JuliusBrussee/caveman · error

duplicate provider/model/region row %s/%s@%s

Error message

duplicate provider/model/region row %s/%s@%s

What it means

Thrown by the catalog YAML loader when two entries share the same provider+model+region triple (checked via a seen-map keyed on the NUL-joined triple). The triple is the pricing lookup key, so duplicates would make lookups ambiguous; the load fails instead of letting one row silently shadow another.

Source

Thrown at shared/platform/catalog/catalog.go:299

			if err != nil || capVerified.IsZero() {
				return nil, fmt.Errorf("%s %s/%s: capabilities_verified_at must be RFC3339", label, entry.Provider, entry.Model)
			}
			if capVerified.After(time.Now().UTC().Add(24 * time.Hour)) {
				return nil, fmt.Errorf("%s %s/%s: capabilities_verified_at is in the future", label, entry.Provider, entry.Model)
			}
		}
		if len(entry.Sources) == 0 {
			return nil, fmt.Errorf("%s %s/%s: at least one source is required", label, entry.Provider, entry.Model)
		}
		for _, rawURL := range entry.Sources {
			u, err := url.ParseRequestURI(rawURL)
			if err != nil || u.Scheme != "https" || u.Host == "" {
				return nil, fmt.Errorf("%s %s/%s: invalid HTTPS source %q", label, entry.Provider, entry.Model, rawURL)
			}
		}
		key := entry.Provider + "\x00" + entry.Model + "\x00" + entry.Region
		if _, duplicate := seen[key]; duplicate {
			return nil, fmt.Errorf("duplicate provider/model/region row %s/%s@%s", entry.Provider, entry.Model, entry.Region)
		}
		seen[key] = struct{}{}
	}
	return decoded, nil
}

// catalogCandidates lists the paths to try, in order: an explicit env override,
// the deploy-image and CWD-relative locations, then a walk up from the working
// directory so the catalog resolves when binaries or tests run from subdirs.
// Both repo layouts are tried: the monorepo keeps the catalog under public/,
// the published caveman repo has it at the top level.
func catalogCandidates() []string {
	rels := []string{
		"public/shared/provider-catalog/catalog/current.yaml",
		"shared/provider-catalog/catalog/current.yaml",
	}
	candidates := []string{}
	if p := os.Getenv("CAVE_CATALOG_PATH"); p != "" {

View on GitHub (pinned to 27d5a3981a)

Solutions

1. Search the file for the provider/model/region named in the error and delete or differentiate the duplicate rows.
2. When updating prices, edit the existing row in place (and refresh verified_at) instead of adding a new one.
3. If a provider genuinely has two price books for the same triple, that's unsupported — pick the correct row.

Example fix

# before
- provider: openai
  model: gpt-5.6
  region: us
  pricing: {old}
- provider: openai
  model: gpt-5.6
  region: us
  pricing: {new}

# after — single updated row
- provider: openai
  model: gpt-5.6
  region: us
  pricing: {new}
  verified_at: "2026-08-14T18:30:00Z"

seen := map[string]bool{}
for _, e := range entries {
    key := e.Provider + "\x00" + e.Model + "\x00" + e.Region
    if seen[key] {
        return fmt.Errorf("duplicate row %s/%s@%s", e.Provider, e.Model, e.Region)
    }
    seen[key] = true
}

Fail catalog load at startup; duplicates are a data-authoring error, not something to dedupe silently (the two rows may disagree on price).

When it happens

Trigger: Two YAML list items with identical provider, model, and region (e.g. the same model pasted twice for a price update without deleting the old row, or the same provider+region added by two contributors).

Common situations: Appending an updated price row while leaving the stale row in place; merging catalog branches that both added the same model; a generator emitting one row per region and accidentally repeating a region.

Related errors

• ${label}:${lineNo}: "${text}" is not a finite number
• ${label}: row is missing a string "${field}"
• ${label}: ${key} has no pricing block
• {label}: duplicate identity {'/'.join(identity)}
• %s: provider, model, and region are required

AI-assisted analysis of JuliusBrussee/caveman@27d5a3981a (2026-08-15). Data as JSON: /api/errors/607b2d0d2781c6ec. Report an issue: GitHub.