OhadRubin

A survey of caching stances across reference systems

Nine entries, each from a different family. Close cousins are collapsed: React Query stands in for SWR and Relay, Nix stands in for Bazel and Buck, Dagster stands in for Luigi and Prefect and Airflow. Content-addressed storage (Git, OCI, IPFS) is folded into Nix because for a subsystem designer it's the derivation stance that's instructive; pure object dedup is a weaker commitment.

---

1. Spark — lineage-based recovery with opt-in materialization (also Flink, Beam)

What you reach for: df.cache() or df.persist(StorageLevel.MEMORY_AND_DISK).

OhadRubin

V3 Replication Status: BN vs Direct IC Comparison

What Exists in v3

PICon Evaluation Infrastructure — FULLY PRESENT

• picon_env_v3.py: Complete PICon Internal Consistency environment.
  • PICONInternalConsistencyEnv class extending ChatEnv (multi-turn adversarial interrogation)
  • Questioner system prompt (AVH Forensic Interrogator style)
  • Internal consistency evaluator system prompt (conflict vs plausible per turn)
  • score_transcript() with parallel conflict judging via asyncio.gather

OhadRubin

From 848d3f5d4c2aa5bee58eecad8421b5947ec79334 Mon Sep 17 00:00:00 2001
From: catswe <212922539+catswe@users.noreply.github.com>
Date: Sat, 17 Jan 2026 18:36:09 -0600
Subject: [PATCH 1/9] Add prompt_logprobs

Signed-off-by: catswe <212922539+catswe@users.noreply.github.com>
---
 tpu_inference/runner/tpu_runner.py | 101 +++++++++++++++++++++++++++--
 1 file changed, 94 insertions(+), 7 deletions(-)

OhadRubin

# /// script
# dependencies = [
#   "claude-agent-sdk",
# ]
# ///

from claude_agent_sdk import (
    ClaudeSDKClient,
    ClaudeAgentOptions,
    AssistantMessage,

OhadRubin

# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "slack-bolt",
#     "slack-sdk",
#     "claude-agent-sdk",
#     "aiohttp",
#     "tqdm",
# ]
# ///

OhadRubin

HOWTO: Save a Running Process You Forgot to Start in tmux

You started a long-running job and forgot to use tmux. Now you're stuck. Here's how to save it.

Option 1: Detach with disown (simplest, lose output)

# In the terminal where the process is running:
Ctrl+Z              # Suspend the process
bg # Resume in background

OhadRubin

#!/usr/bin/env bash

json_template_to_canon() {
  # stdin: template with ${VARS}; output: canonical (minified) json
  # Substitutes environment variables in a JSON template.
  #
  # NOTE: Variables MUST be exported (not just set) for this to work.
  # NOTE: Unset variables are left as-is (safe_substitute behavior).
  #
  # Example 1 - single line:

OhadRubin

# /// script
# dependencies = [
#   "ast-grep-py",
# ]
# ///

import sys
from pathlib import Path
from ast_grep_py import SgRoot

OhadRubin

#!/usr/bin/env bash
set -euo pipefail

# gcloud auth login --enable-gdrive-access
# Usage: ./md2drive <markdown_file>
# Example: ./md2drive example.md
# Requires: export FOLDER_ID=<your_google_drive_folder_id>

MD_FILE="${1:?usage: $0 path/to/file.md}"
# Check if markdown file exists

OhadRubin

Create a documentation file that explains code in natural language:

Part 1: Read the input file {filename} and create a new markdown file at docs/{filename}.md

Part 2: Create an initial todo list (let's call it list A) of functions/sections to document, give an complexity rating to each task.

Part 3: Group simple items together into single tasks.

• This will ensure tasks have similar complexity levels.
• Each item should require a similar amount of time to "think about"
• This will result in list B that has a uniform level of complexity for each item.