improvement-loop

Type: types/tag-readme.md

How improvement candidates are searched for, evaluated with the possibility of rejection, and retained so that later operation depends on them. Assign this tag to work on these functions, their coordination and failure, or their boundary with direct updates that adopt a change without a separate candidate-admission decision. The establishing note is a proposal-selection improvement loop requires search, evaluation, and operative retention. A recurring task loop alone is insufficient: the subject must be how a change becomes a candidate, is selected, or affects later operation. Theory-builder covers the stated theories guiding the work and criticism of their content; work on how that criticism controls candidate search can carry both tags. A child of self-improving-systems.

The distinction from continual-learning is how changes are selected versus how learning persists outside model weights. Candidate search, rejection, backtracking, and acceptance belong here even when the retained form is unspecified. A durable artifact alone does not establish a proposal-selection loop. Assign both tags when the note develops both the selection process and learning through retained non-weight changes.

Loop structure

• Test-gated reuse of orchestration strategies — candidate strategies pass separate fit and promotion checks before later reuse

• A retained instruction preserves what testing selected — testing chooses a candidate procedure and retention preserves that evaluated choice for reuse

• A proposal-selection improvement loop requires search, evaluation, and operative retention — the three required functions

• An omitted improvement-loop function and a frozen one need different repairs — five systems with frozen functions, and why a direct update lacks a gate without omitting one
• False-positive generation is filtered; false-positive acceptance becomes operative — why errors at acceptance cost more than errors at generation
• Distinct residue classes require distinct functions in a self-improving architecture — different reasons a decision stays untransferred point to different missing functions

Search and search control

• Open-ended improvement must allocate search before decisive evaluation is available — even a proof-gated loop must first choose what to develop
• Lightweight search control allocates further search without licensing adoption — the authority of a search judgment stops short of an operative change
• Backtracking keeps lightweight search control provisional — restoring an earlier state keeps a heuristic branch choice revisable
• A search controller is tested by what it brings to stronger evaluation — judge the controller by the branches it routes on, not as acceptance claims
• A failure explanation becomes search control only when it changes a later branch decision — the operative test for a retained failure explanation
• Natural-language project state may specialize weight-resident search heuristics — intent, theory, and branch history steer heuristics the model already holds
• The 2026-08-30 Commonplace revision used retained theory to guide computational search — an observed loop: computational search, operator selection

Evaluation and selection

• Choosing what to learn requires both validity and learning-value gates — trustworthy to learn from is a different check from worth learning
• Diagnostic richness constrains outer-loop learning quality — selection needs inspectable failure evidence, not only a winner-picking oracle
• Evaluation automation is phase-gated by comprehension — error analysis and judge discrimination come before automated optimization
• Weakly discriminated qualities tend to be underselected — what an oracle cannot tell apart, selection does not enrich
• Oracle accumulation improves selection for later candidates in its maintained domain — a failure kept as a maintained check improves later selection

Related Tags

• self-improving-systems — the parent: evidence-responsive changes to a system's own organization, whether by proposal selection or direct update
• theory-builder — the theories that guide search; shares the notes on theory-guided search
• software-factory — factory revision is one target a loop can retain changes into
• reflection — loops whose retained changes pass through a self-representation
• warranted-autonomy — which evaluation and acceptance decisions a computational actor is warranted to take
• continual-learning — learning through persistent non-weight changes; it can use proposal selection or direct updates

Other tagged notes

• A repeatable operative path keeps a redesign class open to revision - Operationalizes repeatable operative revision for a named redesign class as a causal path through representation, evidence-bearing determination, admission, installation, dependence, and continuity
• Commonplace as a reflective self-improving system - Commonplace witnesses that a human-inclusive KB can be reflectively self-improving on one pathway despite uneven coverage and human-gated design judgment
• Continual learning requires governing behaviour-changing writes, not just storing content - For deployed systems, persistence is insufficient; continual learning must select, validate, authorize, and coordinate behaviour-changing updates across the representational forms a system can change
• Cost-sensitive formalisms for tentative theory search - Exploratory map of backtracking, learning, and complexity models that expose budgets relevant to search guided by tentative theories
• Factory-learning mechanisms should be compared on the same causal job - Compares factory-learning mechanisms on their shared causal job — experience-responsive retention — while separating update mechanisms from the project-theory function needed for open-ended coherent modification
• Gödel machines are a proof-governed case of reflective self-modification - A Gödel machine admits self-rewrites through proof under its current formalization; this restricts admission without establishing how many useful changes are reachable or how reliably they are found
• Holding a program theory means sustaining coherent search under delayed feedback - Holding a program's theory is tested by whether a partial, tentative account of what the program is for keeps modification search, backtracking, and recovery coherent until delayed evidence arrives, not by whether the first change is right
• Improvements outside the admitted formal language need a pre-formal stage somewhere - An improvement whose concepts have no expression in a loop's admitted formal language is reached only through a pre-formal stage, inside the loop or fixed at design time in the choice of language; translation relocates that stage
• Project-theory possession requires comparing new demands with existing organization - For open-ended modification, project-theory possession includes relating a new demand to existing responsibilities before parallel structure becomes the default; an explicit assimilation branch may counter additive coding-agent patches
• Promote Only When Future Value Exceeds Maintenance Cost - Candidate memory should become durable only when future retrieval or activation value exceeds review and maintenance cost
• Retained system-definition artifacts enable persistent deployment-time adaptation - Retaining evaluated changes to behavior-shaping prompts, rules, tools, and tests gives deployed systems a persistent adaptation path outside model-weight updates
• Six reported self-improvement paths expose bounded redesign surfaces within supplied methods - Comparative evidence separates operative redesign, revision of governing machinery, and contributions to later improvement from declared editability across six reported self-improvement paths including HyperAgents
• System use is an initial selection environment when theory fit lacks a fixed oracle - When no complete fixed oracle decides whether a claim belongs in a working theory, distributed consequences of live system use can provide an initial selection environment
• The self-improving-system definition classifies its boundary cases without ad hoc exceptions - Ten boundary cases run against the self-improving-system definition — each classifies by the stated criteria alone; the stress they apply falls on boundary declaration, not on the membership clauses
• Use Trace Extraction As Meta-Learning - Trace extraction is an after-the-fact learning path that must respect signal quality, review, and readable-artifact versus distributed-parametric learning boundaries