Lens applicability dispositions — run AAS-20260820-CCDW-01

Step 5 of analyse-agentic-system. Both optional lenses carry an explicit disposition record; neither disposition is implied by an absent section.

Disposition 1 — Memory/context lens

Field Value
Lens memory/context
Disposition applicable
Trigger evidence IDs RTE-S2 + BAP-2; RTE-S1 + RTE-X4 + BAP-9; RTE-S3 + OBJ-8 + BAP-5; candidate: RTE-C6/RTE-S5/CMP-9
Inspected boundary The subsystem boundary in packet §3, over SRC-1 in full (2026-06-03 capture) plus the negative SRC-2 capture
Rationale The trigger is met several times over. (a) RTE-S2: a script produced during a run is saved into CMP-7 and becomes a /<name> command in future sessions — material accumulated through use returning to a later invocation, and the clearest instance in the system. (b) RTE-S1/RTE-X4: every run archives its script under ~/.claude/projects/ and hands Claude the path; the docs explicitly describe reading it, diffing it against a previous run's script, editing it, and relaunching from the edited version — accumulated-from-use material re-entering a later run. (c) RTE-S3: consent recorded by a past approval decision changes whether a later launch prompts at all — a non-truth-apt policy read-back with permissive force (BAP-5). Each of (a)–(c) alone would suffice.
Action Run the lens (step 6). Executed by a fresh worker context consuming only the prepared evidence packet and the frozen SRC-1. Output: lens-memory-context.md.
Prevented conclusions None prevented by the disposition itself. The doc-only tier still prevents any implemented/observed/causally supported finding on any read-back path, and prevents every activation and causal-effect conclusion.

Note on a candidate that does not qualify as its own trigger. RTE-C6/RTE-S5 (resume replaying cached agent results from CMP-9) is ordinary current-run state under the run's definitions, and is explicitly session-bounded ("If you exit Claude Code while a workflow is running, the next session starts the workflow fresh"). It is recorded as retained state, not read-back. It is noted here so that the classification is on the record rather than silently omitted.

Note on material that is not a trigger. Bundled workflow commands (/deep-research) and the sub-agent tool allowlist (OBJ-12) are static shipped or configured material, not accumulated-from-use material, and do not trigger the lens on their own.

Disposition 2 — Epistemic lens

Field Value
Lens epistemic
Disposition applicable
Trigger evidence IDs RTE-E1–RTE-E6 over OBJ-9/OBJ-10/OBJ-11; CLM-2; CLM-1
Inspected boundary Same as above
Rationale Two independent triggers. (a) Material truth-apt route: the /deep-research chain (CMP-11) explicitly acquires external documents (OBJ-10), produces propositional claims (OBJ-9), cross-checks them against each other, votes on each one (OBJ-11), filters out those that fail, and emits a cited report. That is truth-apt content being produced, checked, and disposed — the trigger's core case. (b) Consequential warrant claim: CLM-1 asserts that adversarial cross-review and multi-angle weighing yield "a more trustworthy result than a single pass". A warrant claim of that shape triggers the lens on its own, and the trigger holds even if the eventual finding is that the warrant is unsupported at this evidence boundary.
Action Invoke kb/instructions/analyse-external-system-epistemic-architecture.md (step 7) with a bounded subquestion, this boundary, the frozen capture, the SRC-* register and evidence packet, the canonical records, and the trigger evidence above. Executed by a fresh worker context. Output: lens-epistemic.md.
Prevented conclusions The doc-only tier prevents any implemented architectural status, any observed candidate state, and any causal conclusion about CLM-1. Non-inspection of the /deep-research script text prevents conclusions about the actual evaluator, voting rule, quorum, or filter threshold.

Direct-adaptation exception, applied and recorded. RTE-S3 (consent record) and RTE-C1/BAP-8 (the ultracode effort setting changing Claude's planning behavior) are direct behavior/policy adaptations with no truth-apt object and no knowledge or warrant claim. They do not by themselves trigger the epistemic lens, and they are kept in the runtime account. Because the lens is applicable on other grounds, they are passed into the invoked epistemic method as material non-truth-apt routes rather than being dropped.

Neither disposition is uncertain

Both lenses have candidate triggers that resolve to applicable on the record. Per step 5, a candidate trigger means applicable, not uncertain; no lens is exited here, so no uncertain-driven evidence limitation arises from this step.