Assess learning claims during ingest

Type: types/instruction.md

Write the Learning Claims (our opinion) section of an ingest report so a later reader can judge what the source changes in our understanding of learning and adaptation: how its mechanism works, what the evidence supports, and whether it strengthens, limits, or challenges Commonplace's current account. Preserve surprising mechanisms and distinctions even when our vocabulary does not yet accommodate them.

Use this instruction within draft-ingest-report.md when the source's subject or mechanism is learning or adaptation, including a conceptual account with no empirical results. Keep the caller's source boundary, output contract, and occasion rules. This branch adds exactly one report section, placed after Connections Found under the ingest-report type contract, with its paired learning_claims: true frontmatter field, and does not authorize further source collection.

Read Theory builder as the current comparison basis for the source's system. When fine-grained localization or selective repair matters, also read Addressable theory. Load these definitions at execution rather than copying their content into this instruction. They govern our use of the terms but remain open to challenge by the source.

Read Learning inside a fixed decomposition inherits its mistakes for the boundary between improvement within an effective update space and evidence for choices fixed outside it.

Write for a reader already acquainted with Commonplace's vocabulary. Use our established terms without reteaching them; explain unfamiliar source terms and consequential differences in meaning.

Procedure:

  1. Establish the source's mechanism on its own terms, before mapping it.
  2. Relate its important ideas to Commonplace concepts. Make the mapping explicit, so the reader can see what the source adds to, supports, or puts in question in the current account. Where a mapping is partial or no concept fits, explain the difference using the nearest relevant distinction without forcing equivalence. Distinguish our interpretation from the source's claims.
  3. Judge the source's system against the theory-builder definition condition by condition: localized content, consumption, content-directed criticism, and iteration. Give each condition its own evidence strength and do not infer one from another. Iteration counts when the result of criticism is kept and shapes the next round, including rounds within one run. Record separately how far results persist (within an episode, a run, across runs, or across problems); persistence is graded, not a condition. Judge learning, meaning improved capacity for future action, as a separate claim: meeting the conditions does not establish it, and failing one does not rule out other improvement. Choose the further analytical distinctions that matter for this source and the KB's goals. Where relevant, the current account distinguishes using a theory from revising it. When they affect the learning judgment, identify the signals and histories available to the learner, the operations it can compose, the mappings its hypothesis class can express, and the representations or partitions fixed outside its effective update space. Separate improvement within that space from evidence for the fixed decomposition. Include only distinctions that affect interpretation or reuse.
  4. Keep each conclusion at the strength of its evidence. A plausible mechanism, an observed improvement, and evidence that the mechanism caused the improvement support different judgments. Missing evidence leaves a question open; it does not establish absence.
  5. Write the section and set learning_claims: true in frontmatter. It is sufficient when it makes the source's contribution, its relation to our concepts, and its evidential limits clear, including any reason to revise our concepts. Include only distinctions that change that judgment. Leave unresolved mappings explicit rather than forcing a classification or repairing the KB's theory during ingest.

Relevant Notes: