CBE-IX repeats bounded capability analysis over the updated target. It stops when no further legitimate target-owned expansion remains under the available evidence, or when a distinct blocking condition prevents valid progress.
The base CBE classifications still apply. CBE-IX adds a frontier of unprocessed target-owned questions, an iteration ledger, duplicate detection and an identity check after every material change.
Delegated operations, rejected candidates, semantic duplicates and already exhausted responsibilities stay out. A peer requires its own separately scoped analysis.
Input target signature, frontier before and after, gaps, candidate verdicts, accepted delta, handoffs, unresolved items, material progress and convergence status.
Invariant: expansion must preserve the target's original responsibility. Rewording an old responsibility is not progress.
expand(Tn) = Tn describes a semantic fixed point: another valid pass adds no new target-owned responsibility or unprocessed target-owned SPLIT fragment. Textual equality alone is insufficient.
A hard resource limit is reported as an external limit. An unresolved architectural ownership decision that makes further expansion risk peer takeover is reported as a boundary blocker. Neither is relabeled as successful exhaustion. Unavailable evidence that leaves no further valid expansion path may instead remain explicit residual uncertainty.
The invariant remains: accept documents and produce normalized ingestible records. The existing acceptance, extraction, normalization and submission responsibilities are preserved. I1 replays the accepted snapshot from the CBE ingestion example. The new questions in I2 concern ambiguous parser matches and conflicting classification returns. This synthetic walkthrough assumes supplied format support, parser precedence, document identifiers, an ingestion record set and a new/duplicate submission rule. The Malware Scanner contract supplies clear, flagged or unavailable with a correlation identifier. The pipeline preserves the return and holds submission unless clear. These finite inputs and return rules are teaching assumptions, not a claim of real scanner execution.
| Pass | Target delta | Peer boundary | Next question |
|---|---|---|---|
| I1 | KEEP format detection, parser selection, ingestion-level duplicate detection and safety-gate coordination; KEEP attaching a returned classification or preserving its unresolved state after SPLIT. This is the CBE snapshot. Acceptance, extraction and normalization were reviewed. New/duplicate handling and clear/flagged/unavailable scanner returns are locally exhausted under the supplied fixture rules. Only the named parser and classification follow-up gaps remain. Downstream submission is locally exhausted against the explicit handoffs. | DELEGATE the malware detection algorithm to Malware Scanner, policy decisions to Retention Policy Service and embedding generation to Semantic Indexer. | What if several parsers match? What if the policy service returns conflicting classifications? |
| I2 | KEEP applying supplied parser precedence or reporting an unresolved choice; KEEP preserving conflicting classification returns as an explicit conflict. | No peer logic enters the target. | Run a fresh gap scan and the seven convergence checks. |
| I3 | No semantically new gap or KEEP delta under the supplied example evidence. | Handoff contracts remain explicit. | EXHAUSTED_WITH_HANDOFFS after C1-C7 pass. |
Three passes are the length of this example, not the method's iteration limit. A different target or new evidence can require a different path or reopen the analysis.
Step through a curated trace. Change the case to see why a boundary blocker, a cycle or an execution limit must not produce an exhaustion claim. The cases illustrate decisions; this page does not automatically establish semantic completeness. The initial state is entry 1 in the visible log; trace entries include setup and stop events as well as analysis passes.
Switching cases resets the trace. All transitions are local to this page.
Residual uncertainty describes the exhaustion of available bounded alternatives, not complete knowledge. New authoritative evidence can justify re-entry with the previous ledger preserved.
Start with Capability-Gap and CSR when existence or coverage is unclear. Use SORR to resolve ambiguous ownership. Apply CBE-IX when repeated internal expansion is justified; use CBE for a bounded analysis without the exhaustion requirement.
Dialogue Lifecycle governs transitions and 5PP Variation A governs scoped work and its audit. A CBE-IX terminal state records the analysis outcome. A subsequent implementation still needs its own accepted scope and authority.
Explore the composed solution pattern or test this distinction in the existing Blueprint reader labs.
Derived from the author-provided Capability Boundary Expander - Iterative Exhaustion AI Execution Prompt, updated 24 August 2026. The frontier, ledger, C1-C7 checks and eight terminal names follow that source. The trace cases are synthetic teaching examples.
Source mapping and reproducibility notes record the supplied source snapshot. No model call, capability installation, peer operation or repository change occurs in these labs. A finite curated walkthrough is not a general convergence proof.