LLMS-TLDR 0.2
/llms-tldr.txt is an optional experimental Cybermaps publication for testing deterministic, token-budgeted site briefings.
It is not part of the llms.txt proposal, an IETF standard, an MCP resource, an Agent Skill, a search index, or an AI-generated summary.
Stable order
- Configured pinned resource IDs in publisher order
- Configured content-type order
- Stored modified time descending
- WordPress object ID ascending
Literal extraction
Each candidate contains the title, public URL, content type, valid stored modification time, and a fixed-length literal extract.
Cybermaps strips stored markup but does not execute shortcodes or server-rendered dynamic blocks.
Whole-entry budgeting
The disclosed estimator is:
estimated_tokens = ceil(UTF-8 bytes / 4)
The configured range is 1,000–200,000 approximate tokens; the default is 80,000.
Cybermaps includes a candidate only when the complete entry and updated coverage header fit. An entry is never cut midway by the budgeter. The final header reports selected, eligible, omitted, and partial counts; partial remains zero in version 0.2.
Explicit non-features
The generator does not:
- score quality;
- infer entities;
- cluster topics;
- deduplicate meaning;
- calculate information gain;
- execute a model; or
- claim downstream adoption.
Physical publication in all mode bypasses PHP and cannot be counted by Discovery Analytics.
The full technical rationale is available in the 0.2 draft paper.