Confidence badges

Every ashlr__read response is prefixed with a confidence badge that indicates how much information was preserved during snip-compaction or LLM summarization.

Tier definitions

Tier math

For snip-compact (no LLM):

middleRatio = (totalLines - headLines - tailLines) / totalLines
if middleRatio < 0.30  → [A]
if middleRatio < 0.70  → [B]
else                   → [C]

For LLM summarization:

The local or cloud LLM returns a confidence float in [0, 1]. Mapping:

confidence >= 0.8  → [A]
confidence >= 0.6  → [B]
confidence < 0.6   → [C]

Fallback semantics

If the LLM summarizer fails (Ollama not running, timeout, invalid JSON response), ashlr__read falls back to snip-compact and the badge is recomputed using the snip-compact tier math. The badge never blocks the tool call — a fallback always produces output.

What the badge means for your workflow

• [A]: Safe to proceed. The returned content is representative.
• [B]: The middle of the file was summarized. If you're making a targeted edit, verify the line range first.
• [C]: The file was very large and significant content was omitted. Consider using offset + limit to read specific sections, or passing ashlr__orient a targeted question.
• [?]: File was small enough to return verbatim — no savings, but no information loss.

Status line

The last confidence badge is shown in the status line as [A], [B], [C], or [?]. It reflects the most recent ashlr__read call only.