PAPERS

One size pass, one resize, then a write pass that never checks capacity. Every byte the schema already knows — keys, quotes, colons, commas, indentation — is packed into integer constants at compile time, and nearly every copy is a compile-time power-of-2 size that lowers to a single load and store. Covers branchless bools (one subtraction, one 8-byte store), integers written 4 digits per store from compile-time tables with division replaced by multiplies, and SIMD-then-SWAR string escaping. Results across five platform/compiler builds: fastest in 111 of 115 converged tests against Glaze and simdjson's C++26 reflection writer, with full-document leads of 1.1–1.7× minified and about 3× prettified on the closest build.

A schema-free, tape-driven parser shaped like simdjson's On Demand API, but re-accessible. In-order reads cost a cursor increment and a key compare; out-of-order reads go through a SIMD SwissTable-style field index that is only built once an object has skipped enough keys to earn it. Covers the shared forward cursor, zero-copy strings with a fixed-slot arena, why whole-document indexing made the target workload 2.3x slower, and the per-platform threshold sweep. Results across five platform/compiler builds: 289 wins, 15 ties, 44 losses against simdjson On Demand, including 69 wins and 2 losses in 76 full-document reverse-order tests, across in-order, reverse, sparse and sparse-reverse reads. A new section covers multi-document streams through iterateMany, the counterpart of simdjson's iterate_many: 84 wins, 4 ties, 9 losses across NDJSON and comma-separated streams, including 22 wins in 25 results on ports of simdjson's own streaming benchmarks and 28 wins in 32 reverse-order stream results.

Two-stage SIMD parsing is a specialized tool, not a mandatory front door. Full-document parses into reflected types skip the structural tape entirely and take a single fused pass; the tape is reserved for partial reading, prettifying, and minifying — where it actually pays for itself. Covers the per-(ISA × compiler)-tuned step geometry, the folded stepped drain, and a distributed UTF-8 validation scheme that costs zero additional memory bandwidth. Direct comparisons to simdjson throughout, including the honest finding that both libraries converge on the same tzcnt-burst extraction primitive.

Six architectural divergences from simdjson's stage-1 structural indexing: a batched drain that decouples mask production from tape emission, per-(ISA × compiler)-tuned step geometry, UTF-8 validation fused into string unescaping and carried across SIMD width transitions, and a compile-time minified specialization. Includes a case study on relocating control-character validation from stage 1 to stage 2, and full cross-platform results — 99 wins, 1 tie, 12 losses against simdjson across 112 AVX2/NEON tests, with a perfect 23-0-0 sweep under MSVC, plus 56-1-12 across 69 tests on three AVX-512 targets.

A C++20 configuration pattern where settings are passed in any order, duplicates are a build error rather than a silent overwrite, and the entire machinery erases completely under consteval — field values appear as immediates in the output. Includes the one subtlety that makes or breaks the guarantee: why the uniqueness concept must constrain the whole argument pack, not each argument individually.

A filtered, tag-dispatched aggregate over heterogeneous types with per-component boilerplate reduced to a single inheritance line. Dispatch overhead is zero — verified by assembly diff against a hand-written baseline at -O1 and above. Includes compile-time scaling measurements across GCC, Clang, and MSVC, and an honest account of where the pattern degrades and when a plain std::tuple is the better tool.

Removing hardware IDIV from indexing-bound kernels by hoisting divisors into the type system. Measured 12–30× operation-level speedup on RTX 5070 Ti across 16 divisors at 32-bit and 64-bit widths. Includes the honest finding that all three magic-number delivery mechanisms are statistically tied — the win is Granlund–Montgomery itself, not the compile-time encoding.

Cathedral Architecture — the methodology these papers apply.

The libraries the methods are tested against.