D243 - The work list removed what a run solved, not what anything could name

decided · 2026-08-25 · found by comparing two committed files that both claim to be generated

symbols/wanted.txt says of itself that entries "disappear as the vocabulary grows to explain them". 116 of its 3829 entries were hashes symbols/generated.json could already name, and 15 more were names a guest_module! declares.

The rule was for solved in found { wanted.remove(...) } - only what this run solved. Its comment claimed to also drop "hashes carried over from an earlier run that a later vocabulary finally explained", and could not: a named hash is excluded from every later search, because being named is exactly what excludes it, so it is never solved again, so it is never removed. Once on the list, permanently on the list.

Two changes:

• Service::is_named extracts the condition unnamed_imports already used - both sources, the registry and the symbol database, which is why the fix removed 131 rather than 116.
• wanted_now is a pure function taking that as a predicate, so the rule is testable without a corpus, a database and a search. The bug was invisible while it was one function with the file writing, which is principle 8's shape argument arriving as a bug report rather than as a preference.

The rule is now "cannot be named now", not "was never solved". Those read alike. Pinned by a_hash_something_can_already_name_leaves_the_work_list, confirmed to fail against the old rule before being kept.