Graphify 0.9.67 fixes Windows restart regression, preserves fixed hash seed

Graphify released version 0.9.67 on September 23, fixing a Windows regression introduced in 0.9.66 the previous day. According to the maintainer, the re-execution logic added to set a fixed hash seed prevented `update`, `extract`, `cluster-only`, and `label` from restarting correctly when invoked through console `.exe` launchers. For teams that upgraded to the previous version, this update affects whether graph workflows can resume. [Release notes](https://github.com/Graphify-Labs/graphify/releases/tag/v0.9.67)

The fix retains the `PYTHONHASHSEED` setting while replacing re-execution of `argv[0]` with `python -m graphify`. Python’s documentation explains that a fixed integer seed produces repeatable hash values for types affected by hash randomization, while `-m` locates and executes the specified module. From an engineering perspective, the new entry point lets tasks re-execute without reusing the platform-specific launcher path, while preserving the previous version’s aim of controlling hash variation. [Python command-line documentation](https://docs.python.org/3/using/cmdline.html)

This directly affects teams that integrate code graphs into agent workflows. Graphify extracts code relationships locally from syntax trees, enabling assistants to query paths and dependencies. If an update or extraction command does not complete, downstream consumers need to verify which run produced the artifacts they are reading; the continued presence of an existing graph file alone does not establish that an update succeeded. This deployment check is inferred from the tool’s workflow; the release notes do not confirm any data corruption. [Package and usage workflow](https://pypi.org/project/graphifyy/0.9.67/)

The same release also fixes spurious import cycles when a Python package directory and a module with the same name coexist, and makes absolute package imports resolve to nodes within the scanned scope. When names are ambiguous across directories, it avoids choosing an arbitrary target. These fixes can change the relationship graph, so differences before and after an upgrade do not necessarily mean the source code has changed. [Import resolution changes](https://github.com/Graphify-Labs/graphify/releases/tag/v0.9.67)

PyPI lists a source distribution and a wheel uploaded that same day. The official package name is `graphifyy`. [Release history](https://pypi.org/project/graphifyy/0.9.67/) Windows users can upgrade in an isolated environment, rerun the affected commands, check exit statuses and artifact updates, and then compare results using identical inputs. Validation should cover both the console launcher and the module entry point separately, and record the interpreter version and package source to prevent older installations in other virtual environments from entering the comparison. Teams with scheduled automatic rebuilds should also verify which environment those jobs actually use.

This fact-check confirmed the release records and the maintainer’s description, but did not independently reproduce the various Windows launch methods. A fixed hash seed also does not guarantee identical model-generated content or graphs across versions. Teams still need to validate launch compatibility and output consistency separately.