01
AI systems & compute
Model bodies, tokenization, native training, quantization, and selective compute for efficient, plural intelligence.
Research program
Research you can follow by question.
The Hidden Canopy's public research connects AI systems and compute, governance, evaluation, mission assurance, and institutional research into one traceable program.
These releases are research artifacts. Access and licensing remain governed by the settings on each Hugging Face repository.
Research areas
Choose a path through the work.
Start with the question that brought you here. Each area points to a small number of stable pages before the complete publication and release index.
02
Governance & RegOS
Attributable authority, evidence, human review, consequence-time revalidation, and the boundary between interpretation and action.
03
Evaluation & mission assurance
Long-horizon, partial-information evaluation and physical trust research for systems that must remain legible as conditions change.
04
Institutional & social research
Leadership, correctability, collaboration, hierarchy, and the social conditions that shape whether authority remains accountable.
Selected public writing
A short trail before the full index.
These pieces give the program context without asking a first-time visitor to sort every release at once. The complete model, dataset, and article records remain available below.
IDA Leaves Hopper
Porting the IDA native training engine from Hopper-class infrastructure to consumer Blackwell hardware.
Read on Hugging Face →Beyond the Model Checkpoint
Why durable AI systems need institutional governance beyond the model artifact itself.
Read on Hugging Face →The Discomfort of Leadership
Responsibility, correctability, and the distance between authority and consequence.
Read the publication overview →Complete index
Open the full research library when you need it.
The release and article surfaces are intentionally grouped so the page stays readable without hiding the public record.
Model releases and permanent research4 model records · 7 entries
Current model releases
IDA Edge
52.8M parameters · updated June 11, 2026
Local edge intelligence body. A compact paired body for local inference, governed memory, recurrent state, local attention, and cognitive routing.
Read the release overview →IDA AI
1,011.8M parameters · updated June 11, 2026
Larger interpretive intelligence body. A paired body for text generation and governed interpretive work inside the IDA family architecture.
Read the release overview →IDA MoE
Updated June 11, 2026
Mixture-of-experts research body. A release exploring personality council behavior, escalation reserve, governed memory, and cognitive routing.
Read the release overview →IDA Lattice BPE 32K
32K vocabulary · updated June 8, 2026
Tokenizer. The IDA lattice tokenizer supporting the family's causal-language-model training path.
View the tokenizer repository →Permanent research pages and publications
IDA Family
Roles, lineage, governed memory, dissent preservation, and the relationship between the public research bodies.
Read the family overview →IDA Edge, IDA AI, and IDA MoE
Open the individual pages for the compact edge, larger interpretive, and mixture-of-experts research bodies.
Read the release article →The Fallacy of “I'm Not Racist”
A paper on behavior, institutional effects, political alignment, hierarchy, neutrality, tokenism, and selective inclusion.
Read the publication overview →Collaborative Rivalry Needs Rules
How nations can compete without turning preventable catastrophe into a strategic advantage.
Read the publication overview →The Discomfort of Leadership
Responsibility, correctability, and the distance between authority and consequence.
Read the publication overview →Machiavelli: From the Battlefield to the Boardroom
An essay on strategic power, institutional intelligence, dissent, and adapting before failure makes revision unavoidable.
Read the preprint on SSRN →Open World Harness
Controlled, long-horizon evaluation under partial information, with ground truth held by the evaluator.
Public writing indexLive, cached, and curated records
The automatic public-feed sync loads in the background. Curated links remain available if a source is unavailable.
Automatic publication feed
Curated publication links