutor` | function | Adapt an environment provider into an `ExecutorFactory` for `createExecutor`. |
 | `providerAsSandboxClient` | function | Adapt a neutral environment provider to the `SandboxClient` interface used by existing loop paths. |
 | `queueOf` | function | Convenience: a `DispatchUnit` factory over a fixed array of tasks, for the common case where |
@@ -1370,7 +1370,7 @@ Import from `@tangle-network/agent-eval` — 10 exports.
 
 ### STATISTICS — significance, intervals, effect size
 
-Import from `@tangle-network/agent-eval` — 53 exports.
+Import from `@tangle-network/agent-eval` — 61 exports.
 
 | Symbol | Kind | Summary |
 |---|---|---|
@@ -1389,12 +1389,17 @@ Import from `@tangle-network/agent-eval` — 53 exports.
 | `mcnemarPower` | function | Power of a McNemar test at a given number of paired observations, the inverse |
 | `mcnemarRequiredN` | function | Number of paired observations needed for a McNemar test to reach a target |
 | `mulberry32` | function | Tiny seedable PRNG (mulberry32) — deterministic resampling/shuffling, not |
+| `pairedBinaryScale` | function | The common positive level `s` such that EVERY value across both paired arms is |
 | `pairedBootstrap` | function | Paired bootstrap on (after − before) deltas. Returns a CI on the chosen |
 | `pairedCohensDz` | function | Cohen's dz for paired observations: mean(after - before) divided by the |
+| `pairedDecisionShape` | function | Which estimator {@link decidePairedPromotion} would use on this data, and the |
 | `pairedDeltaTest` | function | Tests whether a paired candidate-minus-baseline delta clears a threshold. |
+| `pairedDeltaTieFraction` | function | Fraction of paired observations whose delta is an exact tie (\|after − before\| |
 | `pairedEvalueSequence` | function | Run the paired e-value sequence over an in-order delta stream. |
 | `pairedMde` | function | Minimum detectable paired effect (standardised units) for a target paired |
 | `pairedRiskDifference` | function | Paired risk difference (the effect-size companion to {@link mcnemar}): the |
+| `pairedRiskDifferenceExact` | function | Paired risk difference with the EXACT CONDITIONAL interval — the estimator a |
+| `pairedRiskDifferenceScore` | function | Paired risk difference with TANGO'S (1998) SCORE INTERVAL — the estimator a |
 | `pairedSignTest` | function | Exact one-sided sign test over paired differences. |
 | `pairedTTest` | function | Paired t-test — before/after measurements on the SAME items. |
 | `partialCredit` | function | Partial credit: returns 0-1 ratio of current toward target |
@@ -1408,9 +1413,12 @@ Import from `@tangle-network/agent-eval` — 53 exports.
 | `wilcoxonSignedRank` | function | Wilcoxon signed-rank — paired, no distributional assumption on the deltas. |
 | `wilson` | function | Wilson score interval for a binomial proportion. Correct at small n and near |
 | `normalizeScores` | const | Identity: dimensions already follow "higher = better" by prompt convention |
+| `ExactRiskDifferenceResult` | interface | A paired binary effect size with an EXACT interval and the exact test that |
 | `McNemarResult` | interface | Result of a McNemar paired-binary significance test. |
+| `PairedMcNemarEvidence` | interface | McNemar's exact paired-binary evidence, on the two-point path only. |
 | `ProportionInterval` | interface | A binomial proportion estimate with a confidence interval. |
 | `RiskDifferenceResult` | interface | A paired binary effect size (treatment rate − control rate) with a CI. |
+| `ScoreRiskDifferenceResult` | interface | A paired binary effect size with an interval that is valid at a NONZERO |
 
 **Undocumented supporting types** (add a TSDoc line at the declaration to earn a table row): `BootstrapOptions`, `BootstrapResult`, `ClusterBootstrapInterval`, `CorpusAgreementOptions`, `CorpusAgreementPerDimension`, `CorpusAgreementReport`, `CorpusScoreRecord`, `EProcess`, `EProcessOptions`, `EProcessState`, `EProcessStep`, `PairedBootstrapOptions`, `PairedBootstrapResult`, `WeightedCompositeInput`, `WeightedCompositeResult`, `CliffsMagnitude`.
 
@@ -1455,8 +1463,8 @@ Import from `@tangle-network/agent-eval/campaign` — 329 exports.
 | `fsCampaignStorage` | function | Node-filesystem storage — the default. Lazily requires `node:fs` so the |
 | `gepaOptimizationMethod` | function | Turn an optional GEPA installation into an `OptimizationMethod`. |
 | `gitWorktreeAdapter` | function | Git-backed `WorktreeAdapter`: creates isolated worktrees on fresh branches, commits agent changes, and discards losers. |
-| `heldOutGate` | function | Composable held-out gate: ships only when the PAIRED bootstrap CI lower bound |
-| `heldoutSignificance` | function | Significance of the held-out composite lift: ship only when the paired |
+| `heldOutGate` | function | Composable held-out gate: ships only when the lower bound of the DECIDING |
+| `heldoutSignificance` | function | Significance of the held-out composite lift: ship only when the lower bound |
 | `inMemoryCampaignStorage` | function | In-memory storage for filesystem-less runtimes. Artifacts + trace spans |
 | `isProposedCandidate` | function | Type guard: a proposal carrying its rationale vs a bare |
 | `isTransientTransportFailure` | function | True when the error text describes an infrastructure hiccup that should be |

docs/api/runtime.md

@@ -4923,7 +4923,8 @@ Paired tasks that carried both strategies' cells.
 
 > **lift**: `object`
 
-Paired (candidate − incumbent) lift across the holdout tasks.
+Paired (candidate − incumbent) lift across the holdout tasks. `low` and `high`
+ are the bounds that carried the decision; `mean` and `median` are diagnostics.
 
 ###### mean
 
@@ -4945,8 +4946,8 @@ Paired (candidate − incumbent) lift across the holdout tasks.
 
 > `optional` **costSavings?**: `object`
 
-non-inferiority mode: paired (incumbent − candidate) cost SAVINGS per task (usd) —
- positive means the candidate is cheaper; significant iff the CI low clears zero.
+non-inferiority mode: paired (incumbent − candidate) cost savings per task (usd).
+ Positive means the candidate is cheaper; `low` and `high` carried the decision.
 
 ###### mean
 
@@ -18282,7 +18283,7 @@ readonly [`EqualKArm`](#equalkarm)[]
 
 > **promotionGate**(`opts`): [`PromotionVerdict`](#promotionverdict)
 
-Statistical promotion decision over a holdout benchmark: a seeded paired bootstrap (`heldoutSignificance`) whose CI lower bound must clear `deltaThreshold`.
+Statistical promotion decision over a holdout benchmark using the outcome-appropriate interval selected by `heldoutSignificance`.
 
 #### Parameters
 

docs/canonical-api.md

@@ -4,9 +4,9 @@
 Generated signatures and the complete export list live in docs/api/.
 Run pnpm docs:freshness after editing this file. -->
 
-> **Version 0.109.1.**
+> **Version 0.109.2.**
 > [`docs/api/primitive-catalog.md`](./api/primitive-catalog.md) lists every export and import path.
-> `agent-eval` must satisfy `>=0.135.1 <0.136.0`.
+> `agent-eval` must satisfy `>=0.135.2 <0.136.0`.
 > `sandbox` must satisfy `>=0.15.0 <0.16.0`.
 > Portable profile and tool-part types come from `@tangle-network/agent-interface` `>=0.36.0 <0.37.0`.
 >

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-runtime",
-  "version": "0.109.1",
+  "version": "0.109.2",
   "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": {
@@ -162,7 +162,7 @@
   "license": "MIT",
   "packageManager": "pnpm@11.17.0",
   "peerDependencies": {
-    "@tangle-network/agent-eval": ">=0.135.1 <0.136.0",
+    "@tangle-network/agent-eval": ">=0.135.2 <0.136.0",
     "@tangle-network/agent-interface": ">=0.36.0 <0.37.0",
     "@tangle-network/sandbox": ">=0.15.0 <0.16.0",
     "playwright": "^1.40.0"

pnpm-workspace.yaml

@@ -18,9 +18,9 @@ allowBuilds:
 catalog:
   '@arethetypeswrong/cli': 0.18.5
   '@types/node': 26.1.1
-  '@tangle-network/agent-eval': 0.135.1
+  '@tangle-network/agent-eval': 0.135.2
   '@tangle-network/agent-interface': 0.36.0
-  '@tangle-network/agent-knowledge': 6.1.10
+  '@tangle-network/agent-knowledge': 6.1.11
   '@tangle-network/agent-profile-materialize': 0.9.2
   '@tangle-network/sandbox': 0.15.2
   publint: 0.3.22