Deterministic regression

Define the property being tested

Continuous integration builds and tests changes; continuous delivery prepares distribution, and continuous deployment adds release to service. For an agent component, the fast base verifies parsing, contracts and host decisions. Provider downtime must not randomize a business-rule test. Real calls retain their place in identified campaigns with their own prerequisites and outcomes.

Place the double at the model boundary

AgentCiCd.Tests keeps a MAF agent and replaces IChatClient with DeterministicChatClient. Tests choose JSON outputs to exercise a specific branch. This verifies orchestration and policy; the double does not discover a good answer to the user request. An assertion that merely checks for a word in the prompt has limited scope.

Read contracts and effects

Console.Write("Support request > ");
var input = Console.ReadLine();
if (string.IsNullOrWhiteSpace(input)) return;
var result = await SupportTicketAgentFactory.Create(client).HandleAsync(input, stop.Token);
Console.WriteLine(System.Text.Json.JsonSerializer.Serialize(result));
Console.WriteLine("Classification only; no ticket was created. Tests exercise rules, not model quality.");

demos/Demo450_ContractRegression/Program.cs · L7–L12 · 9a895c94534c

Source code · Demo450 · Folder

• demos/Demo450_ContractRegression/Demo450.App.csproj

Demo450 is a live classification console: it asks for input and calls the configured provider. Deterministic regression runs in AgentCiCd.Tests, rather than treating the console as model-free. Policy routes riskScore >=70 to human_review and rejects forbidden actions, out-of-domain scores and blank reasons. A historical creation-test name does not mean a SQLite ticket was written: the console announces Classification only.

A deserializable contract can remain invalid in business terms. Also test missing fields, refusals and errors. For writes, verify permission, exact approval and storage readback in Maf.Backlog.Tests; for resumption, use Maf.Durable.Tests. A judge cannot replace these executable boundaries with a favorable opinion about prose.

Build an interpretable CI result

dotnet build BaseTechnique.slnx -m:1 -c Release

dotnet test tests/AgentCiCd.Tests/AgentCiCd.Tests.csproj -m:1 -c Release --no-build --no-restore

dotnet test tests/Episode17Demo.Tests/Episode17Demo.Tests.csproj -m:1 -c Release --no-build --no-restore

Run from the complete repository root with the pinned SDK and restored locked dependencies. Every build/test retains -m:1. Keep TRX files, exit status and revision; a skipped or interrupted suite has not passed. Contract tests and fixtures remain separate from live campaigns and qualification. Historical Verify-Lot014.ps1 recipes remain archived; these commands target distributed projects.

What application success establishes

Tests establish program behavior under controlled inputs. They do not certify that granite4:3b understands requests, always follows policy or produces faithful answers. Transport tests do not establish graphical WPF acceptance. Actual campaigns, business references and independent human validation remain separate evidence; the latter has not been established.

Move from rules to the model

• Demo450 — Contract Regression

• Demo460 — Candidate Evaluation

• Candidate, judge and qualification