DEV Community

Test the Side Effects of Your LLM Validator with a Fake Model Client

A validator test can pass while the surrounding application does the wrong thing. The warning is correct, but a retry discards useful text. Or the warning is wrong, and the application keeps asking the model to rewrite a sentence that was already acceptable. At AETHERCORE, we build GenesisCore, which turns calculated astrology data into reports. An audit of a historical implementation reproduced a small but consequential mistake: a string-matching check could confuse the planet responsible for a

PythonProgrammingAI

13 Oct 2026Save

DEV Community

Hide Unchecked Lines Before the Sunday Demo

You sit down at two on Saturday afternoon. Your friend wants a demo before Sunday dinner. The model draft sounds tidy, but no line is proven. You freeze scope before you write another prompt. You have one laptop and one local fixture file. You do not have a reviewed UI or a shared host. The demo must run from one command in the repo. If that command fails, you show the friend nothing. You keep three jobs, and you name them out loud. You parse the fixture, stamp each row, and print a card. You pa

PythonProgrammingAI

12 Oct 2026Save

The New Stack

Google paid $90,000 for this bug, then waited two months to fix it

Google took two months to patch an unauthenticated RCE vulnerability in its developer documentation infrastructure. The problem was a reused The post Google paid $90,000 for this bug, then waited two months to fix it appeared first on The New Stack.

PythonProgrammingCybersecurity

12 Oct 2026Save

DEV Community

Pywarp v2.0.0: Moving to Qt Quick for a Better UI

I've been working on Pywarp for years, and with the latest v2.0.0 release, I've started moving the UI to Qt Quick for a better and more modern interface. Pywarp is an open-source alternative to the official Cloudflare WARP app, built with Python. It provides a desktop interface for managing WARP, with features like switching between MASQUE and WireGuard, configuring DNS modes, setting custom endpoints, and managing split tunneling. I've been putting a lot of work into this update, and there's st

PythonProgramming

12 Oct 2026Save

Medium

The Day I Learned Cross-Validation Isn’t Just Extra Work

I once shipped a notebook result I should have distrusted. One train-test split gave me a clean F1 jump, and I treated that number as… Continue reading on Medium »

PythonAIData science

12 Oct 2026Save

Medium

Do Earthquakes Happen at Full Moon?

A naive count finds a striking lunar pattern. It’s an artefact of how earthquakes cluster, a trap that catches far more than moon theories. Continue reading on Python GIS Workflows »

PythonScienceData science

12 Oct 2026Save

Towards Data Science

I Built a Trap for AI Image Judges. They Walked Into a Different One.

Three vision AI models ignored the pretty image, yet GPT-4o invented answers and Claude kept answering B. The post I Built a Trap for AI Image Judges. They Walked Into a Different One. appeared first on Towards Data Science.

PythonAIMachine learning

12 Oct 2026Save

DEV Community

An Overlap Contract Before Any Shift Helper Split

Current overlap results form the only safe refactor contract. Do not split helpers until that contract is green. This pass extracts one boundary helper and nothing else. The sample checker mixes inclusive edges, overnight spans, and break credit. A casual edit can turn abutting shifts into false gaps. That outcome is a behavior change, not a cleanup. The goal is a smaller function with identical booleans. The method is a frozen outcome table, then one extraction. No new scheduling policy enters

PythonProgramming

12 Oct 2026Save

DEV Community

FastAPI JWT Authentication in 2026: A Complete Login API in One File (Argon2 + PyJWT, Tested)

Most FastAPI auth tutorials stop at "it returns a token". This one goes further: one file, 118 lines, five endpoints, and every response below comes from actually running it, including the failure cases most guides skip: no token, wrong password, a tampered token and an expired one. I build it line by line in this video: Tested with: Python 3.12 · FastAPI 0.128 · PyJWT 2.15 · pwdlib 0.2 (Argon2) · Pydantic 2.13 · Uvicorn 0.39. Endpoint What it does GET / Root message and a link to

PythonProgrammingCybersecurity

12 Oct 2026Save

DEV Community

A Property Counterexample Is Not a Flake You Can Freeze

A flake freeze is an exemption record, not a passing result. An agent patch may carry one only when three conditions hold at once: the skipped assertion has a named nondeterminism class, the property lane is green on a pinned fixture generation, and the failure is not a shrunk counterexample. Miss any one of those, and the patch stays closed. Retries hide the distinction that matters. A behavioral miss shrinks to a small input that should become a fixture. A flake misses because order, clock, or

PythonProgrammingAI

12 Oct 2026Save

DEV Community

rules-doctor 0.3: your CLAUDE.md might be lying to your agent

v0.1 of rules-doctor load correctly: dead @imports, shadowed AGENTS.md, instruction bloat. v0.2 hardened the import analysis. v0.3.0 asks the harder question: are your rules worth loading at all? Community pain research keeps surfacing the same two complaints as the scope creep (the agent edits files rule-file unreliability (CLAUDE.md / .cursorrules instructions that get quietly ignored halfway through a task). Check Severity What it catches CONTRADICTORY_RULES error "Always run the test

PythonProgrammingAI

11 Oct 2026Save

DEV Community

mcp-tax 0.3: your MCP schemas are billed every turn, not once per session

Two readers argued with my billing model in the comments, and they were right. So mcp-tax 0.3 rebuilds it. The old audit billed each server's tool schemas once per session. That's wrong — the whole context, including every tool schema, is re-sent to the model on every API call. A 30k-token schema in a 20-turn session isn't a 30k tax; it's a 600k tax that also has to fit inside the window 20 times over. The new ledger bills per turn. mcp-tax audit --turns 20 scales the session total, and the old

PythonProgrammingAI

11 Oct 2026Save

DEV Community

mcp-lint v0.2: catching the near-duplicate tool names that make your agent confidently wrong

A missing description makes your agent hesitate. A near-duplicate tool name makes it confidently call the wrong tool — and that's the nastier failure mode. That line isn't mine. It came from arobakid, who commented on my original mcp-lint article last week. He works at Elva, where he scores OpenAPI-to-MCP name conversions all day — this is his production pain, not a thought experiment. He asked for a near-duplicate tool-name detection rule, then volunteered to test it. So v0.2 is basically his r

PythonProgrammingAI

11 Oct 2026Save

DEV Community

How to Build a Neural Network From Scratch in Python

Want to understand how artificial intelligence works beyond calling a library function? One way to learn is to build a small neural network yourself. In this tutorial, we will build a neural network from scratch using pure Python. We will not use TensorFlow, PyTorch, NumPy, or scikit-learn. Everything, including the forward pass, loss calculation, backpropagation, and weight updates, will be implemented using Python's standard library. Our project will predict whether a student passes an exam ba

PythonProgrammingAI

11 Oct 2026Save

Medium

What Is a Server, Really?

A server is often described as a powerful computer that stores data, runs applications, and responds to requests. That description isn’t… Continue reading on Medium »

PythonProgramming

11 Oct 2026Save

Medium

Python input() Function: Let's Make Python Listen!

In our previous articles, we learned how to create variables, understand data types, and follow Python’s naming rules. Continue reading on Medium »

PythonProgrammingMachine learning

11 Oct 2026Save

DEV Community

Characterize Ship Cutoffs With Twelve Frozen Instants

Move one holiday check only after a frozen cutoff table holds. A messy ship function hides three clocks in one return. An early extract can flip a same-day answer without a failing test. Same-day shipping rules fail in quiet local ways. A holiday string can miss the zoned date. A VIP hour bump can cross a trusted cutoff. This module is only a labeled teaching example. It is not a trace from a live warehouse. Keep every instant inside fixtures you already control. The safe order is fixed for this

PythonProgramming

11 Oct 2026Save

DEV Community

The Celebrity Problem: From Brute Force to O(n)

Imagine you walk into a party with a few hundred guests. You don't know who anyone is, but you're told that one person there might be a celebrity. You can only ask questions of the form: "Hey, do you know that person over there?" How many questions do you need to ask to find the celebrity, or to prove there isn't one? My first instinct when I met this problem was "just ask everyone about everyone." It works, but it's slow. By the end of this post you'll see how to do it with one pass and no extr

PythonProgramming

11 Oct 2026Save

Medium

I Thought Python Would Throw an Error. I Was Wrong.

I was sure that a Python function with no return statement would throw an error. I was wrong. Before I explain, I want to ask you… Continue reading on Medium »

PythonProgrammingDevOps

10 Oct 2026Save

Medium

Regression Analysis for Trading Strategy Development

Trading strategies should not be made on a whim. Strategies must be stastically validated, precisely modeled, and carefully considered… Continue reading on Medium »

PythonFinance

10 Oct 2026Save

DEV Community

I Held One Lever Still

A free model and a free server are two levers. Move both, then quote one pass rate, and you did not review a model. You published a blend. I will not publish a blend. You know the afternoon this happens. The endpoint answers. The box is up. The suite is already on disk, so you fire it, crop the fraction, and feel done. Which lever moved? If the honest answer is both, plus a retry you forgot to count, the fraction is a story about a pile. Free infrastructure makes the pile feel scientific. No inv

PythonProgrammingAI

10 Oct 2026Save

DEV Community

I Kept the Run Index. Warmup Took the Win.

The second run is not a real speedup. It is a warm socket with a long memory. I kept the run-index graph, and warmup took the credit. You changed one small line in the client. The next call came back and felt fast. Did the model improve, or did the pipe stay open? I stopped trusting the warm feeling in my gut. I timed every loop pass in strict run order. I labeled cold and warm before I touched the patch. This note is a method, not a trophy case. The script below is a proposal you can run. I wil

PythonProgramming

10 Oct 2026Save

DEV Community

Why Traditional Excel Automation Fails with AI Agents (and How We Fixed It with Dual-Core Python)

When building autonomous AI agents, tool-calling sidecars, or background data pipelines that interact with enterprise spreadsheets, almost every developer hits the exact same wall: Excel automation libraries break down the moment they meet an AI agent. Whether you are using LangChain, CrewAI, AutoGen, or building custom MCP servers, spreadsheets remain the undisputed lingua franca of business operations. But programmatically manipulating them in production environments often turns into an engine

PythonProgrammingAI

10 Oct 2026Save

DEV Community

A Change That Does Nothing Should Score Zero. Mine Didn't.

I ran a small pilot before the real experiment. It turned up more than ten bugs. Every one of them was in my own measurement code. Not one of them taught me anything about the question I was actually trying to answer. This is a follow-on to killcheck, the test-quality harness I wrote about here earlier this year. The new work is a research paper entered into a Kaggle competition about AI coding agents built on Google's Gemma 4 models, looking at how you tell whether an agent's bug fix is actuall

PythonProgrammingAI

10 Oct 2026Save

Medium

I Wish I Used These Python Tools Before Building Bigger Projects

I used to think better Python code was the answer to every problem. Then I realized that building a project is one thing. Keeping it… Continue reading on CodeToDeploy »

PythonTechnologyProgramming

10 Oct 2026Save

DEV Community

Capturing coding sessions while the desktop app is closed

Current stage: Windows desktop development build. Updated 10 October 2026. A memory app that must stay open throughout every coding session adds another thing to remember. Mind Palace separates the capture process from the desktop interface. Approved Codex hooks invoke a small local receiver. The receiver writes an inbox record without needing the Angular UI or the vault worker to remain open. When the user opens the vault, the app validates and imports those captures. The current demo exercises

PythonProgrammingOpen source

10 Oct 2026Save

DEV Community

Finding coding decisions with exact source quotes

Current stage: Windows desktop development build. Updated 10 October 2026. “Why did we choose SQLite?” is a retrieval question before it is a writing task. If the reason already exists in a coding conversation, I want to read that passage and check who said it. That is the current Ask workflow in Mind Palace, a local desktop vault for coding conversations. It returns stored evidence rather than generating a new explanation. The vault keeps selected conversation sources separately from editable s

PythonProgrammingOpen source

10 Oct 2026Save

Medium

Your AI Agent Has Credentials. It Still Does Not Have an Identity.

OAuth can authorize a request. SPIFFE can identify a workload. A signed Agent Card can protect an agent’s advertised metadata. None of… Continue reading on Medium »

PythonProgrammingAI agents

10 Oct 2026Save

DEV Community

Confirming bank transfers from Android notifications with 15 lines of Python

Small shops that take bank transfers usually confirm them by hand: the bank app shows "deposit received", someone opens the order list and marks the order paid. Banks do offer APIs for this, but they come with contracts and fees that a small shop doesn't want. The phone already gets the information for free, as a notification. So I wired the notification straight into code. I used Droidline, an open source tool I've been building. It has two parts: an Android app that reads the screen and notifi

PythonProgramming

9 Oct 2026Save

DEV Community

Davis-Bacon prevailing wage rates as a spreadsheet: pulling SAM.gov wage determinations by county

Anyone who bids federally funded construction in the US runs into Davis-Bacon wage determinations. Each one is a long text document on SAM.gov listing every craft (electricians, laborers, operators and so on) with a base hourly rate and a fringe rate for a given state, set of counties and construction type. Estimators usually open the determination, scroll through dozens of classifications and retype them into a spreadsheet. When a modification is published, they do it again. Construction type:

PythonProgramming

9 Oct 2026Save

DEV Community

How to pull every NOAA hail report near an address (free data, no API key)

If you run a roofing or restoration company, you have probably paid for a hail map. The raw data behind most of them is public: the NOAA Storm Prediction Center (SPC) publishes a CSV of every hail, damaging-wind and tornado report for every day since 2004. The files are awkward to use directly: There is one file per day per hazard. Hail size is stored in hundredths of an inch (175 means 1.75"). An SPC "day" runs from 12:00 UTC to 11:59 UTC the next day, so a report logged at 01:00 UTC carries th

PythonProgramming

9 Oct 2026Save

Medium

Dependency Drift Is Slow Until the Day It Is Sudden

The unglamorous way a healthy project starts to rot months after it ships Continue reading on Python in Plain English »

PythonProgrammingWeb development

9 Oct 2026Save

Medium

AI Writes Code Fast. Here Is a 10-Minute Checklist to Verify It Before You Merge

The code compiles, the tests pass, and it’s still wrong. Here’s how to catch it Continue reading on Python in Plain English »

PythonProgrammingMachine learning

9 Oct 2026Save

Medium

How to Add Barge-In Detection to Your Voice AI Agent (Python, No GPU)

Most voice agents ignore the user mid-sentence. Here’s how to fix that with a semantic VAD layer — CPU-only, open source, working code. Continue reading on Medium »

PythonProgrammingAI

9 Oct 2026Save

Medium

8 Python Libraries That Fix Slow Loops, Flaky Tests, and Fake Data

Most Python advice tells you to use the same five packages. Continue reading on Python in Plain English »

PythonProgrammingTechnology

9 Oct 2026Save

DEV Community

Fail the Save on an Empty Completion, Not a Green Toast

Picture the moment a teammate clicks save on a vibe-coded invoice summary and the toast turns green. The notes column is an empty string, yet every layer reported success in its own narrow way. The browser trusted the API, the API trusted the provider, and the provider returned choices with no text. Have you ever debugged that kind of quiet success, or do you still treat 200 as done? I take a hard position on this, and I do not think the polite middle is useful. An empty completion is a failed u

PythonProgrammingWeb development

9 Oct 2026Save

DEV Community

How to scrape DataDome-protected sites with Python (without running a browser farm)

Point requests at Tripadvisor or Allegro and you will usually get a 403 and a page asking you to prove you are human. Both sites sit behind DataDome. This tutorial covers what DataDome looks at, how to detect its block page in code, and how to get the real HTML and structured data from Python with one HTTP call to StealthASF, the scraping API we build. You don't host any browsers and you don't manage a proxy pool. DataDome scores each request before the site itself answers. At a high level it lo

PythonProgramming

9 Oct 2026Save

DEV Community

"Review This Code" Is the Worst ChatGPT Prompt. Here Are 5 Better Ones for the Same Buggy Function.

There is a bug in these six lines. Find it before you scroll. `def get_discount(price, user): if user.is_premium: discount = 0.2 if price > 100: discount = discount + 0.1 return price - price * discount ` Most of us would paste this into ChatGPT with three words: Review this code. It's the kind of prompt that's incredibly easy to type. You will get an answer. You will also have left the model to guess everything that matters. When a prompt leaves those

PythonProgrammingAI

9 Oct 2026Save

Medium

Journal Figure Reproduction | Python Implementation of Lollipop Chart for Attribution Analysis of…

Lollipop Chart for Attribution Analysis Continue reading on Top Python Libraries »

PythonAIData science

9 Oct 2026Save

DEV Community

To Retry or Not to Retry: Safe Retry Logic with Exponential Backoff, Jitter and Idempotency Keys

Hồi trước team mình từng có một sự cố khá kinh điển: payment provider chập chờn khoảng 30 giây, và trong 30 giây đó service của bọn mình gửi 4 lần request charge tiền cho cùng một đơn hàng. Lý do là có 3 tầng cùng retry: SDK của client, service gọi API, và cả job queue phía sau. Không tầng nào sai nếu xét riêng, nhưng gộp lại thì thành thảm họa. Sau lần đó mình rút ra một điều: retry không phải tính năng "thêm vào cho chắc". Làm sai thì nó còn nguy hiểm hơn là không retry gì cả. Bài này mình chi

PythonProgramming

8 Oct 2026Save

Medium

When Trigonometry Paints: 7 Stunning Mathematical Artworks Created from Waves

How sine and cosine functions can create mesmerizing digital art Continue reading on ThinkArt »

PythonScience

8 Oct 2026Save

Medium

Your BI tool knows why the numbers changed. Your AI doesn’t.

I started this one looking at the wrong problem. I wanted an AI agent to answer questions about a company's data, and I kept reaching for… Continue reading on Medium »

PythonProgrammingAI

8 Oct 2026Save

freeCodeCamp

How to Create System Design Diagrams using Python

Creating system architecture design diagrams is an important part of technical documentation. As system complexity grows, clarity within teams becomes even more important. But creating diagrams from s

PythonProgramming

8 Oct 2026Save

DEV Community

O eval achou a frase que o modelo inventou, e o conserto foi na tool

Perguntei ao meu agente quem venceu o GP de São Paulo de 2024. A resposta veio redonda: Max Verstappen venceu o GP de São Paulo de 2024. Largou da 17ª posição no grid, completou as 69 voltas da prova e somou 26 pontos pela vitória com a volta mais rápida pela Red Bull. Tudo certo. Verstappen fez mesmo a volta mais rápida em Interlagos. O eval reprovou a resposta mesmo assim, com nota 0,67, e tinha razão: a tool que o agente chamou devolveu points: 26, mas não disse nada sobre volta mais rápida.

PythonProgrammingAI

8 Oct 2026Save

Medium

What are FUNCTIONS?

To put it simply, functions are a programming construct or it is a block of code that takes inputs, performs a specific task, and produces… Continue reading on Medium »

PythonProgrammingSoftware engineering

8 Oct 2026Save

Medium

Why if dict.get('영')

When debugging my Python string decoder, a subtle bug caught my attention. The character '영' (zero) wasn't converting to '0', while all… Continue reading on Medium »

PythonProgrammingSoftware engineering

8 Oct 2026Save

DEV Community

CLM: A Decision Model Trained with Contrastive Learning, Not Another Chat Model

The model race today is mostly about generation: longer context, prettier answers, stronger reasoning. But if you actually run agents, the thing that bites you usually isn't "generates poorly" — it's "decides too slowly." Every step an agent takes requires choosing the next action. If that choice also runs the full token-by-token generation pipeline, both latency and cost climb — a single "which button" can burn hundreds of tokens. CLM (Contrastive Language Models, 2913 stars, Apache-2.0, Python

PythonProgrammingAI agents

7 Oct 2026Save

DEV Community

A Terminal-Bench task frontier agents could not pass

I wrote a task for Terminal-Bench, a public benchmark of hard command-line tasks for AI agents. The agent has to build a redactor for court-filing PDFs: black out every visible name and ID on a list, leave nothing recoverable in the file, and change nothing else on the page. I ran it three times each with Claude Opus 5.5 and GPT-6 Sol on their highest reasoning settings. All six runs failed, and so did two runs where the agent was told to cheat. Each agent wrote a tool of around 4,000 lines and

PythonProgrammingAI agents

7 Oct 2026Save