strykelang 0.12.21

A highly parallel Perl 5 interpreter written in Rust
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
---
name: zshrs+stryke+fusevm patent strategy — eight omnibus claims
description: Eight independently assertible omnibus claims covering the full zshrs+stryke+fusevm portfolio. (A) Unified-AOT (shell binary + script-AOT trailer-format track). (B) zshrs companion-daemon architecture. (C) fusevm three-stage-JIT runtime + no-GC + Rayon-parallel + AOT + embeddable. (D) stryke language design — 20 dependent claims, meta-claim, three-axis universal-access protocol, 20,000-test Perl-parity corpus. (E) Distributed orchestration + stress-testing infrastructure (agent/controller, cluster() SSH worker pool, heat/stress_cpu/stress_mem/stress_io/stress_test, mTLS, auto-terminate guards). (F) AI primitives as language builtin (`ai` 2-letter primitive, tool fn with build-time JSON schema, mcp_server/mcp_connect declarative DSL, ai_filter/map/sort/classify collection builtins, provider-agnostic with local fallback, cost-aware runtime). (G) Web framework (Rails-grade DSL + thread-per-core io_uring + single-binary deploy + radix-trie compile-time routing + per-request arena). (H) Package registry/manager (Cargo+uv+Nix+Bundler synthesis + AOT-compile-everything-including-deps via Cranelift). Each combination + domain application is patentable novelty. Filing strategy: 8 provisionals at same priority date for ~$520 micro-entity / ~$2400 small-entity USPTO fees.
type: project
originSessionId: 5439464a-01d8-4af4-a95a-c4942b441226
---

Patent direction locked 2026-04-28: file the combination claim as the omnibus, not individual primitive claims. Surface expanded 2026-04-30 to eight omnibuses covering the full zshrs+stryke+fusevm portfolio after design-doc audit (`AI_PRIMITIVES.md`, `WEB_FRAMEWORK.md`, `PACKAGE_REGISTRY.md`, `STRESS_TESTING.md`) revealed claim surface beyond the original four-omnibus framing.

**Three structural moats unify the portfolio:**

1. **Three-axis universal-access protocol** (Patent D, claim #16) — pipeline substrate covers (callable × value-class × reflection-metadata) cross-product with zero categorical exclusion. Defeats design-arounds because missing any cell collapses the imitator into a categorical-exclusion regime that #16 escapes.
2. **20,000-test Perl-parity corpus** (Patent D, claim #20) — empirical specification of `--compat` mode behavior. Defeats "we have Perl-compat too" claims because the corpus IS the operational spec; competitors must reproduce 20,000 tests passing to make a comparable claim.
3. **Systematic-absorption-of-userspace-tooling meta-claim** (Patent D meta + Patents F/G/H) — first scripting language to absorb 25+ categories (git, jq, parallel, visualization, crypto, stats, linalg, networking, compression, serialization, HTTP-server, SQLite, DataFrame, PDF, testing, file-watcher, profiler, formatter, LSP, REPL, dep-manager, docs, AOP, AI primitives, web framework, package manager) as core builtin verbs and language subsystems. Inverts "core minimal, libraries optional" → "core encyclopedic, libraries unnecessary." Defeats "we have feature X too" claims because absorbing one category is trivial; absorbing 25+ as a unified architectural commitment is not.

---

## Patent A: Unified-AOT (shell binary + script-AOT trailer-format track)

A first omnibus claim covering AOT compilation of dynamic-scripting source into self-contained native executables, with two distinct application tracks: (Track 1) zshrs shell-binary AOT, (Track 2) stryke script-AOT.

**The omnibus claim shape (engineering sketch, attorney drafts the legal language):**

A method comprising — (a) AOT-compile shell/Perl source + plugin trees + completion defs into one native executable; (b) embed lookup structures (perfect-hash completion tables, zstyle config, bindkeys) as instruction-accessible constants in `.rodata`; (c) weave AOP advice into machine code at compile time with glob-matched command-name patterns; (d) emit hardware-timestamp-counter-based (rdtscp / CNTVCT_EL0) per-function timing in prologues/epilogues; (e) persist runtime mutable state to a writable section of the executable; (f) atomically replace function chunks via memory-page patching at command-boundary safepoints in response to source-edit events. **Wherein** said executable functions simultaneously as interactive shell, script-deployment artifact, plugin host, completion engine, AOP weaver, and persistent state store.

The "wherein" clause locks the claim to the unified-product novelty.

**Why combination, not individual primitives:**

| Primitive | Prior art (cross-domain) |
|-----------|---------------------------|
| Perfect-hash data in `.rodata` | gperf (1989); RE2 / Hyperscan compiled DFAs; Linux kallsyms |
| Live patching of native AOT code | Linux kpatch / ksplice / kgraft / livepatch (2008-2014); Microsoft Detours (1999); DTrace SDT; ftrace; kprobes |
| Compile-time AOP weaving | AspectJ (2001); gcc `-finstrument-functions`; LLVM sanitizers; GHC `-prof` |
| Image-as-binary persistence | Pharo/Squeak Smalltalk (30+ yrs); SBCL `save-lisp-and-die`; HyperCard |
| Hardware-counter inline timing | DTrace; perf events; rdpmc-based profiling |
| AOT shell compilation | None in shell domain; closest is zsh `.zwc` (bytecode, not native) |

**Why non-obvious (KSR v. Teleflex 2007 test):** the six primitives live in six separate fields. A PHOSITA in any one field has zero motivation to reach for the others. Every shell for 50+ years has been C with manual memory, no VM, no image, no AOP — the invention runs counter to the entire teaching of the field.

### Track 2: Script-AOT trailer-format dependent claims

The script-AOT track applies the same AOT philosophy to standalone stryke scripts (vs. zshrs's shell-as-binary). The output is a copy of the stryke binary with the script source embedded as a zstd-compressed trailer.

```
[zstd payload][u64 compressed_len][u64 uncompressed_len][u32 version][u32 reserved][8B magic b"STRYKEAOT"]
```

| Sub-claim | What it covers |
|---|---|
| **A.x: trailer-format binary appending** | Trailer format with OS-loader-invisible placement past mapped segments — runtime detects via 32-byte magic-suffix sniff at ~50µs |
| **A.y: versioned forward-compat trailer** | `[u32 version][u32 reserved]` fields permit future payload formats (v1 source, v2 bytecode, v3 fully-JIT'd native) to coexist in already-shipped binaries with version-dispatch at load time |
| **A.z: idempotent rebuild semantics** | `stryke --exe foo build other.stk -o foo` strips previous trailer first; trailers never stack |
| **A.w: convention-based multi-file project bundling** | `stryke build --project DIR` bundles `main.stk` + `lib/*.stk` (excludes `t/` test dir by default) into single binary, no manifest required |

**Comparison to existing bundlers:**

| Tool | Output size (hello-world) | Forward-compat versioned format? | Idempotent rebuild? | Detection time |
|---|---|---|---|---|
| Go `go build` | 2-5 MB | n/a (compiled native) | yes | n/a |
| PyInstaller | 10-50 MB | no | partial | ~100-500ms extract |
| `pkg` (Node) | 30-80 MB | no | partial | ~50-200ms |
| PAR::Packer (`pp`) | 5-30 MB | no | no | ~100-500ms |
| GraalVM native-image | 30-100 MB | no | yes | n/a |
| Nuitka | 5-15 MB | n/a (compiled C) | yes | n/a |
| **`stryke build`** | **~13 MB** | **yes** | **yes** | **~50µs** |

Three structural differentiators: (i) versioned magic with reserved fields for payload-format evolution; (ii) sub-millisecond detection (vs PyInstaller's ~100-500ms extraction); (iii) idempotent rebuild semantics. Each is engineering work most bundlers don't attempt because they assume "one payload format forever."

---

## Patent B: zshrs companion-daemon architecture

A second omnibus claim covering the singleton companion daemon for a Unix command shell.

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a singleton daemon process that owns all bytecode-cache mutation, supervises long-running detached jobs surviving originating-shell exit, brokers cross-shell publish/subscribe over a Unix domain socket, and federates with peer daemons over secure remote channels (e.g. SSH multiplex); (b) thin shell-client processes that read bytecode from sharded mmap'd image files via direct user-space pointer dereference (data plane), and signal the daemon via JSON-over-socket IPC for all configuration changes, job submissions, cross-shell dispatch, and event subscriptions (control plane); (c) atomic shard-rename followed by index update with generation counters, allowing clients to detect stale mmap regions and re-mmap without coordination; (d) shell-process registry maintained authoritatively by the daemon via socket-connect enrollment, replacing filesystem-based registries used by prior cross-shell coordination plugins. **Wherein** said daemon is dedicated to a Unix command shell — distinct from text-editor daemons (emacs --daemon), credential-cache daemons (ssh-agent / gpg-agent), terminal multiplexers (tmux / screen), batch-job daemons (pueue), and language runtime images (Pharo / SBCL) — and serves N concurrent shell clients with the data plane operating without per-call IPC.

**Stacked world-firsts the daemon claim secures (dependent claims):**

1. **Shell with a dedicated companion daemon spanning bytecode cache + supervised jobs + IPC + federation.** No prior art in any shell.
2. **Native session-persistent shell-job supervision.** No prior art in any shell.
3. **Native cross-shell pub/sub + dispatch as first-class primitives** (`zsubscribe shell:N.commands`, `zsend --tag prod ...`, federation via daemon-to-daemon).

**Per-shell prior-art comparison:**

| Primitive | Prior art (in any shell) |
|-----------|---------------------------|
| Companion daemon | fish 1.x–2.0.x had `fishd` for **universal-variable sync only**; removed in fish 2.1.0 (2014). No other shell ever shipped one. |
| Shared bytecode cache between shell instances | Zero. zsh `.zwc` is per-file static, not daemon-managed. |
| Daemon-supervised detached jobs | Zero. `nohup`/`disown`/`setsid`/`&` detach but don't supervise. |
| Cross-shell pub/sub | zconvey (zsh plugin, filesystem-IPC + per-prompt polling); not built into any shell. |
| Cross-machine shell federation | Atuin syncs history only, runs as separate REST server, not a shell daemon. |

The daemon-architecture combination as a whole, applied to a Unix command shell, has zero prior art. (zshrs-specific; this strykelang docs slice retains it for portfolio completeness.)

---

## Patent C: fusevm runtime architecture

A third omnibus claim covering the **fusevm three-stage JIT runtime** that powers both zshrs and stryke. Splits off from claims A and B because fusevm is **independently licensable for non-shell uses** (other languages, embedded scripting in non-shell tools).

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a bytecode virtual machine for a dynamic scripting language; (b) a tiered just-in-time compilation system spanning at least three optimization regimes simultaneously: (i) a **linear/template/baseline JIT tier** that compiles basic blocks immediately upon first execution with minimal optimization for fast warmup, (ii) a **tracing JIT tier** that records hot linear execution paths through loops and emits type-specialized native code with inlining and unrolling, (iii) a **block/method optimizing JIT tier** that compiles entire functions with inlining + escape analysis + dead-code elimination + polymorphic-call-site specialization; (c) all three tiers coexisting in a single runtime with a tier-promotion policy that escalates code from baseline → tracing or baseline → block based on observed execution profile; (d) a no-garbage-collection memory model based on Rust ownership semantics + atomic reference counting (Arc), eliminating GC pauses entirely; (e) a Rayon-style work-stealing parallel execution substrate exposed as language primitives (e.g., parallel pipelines, parallel directory walks, parallel field-substitution operations); (f) an AOT compilation path producing self-contained native binaries from the same bytecode source; (g) an embeddable substrate API that allows the runtime to be hosted inside other processes (notably a Unix command shell) with zero process-spawn overhead. **Wherein** said runtime, in combination, achieves per-loop overhead within an order of magnitude of statically-compiled native code while supporting dynamic scripting semantics, native parallelism without a global interpreter lock, and zero-startup invocation when embedded.

**No prior runtime ships all three JIT tiers simultaneously:**

| Runtime | Tracing | Linear/baseline | Block/optimizing |
|---------|---------|-----------------|-------------------|
| LuaJIT | ✓ | — | — |
| PyPy | ✓ | — | — |
| V8 (modern) | — | ✓ (Sparkplug) | ✓ (TurboFan) |
| JSC | — | ✓ (Baseline) | ✓ (DFG, FTL) |
| HotSpot | — | ✓ (C1) | ✓ (C2) |
| GraalVM/Truffle | partial (via partial eval) | — | ✓ |
| **fusevm** | **✓** | **✓** | **✓** |

**Stacked world-firsts (dependent claims):**

1. First runtime combining tracing + linear + block JIT in one substrate.
2. First production no-GC dynamic-scripting runtime with JIT (Roc is research; nothing else qualifies).
3. First runtime exposing parallelism as language primitives (par_pipeline, par_sed, par_walk) rather than library calls.
4. First runtime achieving per-iteration overhead within ~100ns of statically-compiled native Rust while preserving dynamic-scripting semantics. (Empirical, supportable with measurement evidence.)
5. **rkyv-backed mmap'd bytecode cache for warm-start script reruns** — first run of `stryke app.stk` parses + JIT-compiles + serializes bytecode to `~/.cache/stryke/scripts.rkyv` via [rkyv](https://crates.io/crates/rkyv) zero-copy serialization; subsequent runs skip parse + compile entirely, mmap the cache and resume execution directly. Empirically delivers **~11× faster warm-start script reruns** vs cold-start. Cache invalidation is content-hash-driven (source mtime + hash); stale entries silently re-compile. The cache architecture parallels zshrs's daemon-managed rkyv image cache (Patent B) but applies to standalone scripts without requiring a daemon — the script binary itself manages cache read/write, the OS page cache shares pages across concurrent invocations. **No prior dynamic-scripting language ships zero-copy mmap'd compiled-bytecode caching with this architecture** — Python's `__pycache__/.pyc` is a marshaled-bytecode format requiring per-import deserialization (no zero-copy, no mmap-direct-execution); Ruby has no script-level bytecode cache; Lua's `string.dump` is a serialization primitive, not a managed cache. Stryke's combination of (a) rkyv zero-copy archive format, (b) mmap-direct-execution from cache, (c) content-hash invalidation, (d) shared-via-OS-page-cache across concurrent invocations is novel for the dynamic-scripting category.

**Empirical positioning** (from `examples/rosetta/README.md`): "2nd fastest dynamic language runtime ever benchmarked for singlethreaded — behind only Mike Pall's LuaJIT, and beating it on 3 of 8 benchmarks. The fastest on all multithreaded benchmarks. Faster than perl5, Python, Ruby, Julia, and Raku on every benchmark." Combined with the rkyv warm-start cache (#5), repeated invocations of the same stryke script approach native-binary cold-start performance — relevant for shell-embedded one-liner workloads where the same `stryke -e '...'` or `stryke app.stk` invocation runs hundreds of times across a session.

**Closest analog in patent literature:** Sun's HotSpot JIT patents (1999-2005) cover C1+C2 tiered compilation but for a strongly-typed JVM; Oracle's GraalVM/Truffle patents cover partial-evaluation-based JIT for dynamic languages but only one optimization tier. **No patent covers the three-tier combination + no-GC + parallel + dynamic-scripting + embeddable composition.**

---

## Patent D: stryke language design

A fourth omnibus claim covering the **stryke language design** as a unified syntactic + semantic synthesis. Splits off from C because language design is **independently licensable from the runtime substrate** — stryke's syntactic primitives could be implemented on a different runtime; fusevm could host a different language design.

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a programming language for shell-embedded scripting and one-liner workloads, comprising at least: (b) **two-axis positional-argument access syntax** of the form `_N<...<` where N is a positional-slot index (0, 1, 2, ...) and each `<` character represents one level of closure-nesting outward from the current frame, allowing reference to positional arguments of arbitrarily-deep enclosing closures or function frames without named-binding fallback; including (b.i) **four-way cross-tradition aliasing for the default slot 0** wherein `_`, `$_`, `_0`, and `$_0` are accepted as equivalent spellings (Scala bare-`_` ≡ Perl topic-var `$_` ≡ Mathematica/Swift positional ≡ sigil-positional synthesis); (b.ii) **two-way aliasing for indexed slots N≥1**; (b.iii) **orthogonal depth-encoding** wherein the `<...<` depth marker composes uniformly with any spelling choice; (c) **polymorphic literal-typed range syntax** of the form `<literal>:<literal>[:<step>]` where the literal type is inferred from the literal form (integer → integer range, character → character range, roman-numeral → roman range, ISO-8601 date → date range, RFC3339 datetime → datetime range, dotted-quad IP → IP range) and the step semantics are inferred from the inferred type; (d) **encyclopedic bundled standard library** of 3000+ built-in functions covering cryptographic primitives, statistical functions, linear-algebra primitives, networking, date-time arithmetic, compression, structured-data formats, shell-process primitives, type predicates (≥150 distinct is_* predicates), ANSI/terminal styling, and tools-typically-invoked-as-subprocesses-elevated-to-language-builtins (notably git operations and jq queries) — all available without import or package-load step; (d.i) said standard library further providing **parallel short-name aliases** (2-3 character forms) for long-name builtins systematically across the encyclopedic corpus; (e) **a unified threading/pipe operator family** comprising (e.i) pipe-forward operator (`|>`); (e.ii) Clojure-tradition compile-time AST-rewriting threading macros (`->`, `->>`); (e.iii) Racket-tradition threading macros (`~>`, `~>>`); coexisting as semantically-distinct first-class operators; (e.iv) said threading operators supporting **three notation forms per pipeline stage**: bare-function, inline arrow-block, and **anonymous-positional-placeholder** (`f(_, 5)`, `f(5, _)`, `f(5, _, 10)`); (e.v) wherein the **arrow-block sigil `>{ ... }`** is a syntactic primitive distinct from regular closure-block; (f) **Perl-style sigils** (`$`, `@`, `%`, `&`, `*`); (g) **Ruby-derived ultra-terse output verbs** (`p`, `ep`); (h) **parallelism-as-syntactic-primitive** via builtins (`par_pipeline`, `par_sed`, `par_walk`, `par_csv_read`, `pmap`, `pgrep`, `psort`, `pfor`, `pmaps`, `pgreps`, `fan`, `fan_cap`, `pchannel`, `pselect`, `barrier`, `ppool`); (i) **terminal-ASCII data visualization-as-syntactic-primitive** via builtins (`histo`, `sparkline`, `barchart`, `table`); and (j) **systematic absorption of userspace tooling categories into the core language** including version control (git operations as builtins), structured-data query (jq absorbed as builtin verb), parallel processing primitives, terminal visualization, cryptography, statistics, linear algebra, networking, compression, structured-data serialization, and (per Patents F/G/H) AI primitives, web-framework primitives, and package-management primitives. **Wherein** said language is invocable both as a standalone interpreter and as a shell-embedded sublanguage with zero process-spawn cost when embedded; **wherein** the threading-operator family of (e), with its three notation forms (bare-function, arrow-block, anonymous-positional-placeholder), collapses what Clojure achieves with three separate macros (`->`, `->>`, `as->`) into one operator family with richer per-stage notation; **wherein** the systematic absorption of (j) inverts the prevailing scripting-language design philosophy of "core minimal, libraries optional" to "core encyclopedic, libraries unnecessary"; and **wherein** said threading-operator family of (e) operates as a **three-axis universal-access protocol** uniformly covering: (k.i) **the callable axis** — every function-class expression in the language (built-in functions, user-defined functions, closures, lambdas, arrow-block expressions, operator expressions wrapped via arrow-block, and placeholder-substituted call sites) is uniformly addressable as a pipeline stage with zero adapter code, zero registration requirement, and zero type-system constraint; (k.ii) **the value axis** — every value-class in the language (primitive scalars; collections including arrays, hashes, sets; user-defined structures and class instances providing object-oriented programming; nested-structures of arbitrary depth; mixed-type heterogeneous collections) flows through threading uniformly without type-system adapters or "stream-compatible" interface implementations; (k.iii) **the reflection axis** — language metadata including method enumeration, field enumeration, class-hierarchy inspection, dynamic method dispatch by name-string, and dynamic field access/mutation by name-string is itself reified as threadable data and composes with the callable and value axes at no syntactic boundary; and **wherein** the absence of categorical exclusion across all three axes is itself a patentable architectural commitment, in that competitor languages whose pipeline paradigm excludes even one cell from the (callable × value-class × reflection-metadata) cross-product fall outside this claim.

**Stacked world-firsts (20 dependent claims):**

1. **First syntax for two-axis positional-argument access (index × depth) across closure nesting and function boundaries** — `_N<...<` family.
2. **First language to ship four-way cross-tradition aliasing for the default closure slot:** `_` ≡ `$_` ≡ `_0` ≡ `$_0`.
3. **First language with orthogonal depth-encoding decoupled from spelling choice.**
4. **First polymorphic literal-typed range syntax** spanning int + char + roman + date + datetime + IP without trait/protocol boilerplate.
5. **First scripting language with encyclopedic no-import bundled stdlib (3000+) outside Mathematica.**
6. **First general-purpose scripting language with git operations as language builtins.**
7. **First language with parallelism (par_pipeline, par_sed, par_walk) as syntactic-primitive verbs rather than library calls.**
8. **First scripting language synthesizing Clojure threading + Racket arrows + Scala underscore + Perl sigils + Ruby p/string-interp into a unified syntax.**
9. **First scripting language with absorbed jq DSL as a single builtin verb.**
10. **First non-Lisp language to ship pipe operator (`|>`) AND threading macros (`->`, `->>`, `~>`, `~>>`) as semantically-distinct first-class operators.**
11. **First language with three-form threading in one operator** — bare-function, inline arrow-block, anonymous-positional-placeholder.
12. **First `>{ ... }` arrow-block sigil** for inline anonymous transforms in threaded pipelines.
13. **First anonymous-positional-placeholder threading** (`f(_, 5)`, `f(5, _)`, `f(5, _, 10)`).
14. **First scripting language to systematically ship parallel short/long namespaces** (`tm`/`trim`, `rv`/`reverse`, etc.).
15. **First scripting language to ship terminal-ASCII data visualization as language-level builtins** (`histo`, `sparkline`, `barchart`, `table`).
16. **First pipeline substrate operating as a three-axis universal-access protocol** (callable × value-class × reflection-metadata) with zero categorical exclusion. The exclusion-free property is the architectural commitment defended against design-arounds: any imitator who excludes even one cell falls outside the claim.
17. **First scripting language to ship nine purpose-built globally-named introspection hashes** (`%b`/`%k`/`%all`/`%pc`/`%e`/`%a`/`%d`/`%c`/`%p`) populated at compile time from source-of-truth parsing, providing O(1) bidirectional indexing. The triple `%b`/`%k`/`%all` carries the disjoint-union invariant `%all = %a + %b + %k` (callables, keywords, and aliases each in exactly one).
18. **First language to support intra-expression composition of pipe-forward and threading-macro operators** — `~> seed stage1 stage2 |> stage3` mixes `~>` and `|>` in a single expression.
19. **First scripting language to ship bidirectional source-level conversion to its parent-language tradition (Perl) as a built-in subcommand** — `stryke convert` (Perl → stryke) and `stryke deconvert` (stryke → Perl).
20. **First scripting language to ship empirically-validated full Perl 5 compatibility (`--compat` mode) on a JIT'd runtime** — verified by a 20,000-test parity corpus pinning behavior against upstream Perl 5. The 20,000-test parity corpus IS the operational specification of Perl-compat — competitors must reproduce that corpus to make a comparable claim.

**META-CLAIM** (the architectural pattern unifying claims #5, #6, #7, #9, #15, plus Patents F/G/H, plus ~1500 of the 3244 builtins):

> **First scripting language to systematically absorb userspace tooling categories — version control, structured-data query, parallel processing, terminal visualization, cryptography, statistics, linear algebra, networking, compression, structured-data serialization, HTTP-server, SQLite, DataFrame, PDF generation, testing framework, file watcher, profiler, formatter, language server, REPL, dependency manager, documentation system, AOP, AI primitives, web-framework primitives, package-management primitives — into the core language as builtin verbs and language subsystems, rather than library imports or subprocess invocations.**

The category-absorption pattern is itself a defensible meta-claim: a competitor adding ONE absorbed category trivially imitates one feature without infringing the meta. The systematic absorption across 25+ categories is what's claimed.

---

## Patent E: Distributed orchestration + stress-testing infrastructure

A fifth omnibus claim covering language-level distributed fleet orchestration, with `cluster()` SSH worker pools, agent/controller architecture, and bare-metal stress-testing builtins.

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a built-in distributed-execution primitive `cluster()` that opens persistent SSH connections to remote hosts, spawns persistent worker processes (`stryke --remote-worker`), and exposes the resulting worker pool as a language-level value, with `pmap_on $cluster { ... }` distributing work across all slots with work-stealing; (b) an **agent/controller orchestration architecture** wherein `stryke agent --controller HOST:PORT` runs on each fleet node and registers with a central controller, while `stryke controller` provides an interactive REPL with fleet-wide commands (`status`, `fire SECS`, `terminate`, `shutdown`); (c) **bare-metal stress-testing builtins** (`stress_cpu`, `stress_mem`, `stress_io`, `stress_test`, `heat`) that pin every CPU core to 100% TDP for a specified duration, exposed as language-level verbs composable with the threading-operator family; (d) **mTLS controller-agent communication** for production deployment; (e) **audit log of all controller commands** for compliance; (f) **auto-terminate guards** including max temperature thresholds (Celsius), max session duration, and ack timeout fallback. **Wherein** said language is positioned as a "server farms first" language — the first programming language designed from the ground up for distributed infrastructure load testing, capacity validation, and BCP/DR exercises — and **wherein** the same single binary functions as: standalone interpreter, fleet agent, fleet controller, stress-testing tool, and AOT-compile target, with mode selection via subcommand (`stryke`, `stryke agent`, `stryke controller`, `stryke build`, `stryke --remote-worker`).

**Why this is its own omnibus, not a sub-claim of A-D:**

| Existing claim | Covers | Does NOT cover |
|---|---|---|
| Patent A (AOT) | single-binary AOT compile | fleet deployment of those binaries |
| Patent B (zshrs daemon) | per-machine shell coordination + cross-host federation hint | language-level fleet REPL, agent/controller protocol, bare-metal stress builtins |
| Patent C (fusevm) | runtime substrate | distributed execution model |
| Patent D (stryke language) | syntactic + semantic primitives | remote/distributed extension as architectural commitment |

Patent E is the strykelang-side equivalent of zshrs's daemon (Patent B) extended to fleet scale — and adds the bare-metal stress-testing dimension that no shell daemon provides.

**Prior-art landscape:**

| Tool / framework | Year | Language-level orchestration? | Single-binary deploy? | REPL-as-controller? | Stress-testing builtins? |
|---|---|---|---|---|---|
| Puppet | 2005 | no — YAML DSL | no — Ruby runtime | no | no |
| Salt | 2011 | no — YAML DSL | no — Python runtime | partial | no |
| Ansible | 2012 | no — YAML DSL | no — Python runtime | no | no |
| Chef | 2009 | partial (Ruby DSL) | no — Ruby runtime | no | no |
| Spark | 2014 | partial (Scala/Python API) | no — JVM | partial (spark-shell) | no |
| Dask distributed | 2014 | partial (Python API) | no — Python runtime | partial | no |
| Ray | 2017 | partial (Python API) | no — Python runtime | partial | no |
| Kubernetes | 2014 | no — YAML manifests | no — container images | no | no |
| Nomad | 2015 | no — HCL DSL | yes (Go binary) | no | no |
| Fabric (SSH) | 2008 | partial (Python API) | no — Python on remote | no | no |
| stress-ng | 2016 | n/a — standalone tool | yes (binary) | no | yes (sole purpose) |
| **stryke distributed** | **2026** | **yes — orchestration IS the language** | **yes — AOT'd binaries, no runtime needed** | **yes — controller is stryke REPL** | **yes — `heat`, `stress_*` as language builtins** |

**No prior tool ships all five: language-level orchestration + single-binary agents + REPL-as-controller + bare-metal stress builtins + AOT-compile-everything.** Each existing tool fails at least one cell:

- Spark/Ray/Dask: have language-level APIs but require runtime on every node (kills "scp anywhere" deploy)
- Nomad/Consul/HashiCorp: have single-binary deployment but use YAML/HCL DSL, not host language
- Fabric: REPL-driven for SSH but requires Python on remote and isn't language-level
- Ansible/Puppet/Chef/Salt: YAML/DSL configuration languages, not general-purpose
- stress-ng: dedicated stress tool but isn't a programming language

**Stacked world-firsts (dependent claims):**

1. **First scripting language with `cluster()` SSH worker pool as a built-in primitive** — opens persistent connections, spawns `--remote-worker` processes, exposes pool as language value.
2. **First scripting language with `pmap_on $cluster` work-stealing distribution** — composes with the standard parallel-builtin family.
3. **First scripting language with agent/controller fleet REPL** — `stryke controller` is a stryke REPL with fleet-wide commands.
4. **First scripting language with bare-metal stress-testing builtins** (`stress_cpu`, `stress_mem`, `stress_io`, `stress_test`, `heat`) that pin CPU cores to 100% TDP.
5. **First scripting language with mTLS controller-agent communication + audit log + auto-terminate guards** as a unified production-grade orchestration architecture.
6. **First scripting language whose same binary functions as interpreter / AOT target / fleet agent / fleet controller / stress tool** with mode selection via subcommand.

---

## Patent F: AI primitives as language builtin

A sixth omnibus claim covering **AI as a language primitive, not a library**. The simplest invocation is `ai $prompt`; the complex form composes from the same primitive — there is no separate "advanced API."

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a built-in `ai` primitive (two-letter, no-import, always-available) that accepts a prompt and optional context value and runs an agent loop to completion (tool call → tool result → next call → final answer); (b) said `ai` primitive supporting **three syntactic invocation forms**: function call (`ai "summarize", $doc`), threading-macro (`~> $doc ai "summarize"`), and pipe-forward (`$doc |> ai "summarize"`); (c) a **`tool fn` declaration** that marks a function as agent-callable, with build-time generation of (c.i) JSON schema from the function signature and parameter types, (c.ii) tool description from the function docstring, (c.iii) automatic registration of all in-scope `tool fn`s with subsequent `ai` calls; (d) a **declarative `mcp_server { ... }` block** that compiles to a spec-compliant MCP (Model Context Protocol) server with one or more tool/resource/prompt declarations, exposable via stdio/websocket/HTTP transports, with a `stryke build --mcp-server` flag emitting a standalone binary; (e) an **`mcp_connect` client** that connects to remote MCP servers via stdio/HTTP/websocket transports, with discovered tools/resources/prompts auto-attaching to subsequent `ai` calls; (f) **AI collection builtins** (`ai_filter`, `ai_map`, `ai_classify`, `ai_sort`, `ai_match`, `ai_dedupe`) that compile collection operations to single batched LLM calls where possible; (g) **provider-agnostic with namespaced extensions** — same `ai` call hits Claude, GPT, Gemini, or local llama.cpp, with provider chosen by `stryke.toml` config and swappable at runtime; (h) **local-fallback-always-works** semantics — stryke binaries can ship with a quantized model linked statically, so `ai` works offline with degraded quality; (i) **cost-aware-by-default** runtime including result cache, automatic batching, parallel rate-limit-aware execution, hard cost ceilings (`max_cost_run`), cost introspection (`ai_cost`), and pre-flight token estimation (`tokens_of`); (j) **deterministic-in-tests** semantics via `ai_mock { ... }` blocks that intercept every AI primitive in scope, with pattern-matching on prompts and structured/string/generator responses; **wherein** the AI subsystem composes with the rest of the language: web-framework controllers (per Patent G), package-manager dependency declarations, cluster-dispatch (per Patent E), effect handlers, capability-based access control. **Wherein** said language is positioned as the first AI-native general-purpose scripting language where AI is a language primitive rather than a third-party library, framework, or SDK.

**Why combination, not individual primitives:**

| Primitive | Prior art |
|-----------|-----------|
| LLM API call | OpenAI Python SDK (2020); Anthropic Python SDK (2022); LangChain (2022); LlamaIndex (2022); every-framework-and-its-cousin |
| Tool/function calling with JSON schema | OpenAI function calling (2023); Anthropic tools API (2024); LangChain Tools |
| MCP (Model Context Protocol) | Anthropic published spec (2024-11) |
| Agent loop | LangChain agents; AutoGPT (2023); BabyAGI (2023); CrewAI (2024) |
| Streaming responses | every modern LLM SDK |
| Cost tracking | LangChain callbacks; OpenAI usage API; manual instrumentation |
| Test mocking | unittest.mock; vcr.py; every-test-framework |
| Local model inference | llama.cpp (2023); ollama (2023); LM Studio (2023) |
| Multi-provider abstraction | LangChain; LiteLLM (2023); aisuite (2024) |

**Why combination is non-obvious:**

1. **AI-as-builtin teaching.** Every existing language treats AI as a library/SDK that must be imported. The accepted convention is "language is general-purpose, AI is a domain library." Stryke's commitment to `ai` as a built-in primitive (no import, always available, two letters) inverts the convention. PHOSITAs in language design are taught to keep stdlibs minimal; this claim explicitly violates that teaching.
2. **MCP-native combined with provider-agnostic abstraction.** No prior language has both (a) a declarative `mcp_server { ... }` block AND (b) automatic MCP-server-to-`ai`-call attachment — they exist as separate library layers if at all.
3. **Build-time JSON schema from function signatures.** TypeScript/Python practice requires manual schema declaration alongside function definition (see LangChain, OpenAI Python SDK). Build-time schema-generation-from-signature is novel.
4. **AI collection builtins compile to batched calls.** Every other framework treats `[f(x) for x in items]` as N separate LLM calls. `ai_map` compiles to a single batched prompt with cost-aware batching. No prior framework ships this as a language-level primitive.
5. **Cost-as-runtime-concern.** Cost tracking exists in libraries; **cost-aware execution as a runtime guarantee** (with hard ceilings that abort the program) is novel. PHOSITAs treat cost as a deployment/billing concern, not a language-runtime concern.
6. **Composition with web framework / package manager / cluster.** No prior AI library composes natively with HTTP frameworks, package managers, and distributed-execution primitives at the language level — they're all separate library imports requiring glue code. Stryke's design has them all touch the same `ai` primitive.
7. **Cross-tradition synthesis.** Combining (a) language-builtin AI, (b) MCP-native, (c) `tool fn` build-time schema, (d) AI collection builtins, (e) local-fallback-always, (f) cost-aware ceilings, (g) deterministic test mocking, (h) cluster-dispatch composition — eight independent AI-architecture commitments unified in one omnibus.

**Stacked world-firsts (dependent claims):**

1. **First general-purpose scripting language with `ai` as a no-import language builtin.** Two-letter primitive, always available.
2. **First language with three-form AI invocation** — function call, threading-macro, pipe-forward all compile to the same primitive.
3. **First language with `tool fn` declaration generating JSON schema + description at build time** from function signature and docstring, with automatic in-scope tool registration.
4. **First language with declarative `mcp_server { ... }` block** compiling to spec-compliant MCP server, with `stryke build --mcp-server` standalone-binary emission.
5. **First language with `mcp_connect` clients that auto-attach discovered tools/resources/prompts to subsequent `ai` calls** with no re-registration step.
6. **First language with AI collection builtins (`ai_filter`/`ai_map`/`ai_classify`/`ai_sort`/`ai_match`/`ai_dedupe`)** compiling to single batched LLM calls.
7. **First language with provider-agnostic AI calls + namespaced provider extensions** — `ai` is one builtin; `[ai]` config picks provider; provider-specific options via namespaced extensions.
8. **First language with statically-linked local-LLM fallback** — `ai` works offline with degraded quality regardless of API connectivity.
9. **First language with cost-aware runtime** — result cache + batching + parallel + hard ceiling (`max_cost_run`) + introspection (`ai_cost`) + pre-flight estimation (`tokens_of`) as runtime guarantees, not library APIs.
10. **First language with `ai_mock { ... }` deterministic test interception** — patterns match prompts; CI fails build on unmatched live calls (`STRYKE_AI_MODE=mock-only`).
11. **First language whose AI primitive composes natively with web framework, package manager, cluster dispatch, and (planned) effect-system + capability-system** at the language level.

**Closest analog in patent literature:** OpenAI's function-calling and Anthropic's tools API have product implementations but limited issued patents. LangChain is open-source library, no patent claims. MCP spec is open-source from Anthropic, no patent claims. **No prior patent covers the eleven-primitive combination applied to a general-purpose scripting language.**

---

## Patent G: Web framework

A seventh omnibus claim covering **Stryke Web** — Rails-grade developer experience compiled to single statically-linked binary with thread-per-core io_uring runtime.

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a web application framework hosted by a dynamic-scripting language runtime (fusevm per Patent C) wherein (a.i) routes are declared via Rails-style DSL (`route :GET, "/", home#index; resources :posts; namespace :api ...`) parsed at build time and **compiled into a radix trie with parameter-capture indices serialized into the binary as a static lookup table**, with match resolving in 50-200ns with zero allocation; (a.ii) controller methods receive request/response/params/session/cookies/flash as in-scope methods (Rails-style ergonomics, explicit method-table entries on `Controller` base class, no method_missing magic); (a.iii) ORM models compose chainably (`Post.published.recent.limit(20)`) with N+1 detection at compile time; (a.iv) view templates compile to native code; (b) a runtime model of **thread-per-core with io_uring on Linux** (glommio-backed, one executor per core, pinned, with `SO_REUSEPORT` for kernel-side load balancing, per-core memory pools, per-core connection state, per-core arena allocator) and tokio M:N fallback on macOS/Windows; (c) **per-request arena allocator** wherein every request allocates from a bumpalo-style arena, all parsed headers + parameters + response body bump a pointer in the arena, response completion drops the entire arena in one `free()` (zero individual deallocations on the hot path); (d) **HTTP/1.1, HTTP/2, HTTP/3 in the same binary** (httparse + custom fast-path for HTTP/1; h2 crate for HTTP/2; quinn for HTTP/3 over QUIC, opt-in via config) sharing the same handler API, with version negotiation via ALPN and Alt-Svc; (e) **rustls + kTLS on Linux ≥ 4.13** for ~2x OpenSSL throughput; (f) WebSocket (`ws "/chat", chat#stream`) and Server-Sent Events (`sse "/events", events#stream`) as first-class route declarations; (g) deployment via single statically-linked native binary produced by `s build --release` (per Patent A.script-AOT track) — `scp target/release/myapp prod:` is the entire deploy pipeline, no PaaS, no Docker, no nginx required, no `bundle install` / `node_modules` / `pip install`; (h) zero install-time code execution on target machine — the binary is the app, the OS is the runtime, nothing else; (i) `s new myapp --web` scaffolding generators producing working CRUD app in under 30 seconds. **Wherein** said framework targets top-3 TechEmpower throughput (>3M req/s plaintext at maturity), <5ms cold start, <15MB idle memory, <4KB per concurrent connection — public commitments tracked in CI with regressions blocking merge.

**Why combination, not individual primitives:**

| Primitive | Prior art |
|-----------|-----------|
| Rails-style routing DSL | Rails (2004); Phoenix (2014); Rocket (Rust, 2017); Sinatra (2007) |
| Thread-per-core with io_uring | seastar (ScyllaDB, 2014); glommio (2020); monoio |
| Compile-time routing trie | Rocket; some actix-web versions; matchit crate |
| Per-request arena allocator | hyper (partial); some C++ frameworks (drogon, lithium) |
| HTTP/1+2+3 in one binary | actix-web, axum (with separate features) |
| Single-binary deploy | Go web frameworks (Gin, Echo, Fiber); Phoenix releases |
| Rust-runtime web framework | actix-web (2017); axum (2021); rocket; warp |
| Rails-DX ergonomics | Phoenix (Rails-inspired but Elixir); Buffalo (Go); Loco (Rust, 2024) |

**Why combination is non-obvious:**

1. **Rails-DX + native compilation teaching.** Every prior framework forces a choice: Rails-DX (Ruby/Phoenix/Buffalo) with runtime overhead, OR native compilation (actix-web/axum) with much-coarser DSL. The PHOSITA in web framework design treats these as a tradeoff curve with no Pareto-dominant point. Stryke web claims to land both: Rails-DX + AOT-compiled native code. The combination requires a host language that has both (a) a Rails-quality DSL surface and (b) a native AOT compile path — almost no language has both.
2. **Single-binary deploy + Rails-DX.** Go has single-binary deploy but no Rails-DX (Gin/Echo/Fiber are explicit, low-magic). Phoenix has Rails-DX but ships a BEAM runtime (not a true single-binary). The combination has no prior art.
3. **Convention-over-configuration in a strongly-typed-runtime context.** Most "convention" frameworks (Rails, Django, Phoenix) live in dynamically-typed languages where convention emerges from runtime introspection. Stryke web has gradual typing (per Patent D) AND convention-over-configuration AND build-time route compilation — three commitments most frameworks pick one or two of.
4. **Thread-per-core io_uring + script-language runtime.** No prior dynamic-scripting-language web framework runs on glommio/io_uring at the runtime level. Python/Ruby/Node runtimes have fundamentally incompatible threading models. fusevm (Patent C) makes thread-per-core feasible because it has no GIL and Rayon-native parallelism.
5. **Deployment story dominance.** Single-binary scp = `s build --release && scp target/release/myapp prod:` — every Rails-DX framework requires more (Rails: bundler+Ruby+Passenger; Django: gunicorn+Python+nginx; Phoenix: BEAM+release-tooling; Express: node+npm+pm2). Stryke web's deploy story is Go-grade.

**Stacked world-firsts (dependent claims):**

1. **First web framework combining Rails-grade DX + AOT-compiled-to-native-code single-binary deploy.** Goes beyond what Rails/Phoenix/actix-web/axum each ship individually.
2. **First scripting-language web framework on thread-per-core io_uring.** No GIL, no Python/Ruby threading limitations.
3. **First web framework with build-time radix-trie routing + Rails-style `resources :posts` DSL** — the DSL expands at compile time to the seven CRUD routes serialized as static lookup tables.
4. **First web framework with WebSocket and SSE as first-class route declarations** in a Rails-style DSL.
5. **First web framework whose entire dependency graph (framework + user code + stdlib) compiles to native machine code via Cranelift** (per Patent C + Patent A.script-AOT).
6. **First web framework explicitly committing to top-3 TechEmpower performance** with public CI-tracked benchmarks — public-commitment performance targets are a deliberate marketing/architectural claim.
7. **First web framework whose `s new myapp --web` produces working CRUD app in <30 seconds with zero PaaS / Docker / nginx / bundle / npm / pip dependencies.**

**Closest analog in patent literature:** Web framework patents are sparse — Rails/Django/Phoenix/actix-web are open-source with no patent claims. Microsoft has ASP.NET patents (mostly process patents around request lifecycle). Google has gRPC patents. **No prior patent covers the seven-primitive combination of Rails-DX + AOT-native-compile + io_uring + radix-trie-build-time + per-request-arena + HTTP-1/2/3-unified + zero-target-runtime applied to a dynamic-scripting language.**

---

## Patent H: Package registry/manager

An eighth omnibus claim covering the **stryke package manager** — Cargo's model + uv's execution speed + Nix's reproducibility + Bundler's lockfile sacredness + npm's `[scripts]` table — with the "kill feature" that `s build --release` AOT-compiles **the entire program (user code + every dep + stdlib) through Cranelift to native machine code**, producing a single statically-linked binary.

**The omnibus claim shape (engineering sketch):**

A method comprising — (a) a TOML-based manifest (`stryke.toml`) with sections for package metadata, dependencies (with semver ranges, exact pins, path-deps, git-deps), dev-deps, arbitrary-named groups (Bundler-style beyond dev/prod), features (per-package scoped, not unified workspace-wide), `[scripts]` table for project-local task running (npm's one good idea), `[bin]` entries for executable targets, and `[workspace]` for first-class multi-package workspaces; (b) a deterministic lockfile (`stryke.lock`) with hash-pinned integrity (Nix-style reproducibility) — every dependency is hash-pinned, two `s install`s from the same lockfile on different machines produce byte-identical store contents; (c) a global content-addressable store at `~/.stryke/store/name@version/` (human-readable paths, not Nix-style hash-paths) with hash-pinning happening in the lockfile; (d) a parallel resolver and parallel fetch/extract/verify (uv-style execution speed, milliseconds not minutes); (e) a unified CLI surface — `s init`, `s new myapp`, `s build`, `s run`, `s test`, `s bench`, `s doc`, `s check`, `s fmt`, `s clean`, `s install`, `s add`, `s remove`, `s update`, `s tree`, `s outdated`, `s audit`, `s publish`, `s yank`, `s search` — one binary, one mental model; (f) **deliberate exclusions** — no per-project deps tree (no `node_modules`/`vendor`/`packages`), no install-time code execution (no `build.rs`/`postinstall`), no hoisting, no phantom deps, no peer deps, no mutable registries (no left-pad re-runs); (g) **first-class private registries** (no centralized monoculture); (h) **the killer feature** — `s build --release` AOT-compiles the entire program (user code + every transitively-resolved dep + stdlib) through Cranelift to native machine code, producing a single statically-linked ELF/Mach-O/PE binary in `target/release/`, no interpreter required on target machine, no JIT warmup, no bytecode at runtime — Perl-grade ergonomics with Go-grade binaries. **Wherein** said package manager combines (i) Cargo's manifest+lockfile+resolver model, (ii) uv's parallel-Rust-native execution, (iii) Nix's hash-pinned reproducibility, (iv) Bundler's sacred-lockfile discipline, (v) npm's `[scripts]` table — picking proven winners from a decade of design experiments while skipping the legacy mistakes (no node_modules / install-time-code / hoisting / phantom-deps / mutable-registries / centralized-monoculture).

**Why combination, not individual primitives:**

| Primitive | Prior art |
|-----------|-----------|
| TOML manifest | Cargo (Rust, 2014); Pyproject.toml (PEP 621, 2020); Bundler-Gemfile is Ruby-DSL |
| Deterministic lockfile | Cargo (Cargo.lock); Yarn (yarn.lock, 2016); Bundler (Gemfile.lock, 2010); pip-tools (requirements.txt, 2010s) |
| Hash-pinned reproducibility | Nix (2003); NPM lockfile-hashes (since 5.0); Cargo Cargo.lock with Cargo-checksums |
| Parallel resolver | uv (2024); pnpm (2017); Cargo's resolver; Renovate |
| Global content-addressable store | Nix; pnpm store; Maven local repo |
| `[scripts]` table | npm package.json scripts; just/Justfile |
| Workspace support | Cargo workspaces; npm workspaces; Yarn workspaces |
| AOT compile-everything-including-deps | GraalVM native-image (2018); Go (compiles deps as part of binary); Rust (similar); Nuitka (2007 Python) |
| AOT compile dynamic-language program-and-deps | Nuitka (Python, partial); GraalVM (JVM); **none for general-purpose scripting languages** |

**Why combination is non-obvious:**

1. **Five-tradition synthesis applied to package management.** Each of (Cargo, uv, Nix, Bundler, npm-`[scripts]`) is the canonical leader in its own dimension. Picking proven-winners from five separate package-manager traditions and unifying them is itself a deliberate cross-tradition design act. Most package managers either (a) inherit from one tradition (Bundler ≈ Gemfile-style, Cargo ≈ Rust-only), (b) accumulate ad-hoc features (npm), or (c) reinvent everything (Nix). Stryke's package manager is the first to do principled multi-tradition synthesis at the design-doc level.
2. **AOT-compile-everything-via-Cranelift for a dynamic-scripting language.** GraalVM native-image does this for JVM bytecode (typed languages); Nuitka does partial for Python. **No prior dynamic scripting language ships `package_manager build --release → native binary including all transitively-resolved deps + stdlib` with this level of completeness.** Stryke's claim is that you write Perl-flavored stryke, declare deps in `stryke.toml`, run `s build --release`, and get a Go-grade binary out.
3. **Deliberate exclusions are themselves a claim.** "No node_modules" / "no install-time code execution" / "no hoisting" / "no phantom deps" / "no peer deps" / "no mutable registries" / "no centralized monoculture" — each exclusion fixes a known package-manager footgun. The set of exclusions IS the design commitment.
4. **First-class private registries.** Centralized monocultures (npm, PyPI) have failure modes (left-pad, name-squatting, supply-chain attacks). Stryke commits to private-registry-first as a structural decision.
5. **Per-package scoped features** vs Cargo's unified-workspace-wide features. Cargo's biggest footgun (a consumer turning on `feature = "yaml"` silently flips it on for every other package in the graph) is fixed by scoping features per-package.

**Stacked world-firsts (dependent claims):**

1. **First package manager to synthesize Cargo + uv + Nix + Bundler + npm-`[scripts]` traditions** with explicit exclusion of node_modules / install-time-code / hoisting / phantom-deps / mutable-registries / centralized-monoculture.
2. **First package manager for a dynamic scripting language to AOT-compile the entire program (user code + every transitively-resolved dep + stdlib) to native machine code via Cranelift**, producing a single statically-linked binary with no target-machine runtime requirement.
3. **First package manager with per-package scoped features** (Cargo's unified-workspace-feature footgun fixed).
4. **First package manager with human-readable global-store paths AND hash-pinned lockfile reproducibility** — Nix's reproducibility without Nix's opaque paths.
5. **First package manager whose `s build` produces three artifacts from one source: standalone interpreter target, AOT-compiled native binary, MCP-server binary (per Patent F)** via flag selection.

**Closest analog in patent literature:** Package manager patents are sparse and mostly about specific algorithms (Microsoft's NuGet, Apple's CocoaPods). **No prior patent covers the multi-tradition synthesis + AOT-native-compile-everything combination applied to a dynamic scripting language.**

---

## Filing Strategy for the Eight-Omnibus Portfolio

| Omnibus | Cost (provisional, micro) | Cost (provisional, small) | Locks |
|---|---|---|---|
| A: unified-AOT (shell + script) | ~$65 | ~$300 | shell-binary + script-AOT trailer-format |
| B: zshrs daemon | ~$65 | ~$300 | shell-coordination architecture |
| C: fusevm runtime | ~$65 | ~$300 | three-stage-JIT + no-GC + parallel + AOT + embeddable |
| D: stryke language design | ~$65 | ~$300 | 20 dependent claims + meta-claim + 3-axis universality + Perl-parity |
| E: distributed orchestration + stress | ~$65 | ~$300 | language-native fleet REPL + cluster + heat builtins |
| F: AI primitives | ~$65 | ~$300 | `ai` builtin + `tool fn` + MCP-native + collection builtins + cost-aware |
| G: web framework | ~$65 | ~$300 | Rails-DX + thread-per-core io_uring + radix-trie-build-time + single-binary |
| H: package registry/manager | ~$65 | ~$300 | five-tradition synthesis + AOT-everything via Cranelift |
| **Total** | **~$520** | **~$2400** | **8 independently-assertible omnibus claims** |

**File all eight under the same priority date** when possible; cross-reference in the specifications so the dependent-claim space spans the full portfolio.

**The eight omnibuses combined create the widest possible moat** — competitors must work around all eight to ship anything architecturally similar to the zshrs+stryke+fusevm trio. Any partial overlap (e.g., a competing language that adds `ai` builtin without the web framework) hits at least one omnibus.

**PCT international filing** within 12 months of any provisional gives priority in 150+ countries — strongly recommended given that zshrs/stryke have global use cases.

**Defensive publication** (IP.com, Defensive Publications repository) for items not patented but where blocking others is desirable — free, locks prior art.

**Public-disclosure clock awareness:** US gives a 1-year grace period from public disclosure; EU and most international jurisdictions bar patent rights from the moment of public disclosure. Anything already publicly committed to the zshrs/stryke/fusevm git repos started a 1-year US clock and may have already barred international rights. **Audit first-public-disclosure dates per omnibus claim element** before deciding which claims still have international priority headroom.

---

## Three Structural Moats — Summary

1. **Three-axis universal-access protocol** (Patent D, claim #16) — pipeline substrate covers (callable × value-class × reflection-metadata) cross-product with zero categorical exclusion.
2. **20,000-test Perl-parity corpus** (Patent D, claim #20) — empirical specification of `--compat` mode behavior; competitors must reproduce 20,000 tests passing to make a comparable claim.
3. **Systematic-absorption-of-userspace-tooling meta-claim** (Patent D meta + Patents F/G/H) — first to absorb 25+ categories (git, jq, parallel, visualization, crypto, stats, linalg, networking, compression, serialization, HTTP-server, SQLite, DataFrame, PDF, testing, file-watcher, profiler, formatter, LSP, REPL, dep-manager, docs, AOP, AI primitives, web framework, package manager) as core builtin verbs and language subsystems.

Each moat is independently load-bearing; combined they defeat design-arounds across categorical (#16), empirical (#20), and architectural (meta-claim) axes simultaneously.

---

## How to Apply This Memory

- Treat each omnibus (A through H) as **independently assertible**. Engineering proposals affecting any one are patent-relevant per that omnibus's dependent-claim space.
- New `z*` builtins → dependent claims under B (zshrs daemon).
- New JIT tier optimizations or runtime architectural changes → dependent claims under C (fusevm).
- New language-level syntactic primitives → dependent claims under D.
- New AOT-binary capabilities → dependent claims under A.
- New cluster/agent/controller/stress-testing capabilities → dependent claims under E.
- New AI primitives, MCP features, tool-fn extensions → dependent claims under F.
- New web framework capabilities (routing DSL extensions, runtime improvements, deploy enhancements) → dependent claims under G.
- New package-manager capabilities → dependent claims under H.
- When a feature spans multiple omnibuses (e.g., a new parallel primitive that's both a fusevm runtime feature AND a stryke language builtin), file under both as cross-claim dependent material.
- Engineering analysis here; legal claims belong to a real patent attorney specializing in compiler/runtime/language patents.

---

**Canonical source-of-truth:** `/Users/wizard/.claude/projects/-Users-wizard-RustroverProjects-zshrs/memory/aot_patent_strategy.md` (zshrs-project memory). This file is the strykelang-repo slice, kept in sync with the canonical when significant updates land. If conflicts exist, the canonical wins; reconcile here.