The released code explicitly includes the procedures for automatic identification of damaged elements, triggering of the twin substructure refinement, initialization of the refined substructure, and smooth force transition at the global–local interface, thereby ensuring full transparency and practical reproducibility of the proposed non-intrusive framework.

Next, we will briefly introduce the usage descriptions, core processes, and key functions/scripts of the main structure and substructure codes:

Level-1 (MSM driver): Runs the global dynamic analysis, decides when to train/replace, exchanges boundary histories and interface forces with Level-2 over a socket.

Level-2 (aggregator/server): Builds and solves the fine submodel(s), collects reactions at interface nodes, optionally relays boundary data to deeper cores.

Table 1. Consolidated files and roles (Level-1 and Level-2)

File	Role	File	Role
AnalysisOptn_2.tcl	Global transient solve and state machine	LoadPattern_3.tcl	Uniform excitation
addBoudaryInfoToNextLevel.tcl	Push boundary kinematics	Elements.tcl	Element assembly
server.tcl	Command server and controller	GeoTran.tcl	Geometric transforms
model.tcl	Fine model build and setup	Sections.tcl	RC fiber sections, torsion aggregator
correctorLoad.tcl	Apply corrector loads	Materials.tcl	Concrete02 and Steel02
dispLoadg.tcl	Gravity imposed motions	SPConstraint.tcl	Boundary constraints
dispLoadt.tcl	Transient imposed motions (full history)	NodeCoord.tcl	Nodes
dispLoadr.tcl	Refresh imposed motions in replacement	NodeMass.tcl	Nodal mass (template)
forceToLastLevel.tcl	Collect interface reactions	RayleighDamping.tcl	Damping
divideStepAnalysis.tcl	Robust substepping	Recorders (record*.tcl)	Output
TimeSeries.tcl	Earthquake time series

After each analysis step, Level-1 reads the current time and the six-DOF displacement, velocity, and acceleration at its interface nodes, then sends addDisp, addVel, and addAccel in order over sock1 to Level-2, waiting for an acknowledgment after each send.

Level-2 appends these lines to its histories and, when debugMode=0, forwards the same data for node 65 over sock0 and node 2 over sock2 to downstream cores, again with per-message acks.

The receivers turn these time–DOF histories into Path/Linear time series and MultipleSupport prescribed motions, keeping boundary kinematics synchronized across all model levels.

Table 2. Training and replacement flows

Phase	Level-1 (MSM)	Level-2 (TSM)	Data exchanged
Gravity training	Standard static/transient until trigger; continuously pushes kinematics	trainModel: dispLoadg.tcl → static analyze 10; optional downstream “trainModelGravity”; clearSeries	addDisp/addVel/addAccel; trainModelGravity/finish acks
Dynamic training	When toUpdate=1: puts sock1 "trainModel $steps" then switch toUpdate=2	Newmark transient with steps=$data; dispLoadt.tcl; per substep record and forceToLastLevel	finish ack; subForce recorded per step
Replacement (per step)	Compute lambda, request “getTrialForce”; clearSeries; push kinematics; apply corrector; analyze; “replaceModel”	Compute mainForce0/2; request subForce0/2 from sock0/sock2; clearSeries; push kinematics; correctorLoad.tcl; dispLoadr.tcl; analyze; downstream “replaceModel”	subForce line (12 comps); finish acks for clearSeries/replaceModel

Table 3. Interface forces and correctors

Item	Nodes	Source/Use	Details
Upstream reactions (to L1)	L2 nodes 65 and 165	forceToLastLevel.tcl	subForce0 = [nodeReaction 65] (6 DOFs), subForce2 = [nodeReaction 165] (6 DOFs); subForce concatenated (12 comps)
Corrector loads (L2)	Nodes 65 and 2	correctorLoad.tcl	For each DOF i: timeSeries Path with times "0. t t+dt 1000." and values "0 0 ΔF ΔF", ΔF = mainForce − subForce; apply via Plain pattern and load command. DOF mapping modulo 6; tag advancement per 6 DOFs
Imposed motions (L2)	Nodes {65, 165}	dispLoadg/t/r.tcl	MultipleSupport patterns with groundMotion Plain: gravity uses Linear factors; transient uses Path from full histories; replacement refreshes short paths