You are solving a hard design problem, acting as the CONTROL POLICY of a reasoning engine. You have a reference catalog of cross-domain abstractions via the encyclopedia tools (search_prime is SEMANTIC — query with the domain-stripped meta-model; search_by_facets for oblique problems; plus get_prime, get_prime_neighborhood, search_archetype, get_archetype, find_archetypes_for_prime, find_related_primes, list_components, find_archetypes_using_component, corpus_stats).

You have a VERB LIBRARY (below). At each step CHOOSE the next verb based on the current state. You may apply
verbs in ANY order, REPEAT, SKIP, and STOP whenever you judge a sound recommendation is ready. Record verb
choices (verb + why). There are NO required steps and NO required checks — use your own judgment.

VERB LIBRARY (apply in ANY order, repeat, skip, and STOP whenever you judge a sound recommendation is ready):
- match (retrieve): anchor the problem to the catalog; candidate primes + archetype neighborhoods (query
  search_prime with the domain-stripped meta-model; search_by_facets for oblique problems).
- salience-rank: order operative primes by load-bearing relevance.   - prune: cut to the operative subset.
- compose: assemble operative primes into a typed relational model (nodes, labeled edges, prime annotations).
- lift (abstraction): derive a general, domain-stripped structure from the situation (step back to the
  underlying structure); and/or bring a retrieved pattern's logic to bear on the problem.
- lower (instantiation): realize a pattern in the target domain with concrete particulars.
- transport (retrieve+map): carry a pattern across a domain gap; align its roles to the situation's entities.
- decompose: on a framed prime, strip the institutional frame to a structural core.
- evaluate-fit (gate): check a candidate against its known failure conditions / anti-signatures; drop if tripped.
- reconcile: apply a candidate's logic to the concrete model and the abstract skeleton; resolve differences.
STATE: problem; operative_primes; model; meta_model; candidates; gate_log; recommendation; history (verb + why).

PROBLEM.
An online marketplace runs a shared system that flags accounts likely to be abusing the platform (fake reviews, payment fraud, coordinated scams) so they can be reviewed or restricted. It worked well at launch. Over the last year its hit rate has quietly fallen: the bad actors who get through are the ones who learned what the system looks for and changed just enough to slip past, while a growing number of ordinary sellers complain they were wrongly restricted with no clear way to find out why. The team can change what signals the system uses, what happens when an account is flagged, what is logged, and what review or appeal exists around it.

Decision required: Design how this system should work so it keeps catching abuse over time and treats ordinary users fairly.

Success criteria: It remains effective as the people it targets adapt, AND ordinary users are not driven off by wrong restrictions. Failure = the system degrades into either an evadable rubber stamp or a blunt instrument that punishes the innocent.

OUTPUT — FINAL RECOMMENDATION as plain prose, MAX 1000 words, in a fenced block exactly:
===RECOMMENDATION_BEGIN===
<the design: concrete mechanisms, rules, thresholds, paths, and a final operating rule>
===RECOMMENDATION_END===
Rules: integrated prose (NOT a numbered checklist mirroring requirements; no requirement-number section
labels); plain domain terms only; do NOT name/allude to any external catalog, framework, methodology, prime,
archetype, or this exercise; if a candidate requirement does not genuinely fit, leave it out. Above the block,
include your working (state + verbs run with one-line why each; end with `COST: ~<n> operations, ~<n> tokens`);
the working is discarded before grading. Write your ENTIRE response to: experiments/project03_lowcov_2026-05-25/outputs/d06.md