ltm: model_detected_loops has no stateless gate -- stateless-but-cyclic shapes report unscoreable loops and silently drop pins

[bpowers]

Problem

model_detected_loops (src/simlin-engine/src/db/analysis.rs, ~line 2082; post-#746 it reuses the scored loop builder) has no stateless gate. For stateless-but-cyclic model shapes it still REPORTS enumerated loops, while the scored surface (model_ltm_variables, src/simlin-engine/src/db/ltm/mod.rs) bails via its stateless early return and emits zero LTM variables -- so the reported loops are detected-but-never-scorable. And because the bail happens before pin processing, a SetLoopName pin over one of those (reported!) cycles is silently dropped with no diagnostic -- the #770 silence class, except here the user can see the loop in the detected surface, making the silence more surprising.

#749 fixed the parent-level-PREVIOUS instance of the surface disagreement. The remaining stateless-but-cyclic shapes post-#749:

(a) Instantaneous cycles through stockless PASSTHROUGH modules. The parent-level cycle is broken at the module boundary, so it is not a compile-time CircularDependency -- but there is no state anywhere in the model.

(b) MODULE-INTERNAL-PREVIOUS cycles. Deliberately uncounted by model_is_stateless so the detected and scored surfaces agree on what counts as state (see the sibling issue on module-internal PREVIOUS, filed in the same #749/#750 batch).

For both shapes: model_ltm_variables returns zero vars with no invalid-pin warnings, while model_detected_loops reports the enumerated loops.

Note this also narrows #770's mitigating-scope claim that "a stateless model with a CYCLE is a loud compile-time CircularDependency anyway": shapes (a) and (b) are stateless, cyclic, and compile cleanly, so the silent-pin-drop case is not limited to acyclic models.

Why it matters

• Consistency: phantom structural reporting -- the detected surface presents loops the scoring pipeline will never score, with nothing telling the user why.
• Diagnostics contract: a dropped pin should always produce a named Warning (ltm: pin whose only stock is module-internal is wrongly rejected as 'contains no stock' #673's result.invalid machinery, ltm: pin on a genuinely stateless model bypasses validation -- typo'd pin gets no invalid-pin warning #770's framing); here it produces silence on a loop the tool itself reported.
• Severity low: stateless-but-cyclic models are rare; no wrong numbers.

Components affected

• src/simlin-engine/src/db/analysis.rs (model_detected_loops, no stateless early-out)
• src/simlin-engine/src/db/ltm/mod.rs (model_ltm_variables stateless gate / model_is_stateless)
• src/simlin-engine/src/db/ltm/pinned.rs (model_pinned_loops, unreachable on this path)

Possible approaches

1. Give model_detected_loops the same model_is_stateless gate, returning pins-only or empty, mirroring the scored surface's semantics (the two surfaces then agree that stateless models have no scoreable feedback); or
2. Keep the detected output but emit the ltm: pin on a genuinely stateless model bypasses validation -- typo'd pin gets no invalid-pin warning #770-style "pins ignored: model has no state" Warning from a surface that CAN accumulate diagnostics, so the asymmetry is at least loud.

Either way the resolution should be coordinated with #770 (same warning text/mechanism) so stateless pin handling is uniform.

Cross-references

• ltm: pin on a genuinely stateless model bypasses validation -- typo'd pin gets no invalid-pin warning #770 -- pin on a stateless model bypasses validation (the warning-silence sibling; this issue adds the detected-surface asymmetry and the cyclic shapes)
• ltm: stockless PREVIOUS-lagged cycle is scored by the enumerator but rejected by pin validation (false 'no stock' rationale) #749 -- the parent-PREVIOUS instance of the surface disagreement (fixed in this batch)
• ltm: arrayed-cycle detected/scored loop-id namespace overlap lets the runtime id join read another loop's series #746 -- moved model_detected_loops onto the scored loop builder (no stateless gate added)
• Part of epic epic: Loops that Matter (LTM) -- tracking known issues and improvements #488 (Core algorithm group)

How discovered

Identified as a residual during GH #749/#750 work on branch ltm-fix-batch-2 (commits 69e84d8e/03a9ec1f), 2026-06.

Absorbed #770

#770 (pin on a genuinely stateless model bypasses validation -- typo'd pin gets no invalid-pin warning) was closed as consolidated into this issue: both need the same warning mechanism on the same gate, and this issue's fix (a stateless gate or warning shared across the detected/pin surfaces) delivers #770's typo'd-pin warning as a side effect. The adjacent cleanup from #770's body also rides here: compile_ltm_implicit_var_fragment's dead _dep_graph parameter (db/ltm/compile.rs ~1683), which both callers compute and pass for nothing.