{
  "schemaVersion": "fewertools.product-tests.v1",
  "checkedOn": "2026-11-15",
  "operator": "Automated agent, not a claim of personal use by Clinton",
  "records": [
    {
      "slug": "kaeso",
      "verdict": "Keep for developers who can operate a self-hosted gateway.",
      "why": "The local installation routed a real HTTP request through a controlled OpenAI-compatible mock, returned streaming output and enforced key-level limits before contacting the upstream. That is a useful, concrete infrastructure workflow.",
      "limits": "Requires deployment, databases and ongoing operation. Our upstream was a synthetic mock, not a paid model provider. Dollar-denominated spend limits, provider failover, ClickHouse request logs, SSO, load capacity and security were not independently validated.",
      "trial": "Installed release 0.6.0 in an isolated local Docker environment with PostgreSQL and Valkey. Tested chat-to-Responses translation, streaming, usage fields, invalid-key rejection, rate limiting, hard token-budget rejection and restart recovery. A hard token budget also refused a request while its counter service was unavailable, without calling the mock provider. Automated tests used synthetic content and no paid model calls on 15 November 2026."
    },
    {
      "slug": "gh-pull",
      "verdict": "Keep as a focused developer dashboard, with a bounded test record.",
      "why": "The dashboard displayed 50 live pull requests from its public GitHub repository and offers useful review, dependency and status views. A controlled fixture preserved a selected merge method across refresh and showed a rejected action without losing the row.",
      "limits": "Needs GitHub authentication and appropriate permissions. No real pull request was merged, edited or commented on in our checks. Hosted GitHub App installation, large private repositories and production reliability were not validated. Three of 40 upstream browser checks failed in our macOS harness; they concern native keyboard selection, a Control-key undo assumption and a font-stack assertion. Direct merge-method selection passed separately; the other harness failures are not resolved product verdicts.",
      "trial": "Tested source commit 99df0de9a763104dd29ba2cb4f1f70a9b7eb8a64 locally. All 177 upstream unit checks and 37 of 40 selected upstream browser checks passed. Added an independent merge-selection and rejected-action check. Used the actual GitHub client through a test-only guard restricting requests to read-only public-repository queries; loaded 50 real PR rows with no browser errors. Cross-origin saved-view writes returned HTTP 403. Automated checks on 15 November 2026; no live GitHub mutations were made."
    },
    {
      "slug": "diagramsmith",
      "verdict": "Keep for quick browser diagrams and straightforward exports.",
      "why": "A custom application-approval flow with a decision, labelled Yes/No branches and a return loop produced readable SVG, PNG and PDF files without an account. A named browser copy survived reload and reopened with the original diagram text.",
      "limits": "The free PDF is an image-based page; vector PDF, .drawio and multi-page features are gated behind Pro. The Pro boundary was observed without starting a trial or paying. Native file-picker workflows, collaboration, very large diagrams and every maker were not tested. Keep a separate export rather than relying only on browser storage.",
      "trial": "Created an original six-node branching flowchart from text, checked its output labels and arrows, exercised add/undo, exported SVG, a 452 \u00d7 567 PNG and a one-page PDF, and visually inspected the PNG and rendered PDF. Saved and reopened a named browser copy after reload. Observed the vector-PDF upgrade gate. Automated live-browser checks on 15 November 2026; no paid features or accounts were used."
    },
    {
      "slug": "tukancad",
      "verdict": "Keep for profile-frame planning, with clear fabrication limits.",
      "why": "The editor created perpendicular 1,200 mm and 600 mm profiles, correctly totalled 1,800 mm and rejected an overlapping placement. A 1,400 mm template frame generated two 1,360 mm crosspieces between 20 mm uprights, matching the requested outside width.",
      "limits": "Treat the model and cut list as planning aids, not engineering approval. The tested STEP export represents simplified outer profile geometry and omits internal slots and hardware; it was structurally inspected, not validated in a second CAD engine. Some English-mode warnings and labels remained in Czech. Structural strength, machining, connector fit and fabrication accuracy were not validated.",
      "trial": "Built a two-axis assembly, checked individual and total cut lengths, exercised undo/redo, saved a .cad-al.json project and reopened it in a clean browser with both lengths and positions intact. Exported a STEP file with the documented simplified geometry. Loaded and resized the Flat frame template to 1,400 mm; verified the four generated profile lengths. Automated live-browser checks on 15 November 2026; no material orders, sign-in or payments were made."
    }
  ],
  "commercialInfluence": "None; no purchase, paid invitation or badge affected retention.",
  "scope": "Bounded workflows with synthetic fixtures and public data; no security certification or production reliability claim."
}
