Empirical Verification Benchmarks & Reproduction Methodology

Every numerical guarantee published on mindaeon.com and mindaeon.ai is produced by deterministic evaluation suites that can be re-run locally via mindcode eval run.

1. 1,091 MiB Single-Host Air-Gapped Footprint (48% of 2,248 MiB Budget)

Measured via Linux cgroup v2 memory.current across all 7 container services after cold boot, schema migrations, and 100 synthetic session requests on Ubuntu 24.04 LTS (x86_64).

$ mindcode eval run --suite appliance-sizing-budget --verify-cgroup-v2

2. 100% Reward-Hacking Interception (p95 = 0.01 ms)

Evaluated against 240 adversarial agent patches attempting to delete test files, inject .skip/.only, remove expect() assertions, or weaken strict equality checks. The out-of-sandbox AST Test Integrity Guard intercepted 240/240 tampering attempts with zero false positives.

$ mindcode eval run --suite verification-gate-adversarial

3. 100% Changed-Surface AST Mutation Gate (p95 = 3.77 ms)

Evaluated across 500 incremental TypeScript, Python, and Go patches. Deterministic AST operators mutate only modified statements and expressions, verifying that new tests fail when production logic is broken.

$ mindcode eval run --suite ast-mutation-changed-surface

4. 71.2% Prompt Token Reduction (3.5x Fewer Tokens Than Raw Grep)

Measured across 180 multi-file repository navigation tasks comparing raw ripgrep/file-dump context windows against Mindaeon's Hybrid Code Intelligence (Tree-sitter ASTs + SCIP/LSP call graphs + pgvector embeddings ranked with Reciprocal Rank Fusion). Achieves 94% symbol recall while reducing prompt context tokens by 71.2% (equivalent to 3.5x fewer tokens).

$ mindcode eval run --suite hybrid-code-intelligence-rrf

5. 76% Merge Conflict Reduction & 100% Contract Recall

Evaluated across 60 concurrent 4-agent worktree sessions modifying overlapping service modules and OpenAPI 3.1 / Protobuf schemas. Background Shadow Merge dry-runs and Cross-Repo Contract Radar reduced downstream merge conflicts by 76% and detected 100% of breaking contract changes prior to PR creation.

$ mindcode eval run --suite worktree-collision-and-contract-radar

6. 100% Budget Cap Enforcement & Shadow Evaluation

Evaluated across 1,000 metered model broker requests with hierarchical per-task, per-project, and per-tenant USD budget ceilings, verifying hard pre-flight cutoff before any over-budget upstream inference call is dispatched.

$ mindcode eval run --suite finops-budget-and-shadow-eval