Added an introduction and an onboarding chat to the BeyondFacts Lab (#1178)

app/Tests/BeyondFactsLab/BeyondFactsLabStoreThreadTests.cs

@@ -35,7 +35,7 @@ public sealed class BeyondFactsLabStoreThreadTests
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
 
-        var thread = await store.LoadThreadAsync(new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK));
+        var thread = await store.LoadThreadAsync(new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK));
 
         Assert.That(thread, Is.Null);
     }
@@ -45,10 +45,11 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var location = new BflThreadLocation(created.ProjectId, BflStep.KNOWLEDGE);
+        var location = new BflStepThreadLocation(created.ProjectId, BflStep.KNOWLEDGE);
         var thread = WithMessage(BflThreadFactory.Create(location), "Worauf beruht meine Expertise?");
 
-        var changed = await store.SaveThreadAsync(location, thread);
+        await store.SaveThreadAsync(location, thread);
+        var changed = await store.LoadAsync(created.ProjectId);
         var loaded = await this.CreateStore().LoadThreadAsync(location);
 
         using (Assert.EnterMultipleScope())
@@ -57,7 +58,7 @@ public sealed class BeyondFactsLabStoreThreadTests
             Assert.That(loaded!.ChatId, Is.EqualTo(location.ChatId));
             Assert.That(loaded.Blocks, Has.Count.EqualTo(1));
             Assert.That(((ContentText)loaded.Blocks[0].Content!).Text, Is.EqualTo("Worauf beruht meine Expertise?"));
-            Assert.That(changed.ModifiedAtUtc, Is.GreaterThan(created.ModifiedAtUtc), "A conversation is work on the preparation.");
+            Assert.That(changed!.ModifiedAtUtc, Is.GreaterThan(created.ModifiedAtUtc), "A conversation is work on the preparation.");
             Assert.That(File.Exists(Path.Combine(this.RootDirectory, created.ProjectId.ToString("D"), "steps", "02-knowledge", "thread.json")), Is.True);
         }
     }
@@ -67,9 +68,9 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var thread = BflThreadFactory.Create(new BflThreadLocation(created.ProjectId, BflStep.STRATEGY));
+        var thread = BflThreadFactory.Create(new BflStepThreadLocation(created.ProjectId, BflStep.STRATEGY));
 
-        await Assert.ThrowsAsync<ArgumentException>(async () => await store.SaveThreadAsync(new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK), thread));
+        await Assert.ThrowsAsync<ArgumentException>(async () => await store.SaveThreadAsync(new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK), thread));
     }
 
     [Test]
@@ -77,7 +78,7 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var location = new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
+        var location = new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
         await store.DeleteAsync(created.ProjectId);
 
         await Assert.ThrowsAsync<FileNotFoundException>(async () => await store.SaveThreadAsync(location, BflThreadFactory.Create(location)));
@@ -90,8 +91,8 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var strategy = new BflThreadLocation(created.ProjectId, BflStep.STRATEGY);
-        var framework = new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
+        var strategy = new BflStepThreadLocation(created.ProjectId, BflStep.STRATEGY);
+        var framework = new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
         await store.SaveThreadAsync(strategy, BflThreadFactory.Create(strategy));
         var stepsDirectory = Path.Combine(this.RootDirectory, created.ProjectId.ToString("D"), "steps");
         Directory.CreateDirectory(Path.Combine(stepsDirectory, "01-framework"));
@@ -109,7 +110,7 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var location = new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
+        var location = new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
         var path = Path.Combine(this.RootDirectory, created.ProjectId.ToString("D"), "steps", "01-framework", "thread.json");
         Directory.CreateDirectory(Path.GetDirectoryName(path)!);
         await File.WriteAllTextAsync(path, "{ damaged");
@@ -127,7 +128,7 @@ public sealed class BeyondFactsLabStoreThreadTests
     {
         var store = this.CreateStore();
         var created = await store.CreateAsync("Panel");
-        var location = new BflThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
+        var location = new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK);
         await store.SaveThreadAsync(location, WithMessage(BflThreadFactory.Create(location), "Hallo"));
 
         await store.DeleteThreadAsync(location);
@@ -141,8 +142,8 @@ public sealed class BeyondFactsLabStoreThreadTests
         var store = this.CreateStore();
         var first = await store.CreateAsync("Panel");
         var second = await store.CreateAsync("Interview");
-        var firstStrategy = new BflThreadLocation(first.ProjectId, BflStep.STRATEGY);
-        var secondFramework = new BflThreadLocation(second.ProjectId, BflStep.FRAMEWORK);
+        var firstStrategy = new BflStepThreadLocation(first.ProjectId, BflStep.STRATEGY);
+        var secondFramework = new BflStepThreadLocation(second.ProjectId, BflStep.FRAMEWORK);
         await store.SaveThreadAsync(firstStrategy, BflThreadFactory.Create(firstStrategy));
         await store.SaveThreadAsync(secondFramework, BflThreadFactory.Create(secondFramework));
 
@@ -157,10 +158,58 @@ public sealed class BeyondFactsLabStoreThreadTests
             Assert.That(locations[firstStrategy.ChatId], Is.EqualTo(firstStrategy));
             Assert.That(locations[secondFramework.ChatId], Is.EqualTo(secondFramework));
             Assert.That(store.GetThreadWrittenAt(firstStrategy), Is.Not.Null);
-            Assert.That(store.GetThreadWrittenAt(new BflThreadLocation(first.ProjectId, BflStep.FRAMEWORK)), Is.Null);
+            Assert.That(store.GetThreadWrittenAt(new BflStepThreadLocation(first.ProjectId, BflStep.FRAMEWORK)), Is.Null);
         }
     }
 
+    [Test]
+    public async Task TheOnboardingIsStoredApartFromThePreparations()
+    {
+        var store = this.CreateStore();
+        var location = BflOnboardingThreadLocation.INSTANCE;
+
+        await store.SaveThreadAsync(location, WithMessage(BflThreadFactory.Create(location), "Passt das Lab zu meinem Interview?"));
+        var otherStore = this.CreateStore();
+        var loaded = await otherStore.LoadThreadAsync(location);
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(loaded, Is.Not.Null);
+            Assert.That(loaded!.ChatId, Is.EqualTo(BflThreadId.ONBOARDING));
+            Assert.That(((ContentText)loaded.Blocks[0].Content!).Text, Is.EqualTo("Passt das Lab zu meinem Interview?"));
+            Assert.That(File.Exists(Path.Combine(this.RootDirectory, "onboarding", "thread.json")), Is.True);
+            Assert.That(await otherStore.ListProjectsAsync(), Is.Empty, "The folder of the onboarding is no preparation.");
+            Assert.That(otherStore.ReadAllThreadLocations()[BflThreadId.ONBOARDING], Is.EqualTo(location));
+            Assert.That(otherStore.GetThreadWrittenAt(location), Is.Not.Null);
+        }
+    }
+
+    [Test]
+    public async Task StartingTheOnboardingOverDeletesIt()
+    {
+        var store = this.CreateStore();
+        var location = BflOnboardingThreadLocation.INSTANCE;
+        await store.SaveThreadAsync(location, WithMessage(BflThreadFactory.Create(location), "Hallo"));
+
+        await store.DeleteThreadAsync(location);
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(await store.LoadThreadAsync(location), Is.Null);
+            Assert.That(store.ReadAllThreadLocations(), Does.Not.ContainKey(BflThreadId.ONBOARDING));
+        }
+    }
+
+    [Test]
+    public async Task AChatOfAStepIsNotStoredAsTheOnboarding()
+    {
+        var store = this.CreateStore();
+        var created = await store.CreateAsync("Panel");
+        var thread = BflThreadFactory.Create(new BflStepThreadLocation(created.ProjectId, BflStep.FRAMEWORK));
+
+        await Assert.ThrowsAsync<ArgumentException>(async () => await store.SaveThreadAsync(BflOnboardingThreadLocation.INSTANCE, thread));
+    }
+
     private static ChatThread WithMessage(ChatThread thread, string text)
     {
         thread.Blocks.Add(new ContentBlock

app/Tests/BeyondFactsLab/BflModelRequirementsTests.cs

@@ -24,6 +24,19 @@ public sealed class BflModelRequirementsTests
         Assert.That(BflModelRequirements.HasEnoughContext(ContextWindow.Of(defaultTokens)), Is.EqualTo(expected));
     }
 
+    [Test]
+    public void TheOnboardingRunsWithASmallWindow()
+    {
+        var smallWindow = ContextWindow.Of(8_192);
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(BflModelRequirements.HasEnoughContext(BflOnboardingThreadLocation.INSTANCE, smallWindow), Is.True, "The onboarding leaves out the guide.");
+            Assert.That(BflModelRequirements.HasEnoughContext(new BflStepThreadLocation(Guid.NewGuid(), BflStep.FRAMEWORK), smallWindow), Is.False);
+            Assert.That(BflModelRequirements.HasEnoughContext(new BflStepThreadLocation(Guid.NewGuid(), BflStep.FRAMEWORK), ContextWindow.Of(BflModelRequirements.MINIMUM_CONTEXT_TOKENS)), Is.True);
+        }
+    }
+
     [Test]
     public void AWindowWhichOnlyAnOperatorCouldRaiseIsNotEnough()
     {

app/Tests/BeyondFactsLab/BflPromptComposerTests.cs

@@ -67,6 +67,55 @@ public sealed class BflPromptComposerTests
         }
     }
 
+    [Test]
+    public async Task TheOnboardingFillsEveryPlaceholderWithoutTheGuide()
+    {
+        var parts = await BflPromptLibrary.LoadOnboardingPartsAsync(AddressForm.FORMAL);
+        var knowledge = await BflPromptLibrary.ReadAsync(BflPromptLibrary.KNOWLEDGE_PATH);
+
+        var prompt = BflPromptComposer.ComposeOnboarding(parts, "Deutsch");
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(prompt, Does.Not.Contain("{{"));
+            Assert.That(prompt, Does.Contain("Antworte auf Deutsch."));
+            Assert.That(prompt, Does.Not.Contain(knowledge.Trim()[..200]), "The onboarding stays short, so it leaves out the guide.");
+        }
+    }
+
+    [Test]
+    public async Task TheFormOfAddressDecidesTheInstructionOfTheOnboarding()
+    {
+        var formal = BflPromptComposer.ComposeOnboarding(await BflPromptLibrary.LoadOnboardingPartsAsync(AddressForm.FORMAL), "Deutsch");
+        var informal = BflPromptComposer.ComposeOnboarding(await BflPromptLibrary.LoadOnboardingPartsAsync(AddressForm.INFORMAL), "Deutsch");
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(formal, Does.Contain("mit „Sie“"));
+            Assert.That(formal, Does.Not.Contain("mit „du“"));
+            Assert.That(informal, Does.Contain("mit „du“"));
+            Assert.That(informal, Does.Not.Contain("mit „Sie“"));
+        }
+    }
+
+    [Test]
+    public void TheOnboardingIsTheSameOnEverySystem()
+    {
+        var parts = new BflOnboardingPromptParts("Sprache: {{RESPONSE_LANGUAGE}}\r\n{{ADDRESS_FORM}}\r\nEnde", " Anrede\r\n");
+
+        var prompt = BflPromptComposer.ComposeOnboarding(parts, "Deutsch");
+
+        Assert.That(prompt, Is.EqualTo("Sprache: Deutsch\nAnrede\nEnde"));
+    }
+
+    [Test]
+    public void TheOnboardingKnowsNoDetailsOfAPreparation()
+    {
+        var parts = new BflOnboardingPromptParts("{{DETAILS}}", "Anrede");
+
+        Assert.Throws<FormatException>(() => BflPromptComposer.ComposeOnboarding(parts, "Deutsch"));
+    }
+
     [Test]
     public void TheLanguageOfTheAnswersIsNamed()
     {

app/Tests/BeyondFactsLab/BflPromptLibraryTests.cs

@@ -24,6 +24,16 @@ public sealed class BflPromptLibraryTests
     /// </remarks>
     private const int MAX_FIXED_PROMPT_CHARACTERS = 90_000;
 
+    /// <summary>
+    /// How many characters the fixed part of the prompt of the onboarding may have: the prompt and the longest form of
+    /// address.
+    /// </summary>
+    /// <remarks>
+    /// About 4,000 tokens. The onboarding asks for no minimum context window, unlike the steps, so it has to leave room
+    /// for a conversation in small models as well.
+    /// </remarks>
+    private const int MAX_ONBOARDING_PROMPT_CHARACTERS = 16_000;
+
     [Test]
     public async Task EveryPromptComesWithTheApp()
     {
@@ -66,4 +76,30 @@ public sealed class BflPromptLibraryTests
 
         Assert.That(core.Length + knowledge.Length + longestStep + longestAddress, Is.LessThanOrEqualTo(MAX_FIXED_PROMPT_CHARACTERS));
     }
+
+    [Test]
+    public async Task EveryOpeningRequestNamesASectionOfItsPrompt()
+    {
+        var core = (await BflPromptLibrary.ReadAsync(BflPromptLibrary.CORE_PATH)).ReplaceLineEndings("\n");
+        var onboarding = (await BflPromptLibrary.ReadAsync(BflPromptLibrary.ONBOARDING_PATH)).ReplaceLineEndings("\n");
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(BflConversationService.OpeningRequestFor(new BflStepThreadLocation(Guid.NewGuid(), BflStep.FRAMEWORK)), Does.Contain("„Dieser Schritt“"));
+            Assert.That(core, Does.Contain("\n# Dieser Schritt\n"));
+            Assert.That(BflConversationService.OpeningRequestFor(BflOnboardingThreadLocation.INSTANCE), Does.Contain("„Gesprächsführung“"));
+            Assert.That(onboarding, Does.Contain("\n# Gesprächsführung\n"));
+        }
+    }
+
+    [Test]
+    public async Task ThePromptOfTheOnboardingStaysShort()
+    {
+        var onboarding = await BflPromptLibrary.ReadAsync(BflPromptLibrary.ONBOARDING_PATH);
+        var longestAddress = 0;
+        foreach (var addressForm in Enum.GetValues<AddressForm>())
+            longestAddress = Math.Max(longestAddress, (await BflPromptLibrary.ReadAsync(BflPromptLibrary.AddressPath(addressForm))).Length);
+
+        Assert.That(onboarding.Length + longestAddress, Is.LessThanOrEqualTo(MAX_ONBOARDING_PROMPT_CHARACTERS));
+    }
 }

app/Tests/BeyondFactsLab/BflThreadTests.cs

@@ -22,7 +22,7 @@ public sealed class BflThreadTests
         using (Assert.EnterMultipleScope())
         {
             Assert.That(BflThreadId.Of(PROJECT_ID, BflStep.STRATEGY), Is.EqualTo(expected));
-            Assert.That(new BflThreadLocation(PROJECT_ID, BflStep.STRATEGY).ChatId, Is.EqualTo(expected));
+            Assert.That(new BflStepThreadLocation(PROJECT_ID, BflStep.STRATEGY).ChatId, Is.EqualTo(expected));
         }
     }
 
@@ -40,6 +40,46 @@ public sealed class BflThreadTests
         }
     }
 
+    [Test]
+    public void TheIdOfTheOnboardingFollowsFromItsName()
+    {
+        // Computed independently, from SHA-256 over "beyondfacts-lab/thread/onboarding". A stored chat keeps its ID, so
+        // this value never changes:
+        var expected = Guid.Parse("b98423a4-8502-8ddd-8e56-df78ccd610ba");
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(BflThreadId.ONBOARDING, Is.EqualTo(expected));
+            Assert.That(BflOnboardingThreadLocation.INSTANCE.ChatId, Is.EqualTo(expected));
+            Assert.That(BflOnboardingThreadLocation.INSTANCE.Key, Is.EqualTo("onboarding"));
+            Assert.That(BflOnboardingThreadLocation.INSTANCE.ChatName, Is.EqualTo("onboarding"));
+        }
+    }
+
+    [Test]
+    public void TheOnboardingSharesItsIdWithNoStep()
+    {
+        var stepIds = BflStep.All.Select(step => BflThreadId.Of(PROJECT_ID, step));
+
+        Assert.That(stepIds, Has.None.EqualTo(BflThreadId.ONBOARDING));
+    }
+
+    [Test]
+    public void ThePlaceOfAStepServesAsAKey()
+    {
+        BflThreadLocation first = new BflStepThreadLocation(PROJECT_ID, BflStep.STRATEGY);
+        BflThreadLocation second = new BflStepThreadLocation(PROJECT_ID, BflStep.STRATEGY);
+
+        using (Assert.EnterMultipleScope())
+        {
+            Assert.That(first, Is.EqualTo(second));
+            Assert.That(first.GetHashCode(), Is.EqualTo(second.GetHashCode()));
+            Assert.That(first, Is.Not.EqualTo(new BflStepThreadLocation(PROJECT_ID, BflStep.FRAMEWORK)));
+            Assert.That(first.Key, Is.EqualTo($"{PROJECT_ID:D}/05-strategy"));
+            Assert.That(first.ChatName, Is.EqualTo("05-strategy"));
+        }
+    }
+
     [Test]
     public void TheIdIsAVersion8Uuid()
     {
@@ -55,7 +95,7 @@ public sealed class BflThreadTests
     [Test]
     public void ANewChatHasNothingButItsPlace()
     {
-        var location = new BflThreadLocation(PROJECT_ID, BflStep.KNOWLEDGE);
+        var location = new BflStepThreadLocation(PROJECT_ID, BflStep.KNOWLEDGE);
 
         var thread = BflThreadFactory.Create(location);
 
@@ -73,7 +113,7 @@ public sealed class BflThreadTests
     [Test]
     public void AChatIsReadiedAsAChatOfTheLabWithoutToolsOrDataSources()
     {
-        var thread = BflThreadFactory.Create(new BflThreadLocation(PROJECT_ID, BflStep.KNOWLEDGE));
+        var thread = BflThreadFactory.Create(new BflStepThreadLocation(PROJECT_ID, BflStep.KNOWLEDGE));
         thread.DataSourceOptions.DisableDataSources = false;
         thread.RuntimeSelectedToolIds = ["web_search"];
         var provider = new AIStudio.Settings.Provider(1, "provider-1", "Test", LLMProviders.SELF_HOSTED, new Model("model", null));

app/Tests/Tools/ProvenanceCleanupServiceTests.cs

@@ -124,7 +124,7 @@ public sealed class ProvenanceCleanupServiceTests
         // switches it on:
         var labStore = this.CreateLabStore();
         var project = await labStore.CreateAsync("Panel");
-        var location = new BflThreadLocation(project.ProjectId, BflStep.KNOWLEDGE);
+        var location = new BflStepThreadLocation(project.ProjectId, BflStep.KNOWLEDGE);
         var referenced = await this.store.Client.StoreAsync(location.ChatId, ProvenanceKind.WEB_PAGE, "of a page the chat of the lab read", null, TOKEN);
         var unreferenced = await this.store.Client.StoreAsync(location.ChatId, ProvenanceKind.WEB_PAGE, "of an answer the person removed", null, TOKEN);
         var thread = BflThreadFactory.Create(location);
@@ -142,12 +142,33 @@ public sealed class ProvenanceCleanupServiceTests
         Assert.That(entries.Keys, Is.EquivalentTo(new[] { referenced.Value }), "The chat of the lab is read like any other chat.");
     }
 
+    [Test]
+    public async Task WhatTheOnboardingOfTheLabReceivedSurvivesTheCleanup()
+    {
+        var labStore = this.CreateLabStore();
+        var location = BflOnboardingThreadLocation.INSTANCE;
+        var referenced = await this.store.Client.StoreAsync(location.ChatId, ProvenanceKind.WEB_PAGE, "of a page the onboarding read", null, TOKEN);
+        var thread = BflThreadFactory.Create(location);
+        thread.Blocks.Add(new ContentBlock
+        {
+            ContentType = ContentType.TEXT,
+            Role = ChatRole.AI,
+            Content = new ContentText { Text = "Laut der Seite …", SourcePassages = [new SourcePassage { SourceUrl = "https://example.org/", ProvenanceId = referenced }] }
+        });
+        await labStore.SaveThreadAsync(location, thread);
+
+        await this.RunAsync(RestoreStateWith(report: null));
+
+        var entries = await this.store.Client.GetEntriesAsync([referenced!.Value], TOKEN);
+        Assert.That(entries.Keys, Is.EquivalentTo(new[] { referenced.Value }), "The onboarding belongs to no preparation, but it is a chat of the lab all the same.");
+    }
+
     [Test]
     public async Task WhatAChatOfADeletedPreparationReceivedIsRemoved()
     {
         var labStore = this.CreateLabStore();
         var project = await labStore.CreateAsync("Panel");
-        var location = new BflThreadLocation(project.ProjectId, BflStep.KNOWLEDGE);
+        var location = new BflStepThreadLocation(project.ProjectId, BflStep.KNOWLEDGE);
         await labStore.SaveThreadAsync(location, BflThreadFactory.Create(location));
         var entry = await this.store.Client.StoreAsync(location.ChatId, ProvenanceKind.WEB_PAGE, "of a page the chat of the lab read", null, TOKEN);
         await labStore.DeleteAsync(project.ProjectId);