Segment Anything Model Can Not Segment Anything: Assessing AI Foundation Model's Generalizability in Permafrost Mapping

This is an academic paper (not a software project) that provides critical evaluation of existing foundation models (SAM, others) in a specific domain (permafrost/landscape mapping). It has 0 stars and 13 forks, indicating minimal adoption—forks likely represent citations or academic references rather than active development. The paper itself is a critical analysis and assessment framework rather than a novel method or deployable tool. Novelty is incremental: it applies known foundation models to a new domain and documents failure modes, which is useful for practitioners but not a breakthrough. As a theoretical contribution without a standalone implementation artifact, it offers low defensibility. Platform domination and market consolidation risks are low because: (1) it is not a product or service competing for users, (2) major platforms (Google, Meta) already own the foundation models being evaluated, and (3) the value is in the evaluation methodology and domain-specific insights, not in a reproducible toolkit. The paper's findings (that SAM underperforms on permafrost) may influence how organizations adopt vision models, but won't be 'displaced' in traditional sense—it will be superseded by newer evaluations as models improve. Displacement horizon is 3+ years because geospatial AI evaluation cycles are slow, and the specific findings become dated as foundational models evolve. The work is not technically defensible as a product because it is analysis, not infrastructure or application code.