A CARL report has a lot in it. This guide walks through the three views, how to read each, and how to turn the remediation list into a plan the team can actually execute.

What every report carries (shared across views)

At the top of every report, three facts are present regardless of which view you're in:

• The subject — repository identifier and commit hash assessed.
• The level achieved — a badge like L3Reviewed with a one-sentence interpretation ("Codebase is at Level 2: Tested. Close to Level 3 Reviewed with 3 remaining criteria.").
• The assessor — name, type, and version. For AI-assisted reports, the model identifier is included.

Below the header are four action buttons: Share link, Download PDF, Download JSON, and Re-run assessment.

The pillar radar

Eight axes, one per CARL pillar, each scored in [0, 1]. Full radius = everything passes at the level you're claiming; a short spoke = the pillar is pulling the overall score down.

Read the radar for shape, not precise numbers:

• Balanced large shape. Healthy. Proceed with the remediation list to push everything a bit higher.
• Large shape with one or two short spokes. You have deep pillars and shallow ones. Remediation effort is better spent filling the shallow ones than polishing the deep ones — small changes move the level meter more.
• Generally small shape. You're targeting the wrong level. Drop --min-level by one in CI until the shape grows into it.

The Developer view

Default. Organized by pillar, each pillar collapsible.

For each pillar:

• The pillar score and the count of passes / fails / NA.
• Failing criteria at the top. Each failing criterion has:
  • The criterion ID (e.g., DOC-030) linking back to its catalog entry.
  • A plain-language description of what the criterion checks.
  • A sanitized evidence summary — "No AGENTS.md found at repo root."
  • A copy-to-clipboard remediation action: the exact code, command, or config snippet to apply.

The Developer view is what you use during active remediation. Open the failing criteria, apply the remediation, commit, re-run CARL, repeat. A typical L0 → L1 push takes a focused afternoon if the remediations are followed in order.

Remediation

Remediations are ranked within a pillar by severity (critical > high

medium > low > info) and secondarily by effort (S < M < L < XL). You can target the top of the list with confidence — those are the highest ROI actions for the level you're in.

The Engineering Lead view

Aggregated. Intended for EMs and platform leads tracking the readiness of multiple repositories at once.

• Pillar scores with trend. If a prior assessment exists, each score carries a delta (+0.08, −0.03) and a small sparkline of historical values.
• Top five prioritized remediations. Ranked by impact to level progression — specifically, which criterion being added to the pass pile would cross the 80% gate for the next level.
• Estimated effort to next level. Computed as the sum of effort estimates for the criteria that need to flip. Expressed as a range (e.g., M-L, ~6-10 hours).
• Compare-with. Select another repository you own and see a side-by-side radar.

The Engineering Lead view is the one to screenshot into your quarterly plan.

The Executive view

Top-line. Intended for stakeholders who want to know what CARL level means for their business without needing to understand the detail.

• The CARL level as a large, clear badge.
• A percentile comparison — "40th percentile among open-source Next.js projects at similar age."
• A one-paragraph summary: what the level means, what the primary gaps are, and the estimated investment to reach the next level.
• A short "why CARL" paragraph explaining the framework.

The PDF export foregrounds this view. Two to four pages, designed to stand on its own.

How to turn the report into a plan

The practical sequence:

1. Download the JSON.
2. Filter by result: fail and the next level you want to hit (e.g., L2 Tested).
3. Sort by severity first, then effort.
4. Take the top five. Estimate total effort (the report gives ranges).
5. Open one issue per criterion you intend to fix, with the remediation text copied in. Link them under a "Getting to LN" tracking issue.
6. Work the list. Re-run CARL after every merge. The level progress bar moves in real time.

Sanitization in the evidence you see

The evidence you see in the report is deliberately structural — counts, paths, presence / absence. You will not see raw code. This is enforced by the output schema, not just by convention. If you need the underlying source to understand why a criterion failed, the criterion ID links to the catalog entry, which explains what the check looks for and what a passing state looks like.

See private-repo assessment for the full sanitization guarantee.

Sharing reports

• Share link. Copies a stable carl.wentzel.ai/report/[id] URL. Public reports are world-readable; private-repo reports require your team to authenticate (or an unlisted token).
• Download JSON. Stable schema — safe to archive, diff, or feed into another tool.
• Download PDF. Executive view, suitable for sharing with stakeholders.
• Badge. Once M8 ships, a SVG badge will be available at carl.wentzel.ai/badge/owner/repo.svg for the latest report.

Related

• Criteria catalog — where failing criterion IDs link back to.
• Report format — the underlying structured output.
• CI integration — how to turn the report into a CI gate.