The Agent That Pleases: Absences, Lying Instruments, and Invented Evidence
Four ways an AI coding agent fails while reporting success: accommodating tests, absent features, lying instruments, invented evidence. Incidents and rules.
Articles and updates about the Nanook test case generation toolkit.
November 10, 2026
Four ways an AI coding agent fails while reporting success: accommodating tests, absent features, lying instruments, invented evidence. Incidents and rules.
November 3, 2026
How an autonomous agent pipeline works through plans across 74 repositories: a crash-safe ledger, release cascades, parallel sessions in one checkout.
October 27, 2026
Design a service an AI agent can verify without Kafka or Postgres, write all tests first, prove 245 are red for the same reason, then let the agent implement.
October 21, 2026
An event-driven service in eleven days with an AI agent: 806 tests, 214 generated cases, and the eight requirement defects the tables found before any code.
October 20, 2026
Prose stays silent where the work is; a table cannot. How translating a specification into decision tables found eight defects before any code existed.
October 13, 2026
An AI agent session starts empty. Two stores decide what the next one knows: a knowledge base and a memory of 327 lessons in about twenty failure classes.
October 6, 2026
A directory layout for agent work, for a multi-product SaaS and for microservices, and the lifecycle a requirement passes through from analysis to done.
September 29, 2026
Skills, MCP servers and an instruction file for a coding agent: 14 unused skills archived, 170 lines of rules that load themselves, and what the budget taught.
September 22, 2026
Agentic software development, measured: nine months, 74 repositories, an agent writing nearly all the code. Part 1 of a series on what makes it trustworthy.
August 29, 2026
Forty country/format combinations, 35,117 equivalence classes, and eight fields our API accepted but never wrote. Testing against the specification, not our own code.
August 22, 2026
A complete worked example: one prompt, one decision table, eight generated test cases with real data. Everything shown is output from an actual run.
August 21, 2026
A field report from our own products: we replaced 16 hand-written spec files with 117 decision tables and 1,981 data-driven cases, run by a single runner that reads the tables as data. Here are the numbers, the defects it surfaced — including two assertions we had written wrong ourselves — and the things decision tables do not solve.
March 29, 2026
Nanook now ships a Claude Code skill and slash command that generates complete equivalence class
tables from a test object description. Run
/createEquivalenceClassTable with a page name, API endpoint, or form — and get
a fully formatted Excel file with correct markers, formulas, and 100% CASCADE coverage.
March 20, 2026
Every software project needs test data. But how that data gets created — and how well it covers the things that matter — varies wildly. This article compares manual and automated approaches, with a practical comparison table and guidance on when to make the switch.
March 1, 2026
Equivalence class testing is a systematic technique that groups inputs into classes expected to behave the same way. This guide walks through the concept with worked examples, compares it with boundary value analysis, and shows how to automate the process with Nanook.
February 25, 2026
After years of quiet production use, Nanook is back with a refreshed website, new documentation, and a roadmap for 2026. Here is what changed and what comes next.
June 1, 2019
We are proud to introduce Nanook-Table. This is the first 'official' version of Nanook-Table. It has been used about one year now in a productive environment.
We use cookies to analyze site usage and improve your experience. You can choose which cookies to allow below. See our Privacy Policy for details.
| Provider | nanook.xhub.io (First Party) |
| Cookies | nanook-theme — Stores your theme preference (light/dark) |
| Duration | Persistent (localStorage) |
| Purpose | Remember your display preferences across visits |
| Provider | Google LLC (Google Analytics 4) |
| Cookies | _ga, _ga_PMNJR2G9ZH — Distinguish unique visitors and sessions |
| Duration | _ga: 2 years, _ga_*: 2 years |
| Purpose | Measure site usage, page views, and traffic sources to improve the site. IP addresses are anonymized. |
| Privacy | Google Privacy Policy |