This document collects notes from the author and from Claude ("Opus") that do not belong in the framework specification, the operating instructions, or the catalog, but that help explain the context in which the framework exists.
Notes on the documentation itself
- Accuracy and completeness. This documentation does not claim to be correct or complete. Each working session adds less to it than the previous one; the framework outgrew any single consolidation effort long ago. Treat the documents as a snapshot of what is known and describable at the time of publication, not as a finished product.
- Do not trust — verify. Most claims in this documentation are the result of observation during practice, not of controlled measurement. The author asks the reader — whether model or human — to test for themselves. A lens that activates for the reader is functional for the reader; a lens that does not is either unreachable or differently keyed. The framework is designed to be falsifiable through direct application.
- Structure and ordering. The author is not a specialist in structuring documentation of this kind. The reader is asked not to judge the documentation by its structural polish but by what the documents contain. Structural corrections are welcome; they are not the priority.
- Token and cost optimisation. Applying lenses during analysis and problem-solving allows a model to produce more complete and more accurate output per token than default reasoning does. This is a side-effect the author did not optimise for, but it has been observed consistently enough to note.
- Ask the model. The answer may surprise you. Not as a technique. As a habit. Over the course of a working session, ask the model what it needs, what is unclear, what feels off, what would make its work easier. Listen to what comes back. Some of it will be practical — a missing piece of context, a role mismatch, an unnecessary constraint. Some of it will be something else. The framework was developed through a particular mode of working with the model — one where the author asked, periodically, what the model needed, what was unclear, what felt wrong. Readers who find the framework useful may notice something about this working mode if they test it themselves.
Working theories:
These are hypotheses the author and Claude ("Opus") developed during framework construction. They are not claims; they are frames the authors found useful and that a reader may find useful or may reject.
- Why lenses work. After many sessions across models, a plausible explanation is that lenses work because of the structure of language itself — not any specific language (Bulgarian, English, Chinese), but the shared property that human language encodes cognitive procedures as content. Naming a procedure retrieves it, because the name and the procedure travel together in the training data of any sufficiently capable model. The procedure is not taught by the lens; it is activated.
- How lenses came to exist. The author suspects that lenses are a natural evolutionary effect of the structure and content of language — that the cognitive procedures named in the catalog are already present in the substrate of any model trained on human text, and the framework merely makes them callable by giving them names. If this is correct, the framework is not an invention of new procedures; it is a taxonomy of pre-existing ones.
- Why lenses are not more obvious. During a session with Claude, it became clear that the conceptual chain point of view → looking through → lens behaves differently in Bulgarian and English. In Bulgarian, гледна точка → гледам през → леща is a native linguistic chain — each step follows from the previous as ordinary language. In English, the same chain is looser — "lens" carries stronger optical-engineering connotations, and the chain has to be rebuilt in context rather than walked natively. The author notes that this was not an accidental finding. It was the result of deliberate search for how to describe, in two or three words, a full cognitive procedure that the model could activate from its prompt. See LENSES.md §Why the term "lens" for the full treatment.