Files
correx/testing/deterministic/src/test/kotlin/RouterContextBuilderTest.kt
T
kami e6084dcbda feat(router): proper grounded system prompts for chat and narration
Replace the placeholder "You are a routing assistant" one-liner with two
purpose-written prompts: a conversational prompt that frames the router as
correx's operator-facing interface to the event-sourced engine (grounded in
workflow state, no fabrication, concise, steering-aware), and a dedicated
narrator prompt for buildNarrationContext that asks for one or two present-tense
sentences naming the stage and outcome.

The longer protected frame shifted token budgets, so the two L2 eviction tests
now size their budget from the measured protected-frame + per-entry cost instead
of hard-coded magic numbers, making them robust to prompt length.
2026-06-03 22:44:32 +04:00

977 lines
45 KiB
Kotlin

import com.correx.core.context.model.ContextLayer
import com.correx.core.context.model.ContextPack
import com.correx.core.context.model.EntryRole
import com.correx.core.context.model.TokenBudget
import com.correx.core.events.events.L3RetrievedHit
import com.correx.core.events.types.SessionId
import com.correx.core.events.types.StageId
import com.correx.core.router.DefaultRouterContextBuilder
import com.correx.core.router.model.RouterConfig
import com.correx.core.router.model.RouterL2Entry
import com.correx.core.router.model.RouterState
import com.correx.core.router.model.RouterTurn
import com.correx.core.router.model.StageOutcomeKind
import com.correx.core.router.model.TurnRole
import com.correx.core.router.model.WorkflowStatus
import kotlinx.coroutines.runBlocking
import kotlinx.datetime.Clock
import kotlinx.datetime.Instant
import org.junit.jupiter.api.Assertions.assertEquals
import org.junit.jupiter.api.Assertions.assertFalse
import org.junit.jupiter.api.Assertions.assertNotNull
import org.junit.jupiter.api.Assertions.assertTrue
import org.junit.jupiter.api.Test
class RouterContextBuilderTest {
private val config = RouterConfig(conversationKeepLast = 3, tokenBudget = TokenBudget(limit = 200))
private val builder = DefaultRouterContextBuilder(config)
private fun buildPack(state: RouterState, budget: TokenBudget): ContextPack = runBlocking {
builder.build(state, budget)
}
private val sessionId = SessionId("test-session")
private val stageId = StageId("stage-1")
private val clock = Clock.System
// --------------------------------------------------------------------------
// Budget enforcement
// --------------------------------------------------------------------------
@Test
fun `build fits within budget when all entries are small`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "hi", clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
assertEquals(10000, pack.budgetLimit)
assertTrue(pack.budgetUsed <= pack.budgetLimit)
}
@Test
fun `build drops entries when budget is exceeded`() {
// Build a state where conversation + L2 entries together exceed the tight budget
val longContent = "x".repeat(800) // ~400 tokens each
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, longContent, clock.now()),
RouterTurn(TurnRole.ROUTER, longContent, clock.now()),
RouterTurn(TurnRole.USER, longContent, clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 100))
// L0 entries (system prompt + workflow status) always fit; L1/L2 should be dropped
assertTrue(pack.compressionMetadata.entriesDropped > 0)
}
@Test
fun `build drops entries oldest-first for L2 memory when budget forces partial drop`() {
// Each L2 entry: "Stage sN (SUCCESS): " + 200-char summary ≈ 55 tokens each.
// Use a config with conversationKeepLast=0 so only L0 + L2 compete for budget.
val configNoConv = RouterConfig(conversationKeepLast = 0, tokenBudget = TokenBudget(limit = 10000))
val builderNoConv = DefaultRouterContextBuilder(configNoConv)
val summary = "x".repeat(200) // ~50 tokens; with prefix "Stage sN (SUCCESS): " ~55 tokens total
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
l2Memory = listOf(
RouterL2Entry(StageId("s1"), summary, StageOutcomeKind.SUCCESS, Instant.parse("2026-01-01T00:00:00Z")),
RouterL2Entry(StageId("s2"), summary, StageOutcomeKind.SUCCESS, Instant.parse("2026-01-02T00:00:00Z")),
RouterL2Entry(StageId("s3"), summary, StageOutcomeKind.SUCCESS, Instant.parse("2026-01-03T00:00:00Z")),
),
)
// Measure the protected frame (L0) and per-entry cost, then size the budget to fit
// exactly two of three L2 entries — robust to system-prompt length changes.
val measured = runBlocking { builderNoConv.build(state, TokenBudget(limit = 10000)) }
val l0Cost = measured.layers[ContextLayer.L0].orEmpty().sumOf { it.tokenEstimate }
val entryCost = measured.layers[ContextLayer.L2].orEmpty().first().tokenEstimate
val pack = runBlocking {
builderNoConv.build(state, TokenBudget(limit = l0Cost + entryCost * 2 + entryCost / 2))
}
val l2Entries = pack.layers[ContextLayer.L2] ?: emptyList()
val retainedIds = l2Entries.map { it.sourceId }.toSet()
// Newest two survive; oldest is dropped
assertEquals(2, l2Entries.size, "Only 2 of 3 L2 entries should fit")
assertFalse(retainedIds.contains("s1"), "Oldest L2 entry (s1) should be dropped first")
assertTrue(retainedIds.contains("s2"), "Second-newest L2 entry (s2) should be retained")
assertTrue(retainedIds.contains("s3"), "Newest L2 entry (s3) should be retained")
assertEquals(1, pack.compressionMetadata.entriesDropped)
}
@Test
fun `build drops conversation turns oldest-first when conversationKeepLast exceeded`() {
val configWide = RouterConfig(conversationKeepLast = 2, tokenBudget = TokenBudget(limit = 10000))
val builderWide = DefaultRouterContextBuilder(configWide)
val longContent = "x".repeat(400)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, longContent, Instant.parse("2026-01-01T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, longContent, Instant.parse("2026-01-02T00:00:00Z")),
RouterTurn(TurnRole.USER, longContent, Instant.parse("2026-01-03T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, longContent, Instant.parse("2026-01-04T00:00:00Z")),
),
)
// Budget allows L0 + 2 long conversation turns; oldest 2 turns are dropped
val pack = runBlocking { builderWide.build(state, TokenBudget(limit = 500)) }
val l1Entries = pack.layers[ContextLayer.L1] ?: emptyList()
// Only the last 2 turns (oldest-first drop) should remain
assertEquals(2, l1Entries.size)
}
// --------------------------------------------------------------------------
// L0 immutability
// --------------------------------------------------------------------------
@Test
fun `L0 system prompt is always present`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.IDLE,
currentStageId = null,
)
val pack = buildPack(state, TokenBudget(limit = 0))
val l0Entries = pack.layers[ContextLayer.L0] ?: emptyList()
assertTrue(l0Entries.any { it.sourceType == "systemPrompt" })
}
@Test
fun `L0 workflow status is always present`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.FAILED,
currentStageId = StageId("failed-stage"),
)
val pack = buildPack(state, TokenBudget(limit = 0))
val l0Entries = pack.layers[ContextLayer.L0] ?: emptyList()
assertTrue(l0Entries.any { it.sourceType == "workflowStatus" })
val workflowEntry = l0Entries.find { it.sourceType == "workflowStatus" }
assertNotNull(workflowEntry)
assertTrue(workflowEntry!!.content.contains("failed"))
}
@Test
fun `L0 entries survive zero budget`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.COMPLETED,
currentStageId = null,
conversationHistory = emptyList(),
l2Memory = emptyList(),
)
val pack = buildPack(state, TokenBudget(limit = 0))
val l0Entries = pack.layers[ContextLayer.L0] ?: emptyList()
assertEquals(2, l0Entries.size) // systemPrompt + workflowStatus
}
@Test
fun `L0 entries survive budget that cannot cover L1 or L2`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "x".repeat(500), clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "x".repeat(500), StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 50))
val l0Entries = pack.layers[ContextLayer.L0] ?: emptyList()
val l2Entries = pack.layers[ContextLayer.L2] ?: emptyList()
assertEquals(2, l0Entries.size)
// The current (last) user turn is always protected and present even at budget=50
assertEquals(1, (pack.layers[ContextLayer.L1] ?: emptyList()).size)
assertEquals(0, l2Entries.size)
assertTrue(pack.compressionMetadata.entriesDropped > 0)
}
// --------------------------------------------------------------------------
// L2 oldest-first eviction
// --------------------------------------------------------------------------
@Test
fun `L2 entries are evicted oldest-first and retained in chronological order`() {
// Each L2 entry: "Stage sN (SUCCESS): " + 100-char summary ≈ 30 tokens each.
// L0 ~30 tokens; budget 90 leaves ~60. Two entries (~30 each = ~60) fit; oldest dropped.
val summary = "x".repeat(100) // ~25 tokens; with prefix ~30 tokens total
val configTight = RouterConfig(conversationKeepLast = 0, tokenBudget = TokenBudget(limit = 10000))
val builderTight = DefaultRouterContextBuilder(configTight)
val baseTime = Instant.parse("2026-01-01T00:00:00Z")
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
l2Memory = listOf(
RouterL2Entry(StageId("s1"), summary, StageOutcomeKind.SUCCESS, baseTime),
RouterL2Entry(StageId("s2"), summary, StageOutcomeKind.SUCCESS, Instant.parse("2026-01-01T01:00:00Z")),
RouterL2Entry(StageId("s3"), summary, StageOutcomeKind.SUCCESS, Instant.parse("2026-01-01T02:00:00Z")),
),
)
// Size the budget from the measured protected frame + per-entry cost to fit exactly two
// of three entries; the third (oldest, s1) is dropped. Robust to system-prompt length.
val measured = runBlocking { builderTight.build(state, TokenBudget(limit = 10000)) }
val l0Cost = measured.layers[ContextLayer.L0].orEmpty().sumOf { it.tokenEstimate }
val entryCost = measured.layers[ContextLayer.L2].orEmpty().first().tokenEstimate
val pack = runBlocking {
builderTight.build(state, TokenBudget(limit = l0Cost + entryCost * 2 + entryCost / 2))
}
val l2Entries = pack.layers[ContextLayer.L2] ?: emptyList()
val retainedIds = l2Entries.map { it.sourceId }
// Oldest entry (s1) should be dropped; s2 and s3 retained in chronological order
assertFalse(retainedIds.contains("s1"), "Oldest L2 entry (s1) should be dropped first")
assertTrue(retainedIds.contains("s2"), "Second L2 entry (s2) should be retained")
assertTrue(retainedIds.contains("s3"), "Newest L2 entry (s3) should be retained")
// Chronological ordering preserved
if (retainedIds.size == 2) {
assertEquals("s2", retainedIds[0], "s2 should appear before s3 in chronological order")
assertEquals("s3", retainedIds[1], "s3 should appear after s2 in chronological order")
}
}
@Test
fun `L2 entries fit within budget are all retained`() {
val config = RouterConfig(conversationKeepLast = 0, tokenBudget = TokenBudget(limit = 10000))
val builder = DefaultRouterContextBuilder(config)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
RouterL2Entry(StageId("s2"), "summary", StageOutcomeKind.FAILURE, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l2Entries = pack.layers[ContextLayer.L2] ?: emptyList()
assertEquals(2, l2Entries.size)
assertEquals(0, pack.compressionMetadata.entriesDropped)
}
// --------------------------------------------------------------------------
// Layer classification
// --------------------------------------------------------------------------
@Test
fun `system prompt is classified as L0`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.IDLE,
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l0Entries = pack.layers[ContextLayer.L0]
assertNotNull(l0Entries)
assertTrue(l0Entries!!.any { it.sourceType == "systemPrompt" && it.layer == ContextLayer.L0 })
}
@Test
fun `workflow status is classified as L0`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l0Entries = pack.layers[ContextLayer.L0]
assertTrue(l0Entries!!.any { it.sourceType == "workflowStatus" && it.layer == ContextLayer.L0 })
}
@Test
fun `conversation turns are classified as L1`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "hello", clock.now()),
RouterTurn(TurnRole.ROUTER, "hi", clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l1Entries = pack.layers[ContextLayer.L1]
assertNotNull(l1Entries)
assertEquals(2, l1Entries!!.size)
assertTrue(l1Entries.all { it.layer == ContextLayer.L1 })
assertTrue(l1Entries.all { it.sourceType == "conversation" })
}
@Test
fun `l2 memory entries are classified as L2`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l2Entries = pack.layers[ContextLayer.L2]
assertNotNull(l2Entries)
assertEquals(1, l2Entries!!.size)
assertTrue(l2Entries.all { it.layer == ContextLayer.L2 })
assertTrue(l2Entries.all { it.sourceType == "stageSummary" })
}
@Test
fun `no entries other than L0 L1 L2 appear in pack`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "hello", clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val presentLayers = pack.layers.keys
assertTrue(presentLayers.containsAll(listOf(ContextLayer.L0, ContextLayer.L1, ContextLayer.L2)))
assertFalse(presentLayers.contains(ContextLayer.L3))
assertFalse(presentLayers.contains(ContextLayer.L4))
}
// --------------------------------------------------------------------------
// Conversation history capping (conversationKeepLast)
// --------------------------------------------------------------------------
@Test
fun `conversationHistory respects conversationKeepLast from config`() {
val config = RouterConfig(conversationKeepLast = 2, tokenBudget = TokenBudget(limit = 10000))
val builder = DefaultRouterContextBuilder(config)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "one", Instant.parse("2026-01-01T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "two", Instant.parse("2026-01-02T00:00:00Z")),
RouterTurn(TurnRole.USER, "three", Instant.parse("2026-01-03T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "four", Instant.parse("2026-01-04T00:00:00Z")),
RouterTurn(TurnRole.USER, "five", Instant.parse("2026-01-05T00:00:00Z")),
),
)
val pack = runBlocking { builder.build(state, TokenBudget(limit = 10000)) }
val l1Entries = pack.layers[ContextLayer.L1] ?: emptyList()
// conversationKeepLast=2, so only the last 2 turns should appear
assertEquals(2, l1Entries.size)
}
@Test
fun `conversationHistory takeLast preserves ordering`() {
val config = RouterConfig(conversationKeepLast = 3, tokenBudget = TokenBudget(limit = 10000))
val builder = DefaultRouterContextBuilder(config)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "first", Instant.parse("2026-01-01T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "second", Instant.parse("2026-01-02T00:00:00Z")),
RouterTurn(TurnRole.USER, "third", Instant.parse("2026-01-03T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "fourth", Instant.parse("2026-01-04T00:00:00Z")),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l1Entries = pack.layers[ContextLayer.L1] ?: emptyList()
assertEquals(3, l1Entries.size)
// The last 3: second, third, fourth
assertTrue(l1Entries.any { it.content.contains("second") })
assertTrue(l1Entries.any { it.content.contains("third") })
assertTrue(l1Entries.any { it.content.contains("fourth") })
}
// --------------------------------------------------------------------------
// Empty state
// --------------------------------------------------------------------------
@Test
fun `build with empty state produces L0 only`() {
val state = RouterState()
val pack = buildPack(state, TokenBudget(limit = 10000))
val l0Entries = pack.layers[ContextLayer.L0]
assertNotNull(l0Entries)
assertEquals(2, l0Entries!!.size)
assertTrue(pack.layers[ContextLayer.L1].isNullOrEmpty())
assertTrue(pack.layers[ContextLayer.L2].isNullOrEmpty())
}
@Test
fun `pack contains correct metadata on empty state`() {
val state = RouterState()
val pack = buildPack(state, TokenBudget(limit = 5000))
assertEquals(5000, pack.budgetLimit)
assertEquals(0, pack.compressionMetadata.entriesDropped)
assertEquals(listOf("L0Immutable", "Conversation"), pack.compressionMetadata.appliedStrategies)
assertTrue(pack.compressionMetadata.truncatedLayers.isEmpty())
}
@Test
fun `pack has correct context pack id and session info`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
)
val pack = buildPack(state, TokenBudget(limit = 1000))
assertEquals("test-session-router-pack", pack.id.value)
assertEquals(sessionId, pack.sessionId)
assertEquals(stageId, pack.stageId)
}
// --------------------------------------------------------------------------
// Content formatting
// --------------------------------------------------------------------------
@Test
fun `system prompt has expected content`() {
val state = RouterState()
val pack = buildPack(state, TokenBudget(limit = 10000))
val systemEntry = pack.layers[ContextLayer.L0]?.find { it.sourceType == "systemPrompt" }
assertNotNull(systemEntry)
val content = systemEntry!!.content
assertTrue(content.contains("correx"), "system prompt names the product")
assertTrue(
content.contains("workflow", ignoreCase = true),
"system prompt grounds the router in workflow state",
)
}
@Test
fun `workflow status entry contains status and stage`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.COMPLETED,
currentStageId = null,
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val workflowEntry = pack.layers[ContextLayer.L0]?.find { it.sourceType == "workflowStatus" }
assertNotNull(workflowEntry)
assertTrue(workflowEntry!!.content.contains("Status: Completed"))
assertTrue(workflowEntry.content.contains("Stage: none"))
}
@Test
fun `conversation turn entry is formatted with role prefix`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "user message", clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l1Entries = pack.layers[ContextLayer.L1] ?: emptyList()
assertEquals(1, l1Entries.size)
assertEquals("user message", l1Entries[0].content)
assertEquals(EntryRole.USER, l1Entries[0].role)
}
@Test
fun `l2 entry is formatted with stage outcome and summary`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
l2Memory = listOf(
RouterL2Entry(StageId("stage-x"), "completed with 3 items", StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l2Entries = pack.layers[ContextLayer.L2] ?: emptyList()
assertEquals(1, l2Entries.size)
val entry = l2Entries[0]
assertEquals("stage-x", entry.sourceId)
assertTrue(entry.content.contains("Stage stage-x"))
assertTrue(entry.content.contains("SUCCESS"))
assertTrue(entry.content.contains("completed with 3 items"))
}
// --------------------------------------------------------------------------
// Token budget accounting
// --------------------------------------------------------------------------
@Test
fun `budgetUsed equals sum of tokenEstimate of all retained entries`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "hello world", clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val allEntries = pack.layers.values.flatten()
val computedSum = allEntries.sumOf { it.tokenEstimate }
assertEquals(computedSum, pack.budgetUsed)
}
@Test
fun `compressionMetadata entriesDropped reflects dropped count`() {
val config = RouterConfig(conversationKeepLast = 0, tokenBudget = TokenBudget(limit = 10))
val builder = DefaultRouterContextBuilder(config)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "x".repeat(200), clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "x".repeat(500), StageOutcomeKind.SUCCESS, clock.now()),
RouterL2Entry(StageId("s2"), "x".repeat(500), StageOutcomeKind.SUCCESS, clock.now()),
),
)
val pack = runBlocking { builder.build(state, TokenBudget(limit = 65)) }
// L0 consumes ~30 tokens, leaving 35 — both L2 entries (each ~130 tokens) dropped
// conversationKeepLast=0 but the single user turn is the protected current turn; it fits and is not dropped
assertEquals(2, pack.compressionMetadata.entriesDropped)
}
// --------------------------------------------------------------------------
// Integration: full lifecycle
// --------------------------------------------------------------------------
@Test
fun `build with full state including all layers`() {
val config = RouterConfig(conversationKeepLast = 2, tokenBudget = TokenBudget(limit = 10000))
val builder = DefaultRouterContextBuilder(config)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "first turn", Instant.parse("2026-01-01T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "first reply", Instant.parse("2026-01-02T00:00:00Z")),
RouterTurn(TurnRole.USER, "second turn", Instant.parse("2026-01-03T00:00:00Z")),
RouterTurn(TurnRole.ROUTER, "second reply", Instant.parse("2026-01-04T00:00:00Z")),
),
l2Memory = listOf(
RouterL2Entry(
StageId("s1"),
"stage one done",
StageOutcomeKind.SUCCESS,
Instant.parse("2026-01-01T10:00:00Z"),
),
RouterL2Entry(
StageId("s2"),
"stage two failed",
StageOutcomeKind.FAILURE,
Instant.parse("2026-01-02T10:00:00Z"),
),
),
)
val pack = runBlocking { builder.build(state, TokenBudget(limit = 10000)) }
// L0: system prompt + workflow status
val l0 = pack.layers[ContextLayer.L0]!!
assertEquals(2, l0.size)
assertEquals("systemPrompt", l0[0].sourceType)
assertEquals("workflowStatus", l0[1].sourceType)
// L1: last 2 conversation turns
val l1 = pack.layers[ContextLayer.L1]!!
assertEquals(2, l1.size)
assertEquals("conversation", l1[0].sourceType)
// L2: both stage summaries fit
val l2 = pack.layers[ContextLayer.L2]!!
assertEquals(2, l2.size)
assertEquals("stageSummary", l2[0].sourceType)
// Budget and metadata
assertTrue(pack.budgetUsed > 0)
assertTrue(pack.budgetUsed <= pack.budgetLimit)
assertEquals(0, pack.compressionMetadata.entriesDropped)
}
@Test
fun `build with null sessionId produces unknown pack id`() {
val state = RouterState(
workflowStatus = WorkflowStatus.IDLE,
)
val pack = buildPack(state, TokenBudget(limit = 1000))
assertTrue(pack.id.value.contains("unknown"))
}
// --------------------------------------------------------------------------
// L3 memory injection (slice 2b)
// --------------------------------------------------------------------------
@Test
fun `L3 hits injected into pack ordered by score descending`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "entry-low",
sourceSessionId = SessionId("other"),
sourceTurnId = "turn-low",
text = "low relevance memory",
score = 0.3f,
),
L3RetrievedHit(
entryId = "entry-high",
sourceSessionId = SessionId("other"),
sourceTurnId = "turn-high",
text = "high relevance memory",
score = 0.9f,
),
L3RetrievedHit(
entryId = "entry-mid",
sourceSessionId = SessionId("other"),
sourceTurnId = "turn-mid",
text = "mid relevance memory",
score = 0.6f,
),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l3Entries = pack.layers[ContextLayer.L3]
assertNotNull(l3Entries)
assertEquals(3, l3Entries!!.size)
// Verify ordering: highest score first
assertEquals("entry-high", l3Entries[0].sourceId)
assertEquals("entry-mid", l3Entries[1].sourceId)
assertEquals("entry-low", l3Entries[2].sourceId)
}
@Test
fun `L3 entries carry recalled memory label in content`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "e1",
sourceSessionId = SessionId("other"),
sourceTurnId = "t1",
text = "some past knowledge",
score = 0.8f,
),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val l3Entry = pack.layers[ContextLayer.L3]?.first()
assertNotNull(l3Entry)
assertTrue(l3Entry!!.content.contains("[recalled memory]"))
assertTrue(l3Entry.content.contains("some past knowledge"))
assertEquals(EntryRole.SYSTEM, l3Entry.role)
assertEquals(ContextLayer.L3, l3Entry.layer)
assertEquals("recalledMemory", l3Entry.sourceType)
}
@Test
fun `L3 hits dropped when budget exhausted and counted in entriesDropped`() {
// Budget tight enough for L0 but not L3 entries with long text
val longText = "x".repeat(800) // ~200 tokens
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "big-entry",
sourceSessionId = SessionId("other"),
sourceTurnId = "t1",
text = longText,
score = 0.9f,
),
),
)
// Budget 50: L0 frame ~30 tokens leaves ~20; L3 entry ~200 tokens -> dropped
val pack = buildPack(state, TokenBudget(limit = 50))
assertTrue(pack.layers[ContextLayer.L3].isNullOrEmpty())
assertEquals(1, pack.compressionMetadata.entriesDropped)
assertTrue(pack.compressionMetadata.truncatedLayers.contains(ContextLayer.L3))
}
// --------------------------------------------------------------------------
// Budget priority: protect system + current user turn, drop oldest first
// --------------------------------------------------------------------------
@Test
fun `system prompt and current user turn survive tight budget`() {
val longContent = "x".repeat(800) // ~200 tokens each
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, longContent, Instant.parse("2026-01-01T00:00:00Z"), turnId = "old-turn"),
RouterTurn(TurnRole.ROUTER, longContent, Instant.parse("2026-01-02T00:00:00Z"), turnId = "mid-turn"),
RouterTurn(TurnRole.USER, "current user input", Instant.parse("2026-01-03T00:00:00Z"), turnId = "current-turn"),
),
)
// Budget 60: L0 frame ~30 tokens -> remaining ~30; current user turn ~4 tokens -> protected.
// Older turns (each ~200 tokens) cannot fit -> dropped.
val pack = buildPack(state, TokenBudget(limit = 60))
val l0 = pack.layers[ContextLayer.L0]
assertNotNull(l0)
assertEquals(2, l0!!.size) // system prompt + workflow status always present
val l1 = pack.layers[ContextLayer.L1] ?: emptyList()
// Current user turn ("current user input") must be present
assertTrue(l1.any { it.content == "current user input" })
// Older turns dropped due to budget
assertFalse(l1.any { it.content == longContent })
assertTrue(pack.compressionMetadata.entriesDropped > 0)
}
@Test
fun `oldest conversation turn is dropped before newest when budget is tight`() {
val config3 = RouterConfig(conversationKeepLast = 3, tokenBudget = TokenBudget(limit = 10000))
val builder3 = DefaultRouterContextBuilder(config3)
val longContent = "x".repeat(800) // ~200 tokens each
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "oldest turn $longContent", Instant.parse("2026-01-01T00:00:00Z"), turnId = "t1"),
RouterTurn(TurnRole.ROUTER, "middle turn $longContent", Instant.parse("2026-01-02T00:00:00Z"), turnId = "t2"),
RouterTurn(TurnRole.USER, "newest turn", Instant.parse("2026-01-03T00:00:00Z"), turnId = "t3"),
),
)
// Budget 100: L0 frame ~30 tokens -> remaining ~70; newest turn ~3 tokens -> protected.
// Middle turn (~200+) doesn't fit -> dropped. Oldest turn also dropped.
val pack = runBlocking { builder3.build(state, TokenBudget(limit = 100)) }
val l1 = pack.layers[ContextLayer.L1] ?: emptyList()
// Newest turn must survive
assertTrue(l1.any { it.content == "newest turn" })
// Oldest turn should be dropped (too large)
assertFalse(l1.any { it.content.contains("oldest turn") && it.content.contains(longContent) })
assertTrue(pack.compressionMetadata.entriesDropped > 0)
assertTrue(pack.compressionMetadata.truncatedLayers.contains(ContextLayer.L1))
}
@Test
fun `L3 and L2 dropped before recent conversation when budget is tight`() {
val longContent = "x".repeat(200) // ~50 tokens each
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "current user message", clock.now(), turnId = "curr"),
),
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "l3-entry",
sourceSessionId = SessionId("other"),
sourceTurnId = "l3-turn",
text = longContent,
score = 0.9f,
),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), longContent, StageOutcomeKind.SUCCESS, clock.now()),
),
)
// Budget 80: L0 frame ~30 -> remaining ~50; current user turn ~5 -> protected, remaining ~45
// L3 entry ~50 tokens -> just barely doesn't fit (50 > 45) -> dropped
// L2 entry ~50 tokens -> dropped
val pack = buildPack(state, TokenBudget(limit = 80))
val l1 = pack.layers[ContextLayer.L1] ?: emptyList()
// Current user message must survive
assertTrue(l1.any { it.content == "current user message" })
// L3 and L2 must be absent
assertTrue(pack.layers[ContextLayer.L3].isNullOrEmpty())
assertTrue(pack.layers[ContextLayer.L2].isNullOrEmpty())
assertTrue(pack.compressionMetadata.entriesDropped >= 2)
}
@Test
fun `truncatedLayers populated with affected layers when drops occur`() {
val longText = "x".repeat(800)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, longText, clock.now(), turnId = "old"),
RouterTurn(TurnRole.USER, "current", clock.now(), turnId = "curr"),
),
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "l3e",
sourceSessionId = SessionId("other"),
sourceTurnId = "lt",
text = longText,
score = 0.8f,
),
),
)
val pack = buildPack(state, TokenBudget(limit = 60))
val truncated = pack.compressionMetadata.truncatedLayers
// At least L1 and L3 should be truncated
assertTrue(truncated.isNotEmpty())
}
// --------------------------------------------------------------------------
// N2/B2: protected-frame-overflow — budget smaller than system + current user turn
// --------------------------------------------------------------------------
@Test
fun `budget overflow - system prompt and current user turn present even when they exceed budgetLimit`() {
// Budget so tight that the protected frame alone (system prompt + workflow status) exceeds it.
// The current user turn must still be included; budgetUsed must report the true total;
// appliedStrategies must contain "BudgetExceeded".
val longUserTurn = "x".repeat(200) // ~50 tokens
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, longUserTurn, clock.now(), turnId = "current"),
),
)
// Budget 1: system prompt + workflow status alone exceed this.
val pack = buildPack(state, TokenBudget(limit = 1))
// System prompt must be present
val l0 = pack.layers[ContextLayer.L0] ?: emptyList()
assertTrue(l0.any { it.sourceType == "systemPrompt" }, "System prompt must always be present")
assertTrue(l0.any { it.sourceType == "workflowStatus" }, "Workflow status must always be present")
// Current user turn must be present
val l1 = pack.layers[ContextLayer.L1] ?: emptyList()
assertTrue(l1.any { it.content == longUserTurn }, "Current user turn must be present even when over budget")
// budgetUsed must be the true total (exceeds budgetLimit)
val allEntries = pack.layers.values.flatten()
val trueTotal = allEntries.sumOf { it.tokenEstimate }
assertEquals(trueTotal, pack.budgetUsed, "budgetUsed must report the true token total, not clamped")
assertTrue(pack.budgetUsed > pack.budgetLimit, "budgetUsed must exceed budgetLimit in overflow scenario")
// BudgetExceeded must be signalled in appliedStrategies
assertTrue(
pack.compressionMetadata.appliedStrategies.contains("BudgetExceeded"),
"appliedStrategies must contain 'BudgetExceeded' when frame exceeds budget",
)
}
// --------------------------------------------------------------------------
// S1: conversationKeepLast = 0 must still include the current user turn
// --------------------------------------------------------------------------
@Test
fun `conversationKeepLast zero still includes the current user turn`() {
val configZero = RouterConfig(conversationKeepLast = 0, tokenBudget = TokenBudget(limit = 10000))
val builderZero = DefaultRouterContextBuilder(configZero)
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "old turn 1", Instant.parse("2026-01-01T00:00:00Z"), turnId = "t1"),
RouterTurn(TurnRole.ROUTER, "old reply", Instant.parse("2026-01-02T00:00:00Z"), turnId = "t2"),
RouterTurn(TurnRole.USER, "current user input", Instant.parse("2026-01-03T00:00:00Z"), turnId = "t3"),
),
)
val pack = runBlocking { builderZero.build(state, TokenBudget(limit = 10000)) }
val l1 = pack.layers[ContextLayer.L1] ?: emptyList()
// conversationKeepLast=0 caps the window to 0 turns, but the protected current turn must appear
assertTrue(l1.any { it.content == "current user input" }, "Current user turn must be present even with conversationKeepLast=0")
// The older turns (capped out) must NOT be present
assertFalse(l1.any { it.content == "old turn 1" }, "Older turns must not appear with conversationKeepLast=0")
assertFalse(l1.any { it.content == "old reply" }, "Older turns must not appear with conversationKeepLast=0")
}
// --------------------------------------------------------------------------
// S4: appliedStrategies includes L3Recall when L3 memory is injected
// --------------------------------------------------------------------------
@Test
fun `appliedStrategies includes L3Recall when lastRetrievedMemory is non-empty`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "e1",
sourceSessionId = SessionId("other"),
sourceTurnId = "t1",
text = "recalled fact",
score = 0.7f,
),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
assertTrue(
pack.compressionMetadata.appliedStrategies.contains("L3Recall"),
"appliedStrategies must include 'L3Recall' when L3 memory is injected",
)
}
@Test
fun `appliedStrategies does not include L3Recall when lastRetrievedMemory is empty`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
)
val pack = buildPack(state, TokenBudget(limit = 10000))
assertFalse(
pack.compressionMetadata.appliedStrategies.contains("L3Recall"),
"appliedStrategies must not include 'L3Recall' when no L3 memory is present",
)
}
// --------------------------------------------------------------------------
// N3: ContextPack.layers is sorted by ContextLayer ordinal
// --------------------------------------------------------------------------
@Test
fun `layers map is sorted by ContextLayer ordinal`() {
val state = RouterState(
sessionId = sessionId,
workflowStatus = WorkflowStatus.RUNNING,
currentStageId = stageId,
conversationHistory = listOf(
RouterTurn(TurnRole.USER, "hello", clock.now()),
),
l2Memory = listOf(
RouterL2Entry(StageId("s1"), "summary", StageOutcomeKind.SUCCESS, clock.now()),
),
lastRetrievedMemory = listOf(
L3RetrievedHit(
entryId = "e1",
sourceSessionId = SessionId("other"),
sourceTurnId = "t1",
text = "recalled",
score = 0.5f,
),
),
)
val pack = buildPack(state, TokenBudget(limit = 10000))
val layerKeys = pack.layers.keys.toList()
val ordinals = layerKeys.map { it.ordinal }
assertEquals(ordinals.sorted(), ordinals, "ContextPack.layers must be sorted by ContextLayer ordinal")
}
}