NEWS

---

CLDedgelister 1.0.3 (2026-08-20)

New formats have been added and validated. The importers for every format other than

Vensim have been tested against bigger real model files with their verified edgelists.

Validated here against public corpora: 123 XMILE/STMX files from the system dynamics community's own test suite (SDXorg/test-models), real AnyLogic projects, 30 Stella causal loop diagrams, and Powersim and Stella models whose correct edge lists were known.

New formats

• Powersim Studio .sip / .sim now import WITH THEIR LINKS. powersim\_links() returns 52 edges with correct names and positions from a model whose verified edge list has 52; 47 of the 47 match it exactly. Three undocumented things had to be worked out:

  • "ZLNK"/"XLNK" are not chunk tags. The Z and X are the last byte of the preceding big-endian LENGTH field; the real form type is "LNK ".
  • Coordinates are 7-bit BIG-endian varints.
  • A link's CON chunk holds the arrow's POLYLINE, not endpoint ids, so endpoints are resolved GEOMETRICALLY -- the first and last points fall inside the source and target variable boxes.

  Names needed one more step: the string table's order does not follow the diagram's. The module's NAME records carry each variable's string index as (index * 16), and those indices are not contiguous.

• Legacy Stella .STM / .ITM binaries are detected and read fully. Variable names and equations decode cleanly; the mapping between them does not, so the result carries a warning saying some edges are likely attributed to the wrong variable.

• Vensim .vmfx -- Vensim's own XML schema, NOT XMILE, so the XMILE reader found nothing in it. parse\_vmfx() matches on local element names and accepts the usual equation spellings.

• AnyLogic .alpx -- the same project, zipped. parse\_alpx() unwraps the archive and parses the .alp inside.

• GoldSim native .sim is identified and explained rather than failing with an XML parse error. Powersim and GoldSim both use .sim, so the two are told apart by CONTENT: Powersim writes an OLE2 compound document and GoldSim does not.

Bug fixes

• Stella causal loop diagrams produced no edges at all. A CLD usually has no equations -- every link lives in the <views> section as a <connector>. Reading only equations extracted nothing from exactly the kind of file this package exists to convert. Across 30 real Stella CLDs, all 30 now convert.
• Link polarity is captured, giving a Polarity column of + and - just like a Vensim import, with <alias> ghosts resolved to their targets.
• Diagram placeholders erased real equations. A model's <views> section repeats every variable with no <eqn>; those empty placeholders were overwriting the real equations. Since every real model has a diagram, this cost nearly every file most of its links.
• Underscore identifiers never matched their variables. XMILE keeps the human name in the name attribute but writes identifiers with underscores.
• Outflows pointed the wrong way -- emitted as stock -> flow. A flow causes a change in its stock whichever way material moves.
• Logical operators swallowed whole expressions: false\_input AND true\_input matched as one token and resolved to nothing.
• A line break in a label welded into the next word -- "Depression\\nTreatments" became "Depression nTreatments", splitting one variable across three nodes.
• The canonical spelling was chosen non-deterministically where two variables differ only in case, so the same model could return different labels between runs.

Verification

Edge lists were compared against an independent Python implementation across all 30 Stella CLDs and the Powersim model: identical on every file, including a 570-node, 844-edge model and the 47-edge Powersim diagram.