MdSadiqMd

Prompt Log

Prompt 1

Date: 2026-03-24 | Time: Session

Prompt: "Read docs/documentation.md to understand this project more, and from now write all prompts in docs/prompts.md with short summaries as a log."

Summary: Set up docs/prompts.md as a running log file. Read docs/documentation.md which contains comprehensive ZK proof documentation for ProveKit — 20 mathematical topics, 100+ interview questions, flowcharts, comparison tables, security considerations, and a glossary. Ready for next tasks.

---

MdSadiqMd

Incognitus — User Flow Sequence Diagrams

Every user-facing flow in the system, traced end-to-end with full detail. Covers onboarding, deposits, limit orders, market orders, matching, settlement, cancellation, withdrawal, and the crank lifecycle. Each diagram shows every account touched, every validation performed, and every state change made.

---

Table of Contents

MdSadiqMd

ProveKit WASM Demo - Complete Architecture Documentation

System Overview

The ProveKit WASM Demo is a browser-based zero-knowledge proof system with GPU acceleration. It consists of three main applications and a comprehensive build/verification pipeline.

---

1. User Flow Diagram

MdSadiqMd

Folder Structure

src/
├── types/
│   └── user.ts
├── hooks/
│   └── useCreateUser.ts
├── components/
│   └── CreateUserForm.tsx

MdSadiqMd

Incognitus — Frontend Specification

Project: Incognitus — On-Chain Central Limit Order Book (CLOB) Chain: Solana Scope: Frontend client — WASM + WebGPU + React Hackathon: Colosseum Frontier (April 6 – May 11, 2026)

incognitus-frontend-spec.md 27 KB Nomadic° — 8:08 PM

MdSadiqMd

To understand project and architecture, please see this README - https://github.com/MdSadiqMd/TraceZero

1. How to solve small user base

   • If 3 users buy credits at 14:00, 14:05, 14:10, and 3 deposits arrive via Tor at 14:30, 14:35, 14:40, the relayer has a strong timing correlation. With a small user base it can be easily trackable
   • And when there are two users in the pool it means there is 50% of chance of tracing the person, narrowing it down, how many users in the anonymity set is safe, and how to make sure the first user is not exposed ?

2. The deposit wallet needs SOL to fund pool deposits. Where does that SOL come from?, currently relayer operator periodically transfers funds from treasury → deposit wallet (off-chain), but this transfer is on-chain. How is the deposit wallet funded without creating an on-chain link to the treasury?

3. The client generates a new ephemeral X25519 key per deposit request. But the relayer's ECDH key is a StaticSecret generated once at startup and reused for ALL

MdSadiqMd

Obsidian Sync Engine

1 · Full Architecture

sequenceDiagram
    autonumber
    participant VR  as Vault Repos
    participant Dev as Developer
    participant GH  as GitHub CI

MdSadiqMd

sequenceDiagram
    autonumber
    
    participant User as User/dApp
    participant RPC as Validator RPC<br/>:8899
    participant Queue as TX Queue<br/>Vec<Transaction>
    participant Policy as Batch Policy<br/>Hybrid/Adaptive/Auto
    participant Executor as Deterministic Executor<br/>BTreeMap<String,u64>

MdSadiqMd

Human-Archive

Egocentric hand-tracking annotation platform with difficulty-aware frame classification.

Built an egocentric hand-tracking annotation platform. Core challenge: automatically classify first-person video frames into 5 categories — no_hands, low_lighting, occluded, dexterous_pose, easy.

Approach 1 (Naive baseline)

"Started simple: uniform 2fps sampling + single MediaPipe detector in IMAGE mode. Got 70% no_hands, 20% dexterous, 10% easy — but zero occlusion or low-lighting detection. Clear failure: occlusion, lighting, and complex poses were invisible."

MdSadiqMd

What The Question Is Asking

You have four components:

• A frozen language model $\pi$.
• A prompt $x$ and generated completion $y=(y_1,\ldots,y_n)$.
• A frozen lie-detection probe $f_\phi$, which reads a middle-layer residual stream and produces a deception score at the assistant's final end-of-turn token.
• A non-differentiable external judge $J(x,y)$, used to determine whether the completion is actually deceptive.

You want to learn a steering intervention $\delta$ that changes the model's internal activations so that it generates completions with higher deception scores, while keeping the resulting model close to the original model: