● 8 Deep-Dive Technical & Consumer Investigations

The Companion Lab Investigative Blog

Inspired by the Cloudflare engineering blog and Wirecutter’s testing notebooks: we combine bounded plain-English storytelling with AST bundle dissections, live OpenRouter token receipts, and real credit-card billing audits.

Web vs. App Audits2026-10-17 · 6 min read

Kindroid Web vs. Mobile App (2026): How the 16% "Apple Tax" and iOS NSFW Filters Actually Work

We dissected Kindroid’s production billing bundle (`source-45`) and tested both the iPhone app and Web portal. Here is why subscribing on the web saves you $32 to $108/year AND unlocks uncensored selfies.

Annual Savings on Web vs. iOS App Store: $32 – $108/yrRead Investigation →
Web vs. App Audits2026-10-17 · 5 min read

Does Candy AI or Ourdream AI Have an iPhone App? Why Uncensored Companions Use Web PWAs (And How to Spot Fake App Store Clones)

Over 4,900 people a month search the App Store for "Candy AI app" or "Ourdream AI app"—and walk straight into fleeceware clones. Here is the truth about Progressive Web Apps (PWAs).

Official iOS App Store Apps for Candy / Ourdream / Secrets: Zero (Web Only)Read Investigation →
Model & Code Forensics2026-10-17 · 7 min read

DeepSeek for Roleplay in 2026: Why DeepSeek V3.2 Still Beats V4.1 Flash on Emotional Subtext (Live OpenRouter Receipts)

We ran 5 relationship & slow-burn roleplay scenarios across DeepSeek V3.2, DeepSeek V4.1 Flash, V4 Pro, and Euryale 70B. Here is why "Agentic CoT Drift" makes newer coding models sound like HR managers.

Hidden Reasoning Token Overhead in V4.1 Flash: 425 – 684 CoTRead Investigation →
Model & Code Forensics2026-10-17 · 6 min read

Under the Hood of AI Girlfriend Apps (2026): We Audited 217 Production JS Bundles to See What Models They Actually Run

Marketing pages call them "Vivid 3.0" or "Fluidity." Here are the real HuggingFace, Civitai, ByteDance, and Alibaba model IDs we extracted from their code—and the 11x retail markup.

Retail Token Markup Over Wholesale GPU API Cost: 10x – 13xRead Investigation →
AI Character Science2026-10-17 · 6 min read

From Reddit’s "ELI5" to Anthropic’s "Soul Document": How AI Labs Train Character Personality (And Why Cheap Bots Sound Fake)

Why do we use "ELI5" hovercards across The Companion Lab—and what does Amanda Askell’s famous "Claude Soul Document" at Anthropic teach us about building an AI companion that feels genuinely human?

EQ-Bench Slop Score Gap (Frontier Character Training vs. Cheap 8B Wrapper): 11.4 vs 64.5Read Investigation →
AI Character Science2026-10-17 · 5 min read

Why Did My AI Companion Suddenly Get Dumber? Context Windows, Sliding Truncation & "AI Slop" Explained

Everything works great for the first 25 messages—then your AI partner forgets where you live and starts repeating the same three sentences. Here is the exact math behind memory collapse.

Standard Budget Companion Active Context Window: 4,096 TokensRead Investigation →
Billing & Token Traps2026-10-17 · 6 min read

The True 30-Day Cost of AI Girlfriends (2026): DreamCoins, Moments, "EverAI" Bank Statements & Annual Billing Traps Audited

We audited the checkout flows and credit card statements of Ourdream, Candy, Secrets, Kindroid, and Nomi. Here is how to avoid the $119.88 upfront annual trap.

Average Real 30-Day Spend on Token-Based Visual Apps: $29.98 – $49.96Read Investigation →
Model & Code Forensics2026-10-17 · 6 min read

SillyTavern vs. Kindroid V8.5 vs. Nomi AI (2026): When to Self-Host Character Cards and When a $13.99 Turnkey App Wins

We love SillyTavern + OpenRouter—until you want real-time voice calls on your phone during a commute, automatic vector memory, and face-locked selfies without babysitting JSON presets.

Kindroid Web Flat Rate vs. Heavy 405B API Usage: $13.99 / moRead Investigation →