Too Broad
"AI Engineering"
Result: shallow everywhere, expert nowhere
Sweet Spot
"Linux Kernel Engineering"
Result: reasons like a veteran engineer
Too Narrow
"Page Cache Eviction"
Result: knows facts, can't reason
Deep expertise across the Linux kernel: 30+ years of commits, mailing list discussions, CVE resolutions, and subsystem evolution.
Good balance: 10 subdomains, 5 interactions, 4 sources.
Page allocation, slab, OOM, cgroups, NUMA, page cache
TCP/IP, netfilter, XDP/eBPF, sk_buff lifecycle
Spinlocks, mutexes, RCU, seqlocks, lockdep
VFS layer, inode/dentry cache, path resolution
Driver model, DMA, interrupt handling
CFS, RT, deadline, CPU affinity
LSM, SELinux, seccomp, capabilities
ext4, btrfs, XFS, journaling
Pipes, shared memory, futexes, io_uring
cpufreq, cpuidle, suspend/resume, thermal
These produce the hardest reasoning problems. The adversarial swarm exploits these to design traps.
Memory allocation under lock: GFP flags and sleeping
Page cache eviction affects VFS read-ahead
Lock contention affects scheduling; priority inversion
Interrupt vs process context locking
VFS interface; fs implementations
30+ years of kernel commits
LKML design discussions
CVE records
Official kernel docs
Drive CoT template generation for training data.
Trace lock acquisition order
Determine lock order, check for deadlock
identify_locks → determine_order → check_inversions → verify_lockdep → assess_risk
Cross-subsystem interaction
Trace how change in A affects B
identify_shared_data → trace_call_paths → identify_assumptions → find_breakpoints → propose_mitigation
CVE root cause analysis
Trace from CVE to root cause to fix
identify_vuln_class → trace_code_path → determine_preconditions → analyze_fix → assess_completeness
Drive adversarial probe generation for the swarm.
Confuse spinlock with mutex
commonUsing spinlock where mutex needed (sleeping under spinlock)
Ignore NUMA topology
subtleAssuming flat memory on NUMA systems
Misunderstand RCU grace period
expert-onlyIncorrect assumptions about when RCU data can be freed
User-space application development
Different discipline
Linux distribution packaging
Too broad
KICE Sources
4
git_repo, mailing_list, database, documentation
AR Probes
~60
3 patterns + 3 failures
Swarm Traps
5
cross-subdomain interactions
Convergence Targets
3 critical
memory-management, network-stack, concurrency-control
CoT Templates (from reasoning patterns)
Corpus Manifest Versioning
Each KICE evolution cycle produces a new corpus-manifest version (v1.0 → v1.1 → ...). Every version records what changed, why, and which model shard it produced. The Nucleus Seal signs the final manifest — full provenance from extraction to graduation.