Chad Brewbaker chadbrewbaker

WHATWG IDNA conformance findings in Ada URL

This review targets only observable compliance with the WHATWG URL parser and Unicode UTS #46. It does not assume that Ada is deployed, and it does not depend on an exploit or production impact.

Tested Ada commit: 4eee4c204d4581518e3d034ef899560ca23d3850.

Executive result

Ada's ordinary URL state machine performed very well. It matched the maintained whatwg-url reference on 101,822 generated parsing cases and 102,880 setter operations.

Grounding Exercise: M5 Gravity — GW170817 Gate + Poland Fiber Program

Cell under grounding: FORCES / Gravity / Liquid Crystal (M5) in MODELS.md. Current status: ⚠️ — coupling mechanism measured (GEM ∝ (b·g)², zero at zero boost, m5_8_2q_delta_scaling.py), dynamical metric not implemented; the de Broglie clock sector IS implemented and validated. This document: the GW170817 radiative gate (§0–6), the terrestrial extension — desk gates on existing data plus the Poland fiber experiments (§7–10) — and the priority logic that orders them by how much model-unknown each uniquely grounds.

The Manifesto Does Not Typecheck

A Reading of Marx and Engels (1848) in the Category of Combinatorial Species

with an appendix of compiler diagnostics and the author's marginalia

Abstract

Chapter N — Certified Resource Bounds for a UTF-8 Transcoder

Draft chapter in thesis register. Numbers measured on a single VM (Xeon 2.80 GHz, 1 core, 4 GB, GCC -O2); every claim is labelled proved, measured, or conjectured. Citation keys marked [verify] should be checked against the published record before submission.

Segment-wise SIMD shuffle under a variable permutation

Problem. Given a 64-byte vector v (one cache line, one zmm), a 64-bit boundary mask m (bit i set ⇔ a segment starts at byte i; bit 0 always set; k = popcount(m) segments), and a variable permutation π of {0..k-1}: produce the segments of v concatenated in the order π[0], π[1], …, π[k-1]. Both m and π vary per call — nothing can be precompiled into shuffle constants.

Hardware. Intel Xeon, family 6 model 207 = Emerald Rapids

Oracle 1: Perfect Retrieval Suppose an oracle answers every factual question instantly. The model never forgets anything. Question: Does theorem proving become easy? If not, then factual knowledge wasn't the issue. This separates memory from reasoning. Oracle 2: Perfect Search Suppose an oracle instantly returns the best next action among all possibilities. Question:

uv run profile.py ~/.hermes/logs/agent.log

	#!/usr/bin/env python3
	# /// script
	# requires-python = ">=3.11"
	# dependencies = []
	# ///
	"""Time travel: SQLite patterns vs Okasaki-style persistence. Value-checked.

	Three ways to get "query the table AS OF transaction T":

	SQLITE-AUDIT history table kvh(k, tx, v) PK(k, tx DESC), in-memory,

	import torch
	import torch.nn as nn
	from tqdm import tqdm

	# --- CONFIG ---
	DIM = 2048
	BLOCK_SIZE = 16
	LR = 0.01
	ITERATIONS = 100

	./llama-server \
	-m gemma-4-12B-it-Q4_K_M.gguf \ # Target model
	-md gemma-4-12B-it-assistant-Q5_K_M.gguf \ # MTP drafter (small ~0.4B)
	--spec-type draft-mtp \ # Enable MTP speculative decoding
	--spec-draft-n-max 4 \ # Typically 3-5 for Gemma 4 MTP (experiment)
	--spec-draft-n-min 1 \
	-c 131072 \ # Context (try 262144 if you have memory)
	--cache-type-k q4_0 \ # 4-bit KV cache
	--cache-type-v q4_0 \
	-ngl 99 \ # Offload all layers to GPU