Paid audits and repeated revision
When does the task benefit of splitting or merging a representation justify paying for evidence and revision?
Read Online Or start with the explanation below.
Why this question matters
A distinction useful for one prediction task may be irrelevant or too expensive for another. This companion makes acquisition, memory, auditing and revision prices explicit before deciding to keep or merge an observable feature.
Setup and results
Simultaneous certificates and penalized selection bounds support conservative split/merge rules. A total-variation bound prices merging loss, while revision costs create hysteresis. An exact two-query example has nonmyopic value despite zero one-step information value. In the recurring-task reference example the gain is only 0.0198773312 over twelve tasks, about 0.33129%; seven of fifteen tested mismatches harm the frozen selected policy versus never auditing.
Limits and negative results
Finite-class learning, optimal stopping and Bayesian control are established antecedents. The examples do not prove a general concept-development theorem or reliable transfer without a distributional premise. A confidence gate can delay useful revision and auditing can be redundant or harmful.
Reading guide
Read the observable interface and task-dependent distinction example, then the certificate cost study. For repeated revision, compare the rational policy envelopes with every sensitivity and frozen-policy loss case, not just the favorable reference setting.
Place in the research programme
The link to Hierarchical Concepts is task-dependent choice of distinctions. The self-models family provides related auditing questions, but this note selects representations from supplied data rather than establishing a cognitive development process.
Edition and review status
Brian Theory is the author; this is an AI-assisted research draft. Documented specialist and hostile AI review rounds informed development, but not every round was blind, independently provisioned or cover-to-cover on the final assembly. These checks are not external human peer review or journal acceptance. Targeted human review and historical novelty assessment remain pending.
Sources and reproducibility
Read the complete manuscript, open its page-faithful PDF reader, or download the exact PDF. Exact LaTeX and figure dependencies contain the manuscript inputs and a checksum manifest. This source-only package does not contain the computational verification programs; their public release and independent execution are not certified here. Read the manuscript’s reproducibility sections for the scope of its finite checks.
PDF SHA-256: 5567d28e249b85d571f80fee3895eb99e433f432ce110d4a56c9c0519f05c6b7. Main LaTeX SHA-256: 774642fb4504753a996d59a3273fc6242731d8e67adac12db005d69d123ab4df.
Related manuscript
When to Split and When to Merge Task-driven observation concepts with paid audits and repeated revisionAI-assisted research draft; internal AI checks documented. External human review and novelty assessment remain pending.
Reading scope: A finite representation library, observable histories and declared supervised validation. Transfer needs a quantitative distribution-change bound.
Site reading copy 1 from 13 November 2026 manuscript with localized corrections and approved author credit; exact edition hashes linked below.
Technical material
How this connects
Open in the knowledge map- Phases
- Manuscripts
- Projects
Conceptual and navigational links, not proof dependencies.

