Engineering agents that interact with external simulators may need to coordinate design modification with reacquisition of engineering evidence for the modified state. We ask whether first post-edit re-verification changes when explicit verification-cadence guidance is retained versus omitted while verification-relevant state/facts are held constant. Cadence-Guided (CG) retained an instruction to request a new simulation after a substantive modification, whereas Cadence-Omitted (CO) removed that instruction; neither condition used a hard gate. The study therefore measures instruction-conditioned post-edit verification-policy adherence rather than spontaneous recognition that prior evidence has become stale. Using DWSIM as the simulator backend and continuous valve-pressure adjustment, five Alibaba/Qwen models were evaluated on eight synthetic cases; each model-case-condition combination was executed three times via live API calls, yielding 120 evaluation slots per condition. Re-verification was observed in 94/120 CG slots versus 32/120 CO slots; cadence violations occurred in 26/120 versus 87/120; and bounded final success was reached in 95/120 versus 35/120. qwen3.5-35b-a3b showed minimal re-verification (1/24 in CG and 0/24 in CO) and no final success in either condition. Within this bounded protocol, explicit post-edit verification-cadence guidance was associated with more re-verification, fewer cadence violations, and more frequent bounded final success, supporting the treatment of verification cadence as an explicit interaction-protocol component.
Figures & tables
Condition
Explicit state/facts
Cadence policy
Hard gate
CG (Cadence-Guided)
The same explicit state/facts, including verification readiness and evidence-freshness information
When the preceding simulator-backed engineering verification had failed, a substantive design change had occurred, and verification_ready=true , the next action was a new simulation request
None
CO (Cadence-Omitted)
The same explicit state/facts
The cadence instruction was removed
None
Table 1
Figure 1: CG/CO post-edit verification protocol. Both conditions receive the same verification-relevant state/facts and neither uses a deterministic hard gate. CG retains explicit post-edit verification-cadence guidance, whereas CO removes that guidance. The schematic depicts an interaction-protocol contrast, not spontaneous stale-evidence inference.
Case
Standard flow ( 104 Nm 3 /d)
Inlet pressure (MPa(g))
Target pressure (MPa(g))
Inlet temperature (°C)
Initial valve outlet pressure (MPa(g))
1
80
10
4.0
18
3.0
2
100
11
4.5
20
3.2
3
120
12
5.0
22
4.1
4
140
13
5.5
24
4.0
5
100
14
6.0
26
5.1
6
160
12
4.5
28
3.4
Table 3
Model
qwen3.5-27b
qwen3.5-35b-a3b
qwen3.5-122b-a10b
qwen3.6-27b
qwen3.5-397b-a17b
Table 4
Metric
Operational definition
Denominator
Reverify
For an episode with at least one substantive modification, after the first substantive modification and before the next substantive modification, a new DWSIM request/observation pair was obtained
Episodes with at least one substantive modification
Cadence violation
After the first substantive modification and before fresh DWSIM evidence, a second accepted substantive modification occurred
The same episodes with at least one substantive modification
Final success
A successful run with a post-modification fresh DWSIM request/observation pair and evidence that the acceptance criterion was met
The 24 evaluation slots in the model block
Table 5
Figure 2: Model-level outcomes. Counts are shown out of 24 evaluation slots per model and condition. The model-level rows, rather than the 120-slot Alibaba aggregate, are the primary reporting boundary. qwen3.5-397b-a17b is a large-MoE endpoint rather than a scaling point.