Skip to main content

AnalyzeProjects Overview

AnalyzeProjects is the index-driven CLI tool that evaluates the entire Aptlantis portfolio and produces the structured JSON this documentation site's data comes from. Its canonical evaluation inventory is ProjectIndex.md — the sole inclusion authority; it will not discover or evaluate anything not explicitly listed there, and it fails fatally on a mismatched project count.

:::info This is the tool behind this site The docs/index.json and docs/processed-summary.md files that power the CTS/DRS/WDS project pages on this site are compiled output from a run of AnalyzeProjects. When you read a project's capabilities, interfaces, or roadmap elsewhere on this site, that data originated here. :::

Who it's for​

A single operator (Herb) who wants a controlled, repeatable evaluation of the whole portfolio — not an open-ended crawler. Every run requires exactly 30 registered projects across a fixed group distribution (12 DRS, 6 CTS, 4 WDS, 8 Standards); anything else is treated as a fatal validation error, not a warning.

The core concept​

  1. Parse and validate ProjectIndex.md — reject anything that doesn't match the required 12/6/4/8 distribution.
  2. For each project, select its <folder>.manifest.toml if present (a manifest is optional, one alternative direct manifest is accepted).
  3. Read the relevant governing standard (DRS, CTS, or WDS) once per group; Standards-group projects use a separate rubric focused on clarity, completeness, and implementability rather than release requirements.
  4. Sample high-value files within configured file-count and character budgets — every transferred block is explicitly labeled complete file or sampled excerpt, so a truncation boundary is never mistaken for a defect in the actual project.
  5. Send the evidence plus governance context to a model (gpt-5-mini-2025-08-07 via OpenAI, or an Ollama-compatible endpoint as fallback) through ModelClient.py.
  6. Validate and normalize the JSON response, write the individual per-project result, then compile the portfolio-wide summary.

Why this shape​

The design is defensive on purpose: ProjectIndex.md as the sole inclusion authority, fatal validation on malformed entries or wrong group totals, and explicit sampling labels all exist to stop the LLM evaluation from silently drifting — either by evaluating the wrong set of projects, or by treating a sampling artifact as a real finding.