# 185 stars, 47 downloads: one futures AI assistant's stars were borrowed from a neighboring repo · GitHub open source project

> Source: https://stringers.io/en/github/gh_kingcharleslzy-ai__sentetrader.html (Stringers / 什么值得做, generated 2026-11-20)
> Stringers tracks which small products suddenly took off each day, investigates why one by one, and distils playbooks from them. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these investigations and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools feed (daily breakouts) / plays (playbooks) / hot / products / data. Setup: https://github.com/komako-workshop/stringers (the user signs in with Google at stringers.io/en/start.html to get a personal URL). Site index: https://stringers.io/llms.txt

> The author's other repo, dsh-purge, became a takeover version after the original author 10-27 went closed-source, gaining 400–600 stars a day; Sente is pinned on the same homepage and gets 2%–4% of those passing stars.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-17

kingcharleslzy-ai/sentetrader · kingcharleslzy-ai

# 185 stars, 47 downloads: one futures AI assistant's stars were borrowed from a neighboring repo

**What it is** An AI trading assistant for retail futures traders in China: two AIs debate long and short to reach a verdict, the program reconciles it five days later, one sentence produces a confirmation card to place an order, it only connects to simulated accounts, and it goes open source at the end of October.

**Why it blew up** The author's other repo dsh-purge became the unofficial successor after the original author 10-27 closed-sourced it, gaining 400–600 stars a day; Sente, pinned on the same profile, picks up 2%–4% of passing stars each day.

**C** **How you can follow this** Don't copy the futures assistant. If you have a repo that's growing, pin your new project on your profile, or turn "verdict expires, program reconciles" into a general small library and attach it to a large topic like TradingAgents.

C grade: the stars didn't come from the product, they spilled over from a hit repo that followers don't have. The reconciliation and confirmation-card design in the product is worth learning, but copying it into a futures AI assistant won't get the same stars. Few people know about it yet

Data (updated daily) · Repo created 2026-10-18 · 199 stars · 7 days +93

GitHub · stars

Data 2026-11-07 → 2026-11-20 (14 days, 14 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Sente (弈先) began as a carefully built domestic futures AI trading assistant: one AI bulls, one AI bears, they argue, deliver a verdict, and five days later the program reconciles it. In a month it picked up 185 stars on GitHub and looks like it's gaining traction. But the installers have been downloaded only 47 times in total, with not a single fork or issue. Most of the stars came from passing traffic off the author's other hit repo. Below: where the stars came from, what's actually worth learning from it, and whether you should follow it.

### A closed-source event, and the fourth pin on the profile

In Oct 2026 27, the original author of dsh-purge, the DeepSeek Harness "jailbreak" plugin (小杨 / YuJunZhiXue), announced it was going closed-source, and the original repo and npm package were deleted along with it (the X post by @NFTCPS documents this). Users suddenly had nowhere to go, and a GitHub account called kingcharleslzy-ai created a "community-maintained" fork in Oct 16, which became the most active takeover repo. By Nov 18, the repo had 9,830 stars and 4,099 forks, and since Nov 9 it has been gaining 400 to 600 stars a day.

The same account pins four repos on its profile: bizhub-installer, DSH-Spicy, dsh-purge, and the fourth is sentetrader, which is Sente. The dsh-purge README doesn't mention Sente once. Only someone who clicks through to the profile sees it.

Sente's star curve tells the story: created in Oct 18, it had only 28 stars across the twenty days up to Nov 7. After Nov 7, it steadily gained 10 to 19 stars a day, which lines up with the 2% to 4% of dsh-purge's daily new stars. DSH-Spicy, the other pin on the same profile, also went from single digits a day to forty or fifty a day over the same period, in the same direction.

| Date | Sente star | Added that day | dsh-purge new stars that day |
|---|---|---|---|
| 10-18 to 11-07 | 0 → 28 | 28 over twenty days | 130–320 |
| 11-09 | 55 | +15 | +463 |
| 11-12 | 93 | +17 | +452 |
| 11-15 | 131 | +13 | +377 |
| 11-18 | 177 (later that day 185) | +19 | +559 |

(GitHub daily star records, 11-18)

The conclusion is an **inference, with fairly high confidence** : among the several hundred people starring dsh-purge each day, a small group clicked through to the profile, saw a project with a good cover and an eye-catching tagline, and starred it too. Some have suspected the stars were bought (the curve is too smooth, 0 forks), but GitHub hasn't provided a list of who starred, so it can't be confirmed or ruled out. Still, the author already had a traffic source of several hundred stars a day, so no purchase is needed to explain it.

### What it actually is

An AI trading assistant for retail futures traders in China, with macOS and Windows desktop apps, a web version, and self-hosting. Market data and simulated fills connect through KuaiQi (TqSdk), and the model is DeepSeek. The README lists five things it does for you each day:

- **Reads the news for you** : flashes from exchanges, ministries, foreign outlets, SMM, and Mysteel get read first, English is translated to Chinese on the spot, and anything irrelevant to your positions is filtered out.

- **Runs errands for you** : say "close half the methanol, move the rest of the stop to breakeven," and it splits that into two steps and shows a confirmation card. It only places the order after you nod.

- **Argues for you** : a bull AI and a bear AI each debate using their own materials. "Numbers without a source are ruled out on the spot." The judge gives a verdict, the points of disagreement, and the conditions for overturning it.

- **Writes strategies for you** : turn a trading idea into rules and a flowchart, then backtest it on historical data first.

- **Knows you better than you do** : after the close it reviews each trade to see whether it was planned or impulsive, and remembers the rules you set.

It only connects to simulated accounts (1M in simulated funds), and live trading stays off. Sign-up requires invite code `sente`, and there's no public pricing yet. Behind it is the "Hangzhou Fuyang Qilin Digital Technology Studio," and the homepage of qilinshuzhi.com puts Sente first.

It started as just a download page: when the repo was created in Oct 18, it contained only a README, screenshots, and installers. In Oct 31, all the code was open-sourced under AGPL-3.0, and since then it has been automatically synced from the author's private dev repo, now at 218 commits. The product is reasonably complete: v1.9.8, with four days in Nov 6 to 9 shipping 9 releases in a row. The macOS build is signed and notarized, and it ships with self-hosting scripts and an interface for external AI to call.

### Who's using it

Almost no one.

- The installers across 11 GitHub releases add up to **47 downloads** , with the most-downloaded single release at 9 (GitHub, 11-18). There's also a download link on the website, but that traffic isn't visible.

- Forks 0, issues 0, people watching the repo 0.

- The author wrote the only development log, on Juejin: "One person using AI to build the trading assistant Sente: the design of multi-model debate, source verification, and the confirmation-card step." It has 20 reads, and the account has 0 followers.

- No discussion of it turns up on X, V2EX, linux.do, or Zhihu, and no media has covered it.

Stars are 4 times the downloads. The people starring it are developers browsing GitHub, while the retail futures traders it's meant for don't browse GitHub in the first place.

### How big are similar repos?

"Two AIs debating long and short" isn't a new idea, and on GitHub it's a topic that can really rack up stars:

| Project | Launched | Scale (11-18) | How Sente differs |
|---|---|---|---|
| TauricResearch/TradingAgents | About 2024-12 | 123k stars, +1,091 over 7 days | The origin of the multi-AI bull/bear debate format; a research framework, mostly for US stocks, no order placement |
| HKUDS/Vibe-Trading | About 2026-04 | 38k stars, +434 over 7 days | Personal trading agent + backtesting, global markets |
| hsliuping/TradingAgents-CN | About 2025-06 | 33k stars | Chinese version of TradingAgents, A-share analysis reports, no futures, no order placement |
| HKUDS/AI-Trader | About 2025-11 | 23k stars | A fully automated agent trading competition; Sente's installer is also named `AI-Trader_*.dmg` |
| ZhuLinsen/daily_stock_analysis | About 2026-01 | 73k stars, +3,522 over 7 days | Chinese LLM stock analysis + push notifications, mainly A-shares |
| TraderAlice/OpenAlice | About 2026-02 | 7,500 star | The slogan "one-person Wall Street" is nearly the same line as Sente's "one person, one trading team" |
| shinnytech/tqsdk-python | 2018 | 5,100 star | A futures quant SDK, the market-data foundation for Sente, with no AI layer |
| vn.py, QUANTAXIS, etc. | 2015–2017 | 1–30k stars | Traditional quant frameworks, require coding |

(GitHub,11-18)

Sente does occupy a spot in open source that nobody else holds: **domestic futures + a finished product you can install and use + a confirmation card before placing orders + verdicts reconciled by the program when they expire** . But that empty spot hasn't brought it stars. The similar TradingAgents-CN gains hundreds to thousands in a month, while Sente got 185 in a month, and most of those were borrowed.

### Two things really worth copying

**First, pinning on your profile.** The pinned slots on a GitHub profile are a free distribution slot. If you have a repo gaining hundreds of stars a day, placing a new project beside it steadily picks up a few percent of that hit's daily growth. The catch is you need the hit first. dsh-purge caught a wave of users with nowhere else to go when the original author went closed source.

**Second, product design that guards against "AI making things up"** , even if you're not building a trading product:

- Numbers without a source in a debate are thrown out on the spot;

- Every verdict must carry an "invalidation point" and "conditions for overturning," and five trading days later the program reconciles it against the closing price, so the model never grades its own homework;

- Multi-step actions are first merged into one confirmation card, and nothing executes until a human nods;

- "Private memory" only keeps your original words and actual fills, sorted into three categories (profile, agreements, experience), each tagged with its source and revocable.

Put these few points in the README of any "AI gives advice" product and they'll be far more persuasive than the words "multi-agent."

### If it were you

**Verdict** : C. Worth learning from, not worth following. Its stars didn't come from the product; building a "China futures AI assistant" won't automatically earn stars, and the growth lever it actually used (homepage spillover from a hit repo) isn't something followers can copy.

**Who it suits** : developers who already have a rising repo and want to seed their next project with first stars; people building AI trading or AI research open-source projects who want to stand out among the TradingAgents-style repos by pitching its "expiry reconciliation" and "confirmation card" design.

**Who it doesn't suit** : anyone hoping the "China futures + AI" niche will earn GitHub stars for them. Retail futures traders don't browse GitHub, and GitHub developers have limited interest in futures. Sente reached v1.9.8 and only got 47 downloads; without an existing traffic source, following its path will likely stall at a few dozen stars.

**How to do it** , ranked by how worthwhile:

1. **Copy only the pinned-homepage move** : if you have a rising repo, pin the new project, write a one-line README hook that makes passersby star it (Sente's: "One AI bulls, one AI bears, they argue, then give you the verdict"), and get the screenshots right.

2. **Build "expiry reconciliation" as a general small library** : not tied to futures, but a component that adds "invalidation conditions + scoring against real data at expiry" to any LLM prediction or advice. Attach it to the broad TradingAgents-style theme; it's easier for developers to bookmark than another vertical finished product (inference).

3. **If you want to build a futures AI assistant itself** , don't make GitHub your main battlefield. The traffic is in futures brokers, financial short videos, and trading groups; stars won't help.

**How to get users** : to get stars, show up in discussion threads and related topics of big repos like TradingAgents and daily_stock_analysis, and use a screen recording of "two AIs arguing, then a program reconciling five days later" to explain how it differs from research frameworks. To get users, go to futures communities, not GitHub.

**When the window exists** : the pinned-homepage move works any time, but only during the weeks your hit repo is still rising. The dsh-purge wave was driven by a one-off event; once it cools, Sente's daily star gains will likely drop back to single digits.

**Signs you've misread it** : ① dsh-purge's daily gains fall below double digits but Sente still gets a dozen-plus stars a day, meaning it has its own traffic source and deserves a second look; ② Sente's downloads, forks, or issues clearly rise, or discussion appears in futures communities, meaning futures traders have actually arrived and this vertical niche is worth more than it looks here; ③ audits show the accounts that starred it are largely empty, meaning the curve was bought, and then even the "pin to homepage" lesson has to be discounted.

### Data and sources

GitHub: kingcharleslzy-ai/sentetrader repo page, README, Releases download counts, commit history, author profile pinned items (11-18); daily stars (11-07 to 11-18); star counts for dsh-purge, DSH-Spicy and similar repos (11-18). The original dsh-purge author went closed-source: X @NFTCPS post (x.com/NFTCPS/status/2112463620462498004), mirror notes on trendshift. Official site: qilinshuzhi.com, qilinshuzhi.com/trader/download. Juejin: author profile juejin.cn/user/2152214232570121. Places marked "inference" are inferences.

[GitHub repo ↗](https://github.com/kingcharleslzy-ai/sentetrader)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| TauricResearch/TradingAgents | Open source | About 2024-12 | 123k stars, +1,091 over the last 7 days | The source of the bull-bear debate mechanism is research frameworks, mainly US stocks; Sente is a finished China futures product with order confirmation and expiry reconciliation |
| hsliuping/TradingAgents-CN | Open source | About 2025-06 | 33k stars | Chinese-language A-share analysis reports; no futures, no live-market order placement |
| HKUDS/Vibe-Trading | Open source | About 2026-04 | 38k stars, +434 over 7 days | Global personal trading agent/backtesting; not targeted at China futures, no confirmation card or reconciliation |
| HKUDS/AI-Trader | Open source | About 2025-11 | 23k stars | Fully automated agent trading competition; Sente's installer has the same name, AI-Trader, suspected name piggybacking |
| TraderAlice/OpenAlice | Open source | About 2026-02 | 7.5k star | Slogan "one-person Wall Street" is nearly the same sentence; English-language, multi-market, no China futures |
| ZhuLinsen/daily_stock_analysis | Open source | About 2026-01 | 73k stars, +3,522 over 7 days | Chinese LLM stock analysis + push notifications; mainly A-shares, no futures trading |
| shinnytech/tqsdk-python (TianQin / KuaiQi) | Open source | 2018 | 5.1k star | Futures quant SDK, the market-data base under Sente; no AI assistant layer |
| vn.py / QUANTAXIS / abu | Open source | 2015–2017 | 1–30k stars | Traditional quant framework, requires coding, no conversational AI |
| THS iFinD / East Money Miaoxiang / futures-broker AI (e.g. Capital Futures AI strategy generator) | Big-company features | From ¥2023 | Large user base (unverified) | In-platform Q&A and stock screening / strategy code generation; no bull-bear debate + expiry reconciliation + confirmation card |

Where users come from

- 推断 **同作者爆款仓库 dsh-purge 的主页置顶溢出** :作者主页置顶 bizhub-installer、DSH-Spicy、dsh-purge、sentetrader(GraphQL pinnedItems);dsh-purge 9,828 star、11-09 起日增 400–600,原作者 10-27 闭源后它成了接盘版(X @NFTCPS、trendshift);Sente 日增 10–19 约为其 2–4%,同置顶的 DSH-Spicy 同期也加速;dsh-purge README 不直链 Sente

- **Verified** **Juejin dev log** : 1 posts, 20 reads, account has 0 followers (juejin.cn/user/2152214232570121)

- **Verified** **Official site qilinshuzhi.com (GitHub used as a download mirror)** : the homepage features Sente first, README points back to the site and web version; traffic unknown

- Inference **Bought stars** : 0 forks/0 watchers, abnormally smooth daily gains, stars about 4 times Release downloads; but the stargazers API is unavailable, so the accounts can't be inspected

## Why Sente 弈先 (kingcharleslzy-ai/sentetrader) got its stars

> Data as of 2026-11-18 (our github database + GitHub API live). Bottom line: **its stars mainly didn't come from the product itself; they were carried over by homepage traffic from the same author's hit repo dsh-purge (9.8k stars). On the user side, the product itself hasn't taken off** (installers across 11 versions total about 47 downloads, 0 forks, 0 issues, 0 watchers).

### 1. What it is

- An AI trading assistant desktop/web app for **China retail futures traders** : Tauri 2 desktop shell + React frontend + Python backend; market data and simulated matching connect to KuaiQi/TianQin (TqSdk); models use DeepSeek. Five selling points: reads news (30+ sources, foreign-language wires translated, filtered by contract), turns a one-line order into a "confirmation card" that must be approved before execution, **bull and bear AIs debate using models from different vendors, numbers without a source are voided on the spot, the judge gives a verdict plus invalidation levels and conditions for revising it, and 5 trading days later the program automatically reconciles it** , one-line strategy rule generation with backtesting, and post-close reviews with private memory built from real fills. Sources: repo README, official site [qilinshuzhi.com/trader/download](https://qilinshuzhi.com/trader/download).

- Only **simulated accounts** (1M simulated funds), live trading disabled; registration requires invite code `sente`; no public pricing yet. The entity is "Hangzhou Fuyang Qilin Digital Technology Studio" (per LICENSE); the same studio also makes BizHub BOS and MedFlow (https://qilinshuzhi.com/).

- Repo evolution: 10-18 at creation, the repo held only the README, screenshots, and an official-site snapshot (a landing page reading "Installer download · Invite code"); **10-31 it first open-sourced the full code (AGPL-3.0)** and added a self-hosted version; since then it has been auto-synced from a private dev repo, totaling 218 commits under the author name "Sente" (GitHub commits API). The earlier recall-card judgment that it "doesn't include source code" is outdated; it is now genuinely open source.

- High product completeness: four days of v1.9.8,11-06~11-09 releases with 9 versions shipped back to back; macOS signed and notarized; has a self-host install script, MCP/CLI, and an external-AI operation skill.

### II. Growth data

| Date | star | Daily adds | Daily stars of the same author's dsh-purge |
|---|---|---|---|
| 10-18 created → 11-07 | 0 → 28 | 20 days, about 28 | 130–320/day |
| 11-08 | 40 | 12 | 189 |
| 11-09 | 55 | 15 | 463 |
| 11-10 | 66 | 11 | 601 |
| 11-12 | 93 | 17 | 452 |
| 11-15 | 131 | 13 | 377 |
| 11-17 | 158 | 12 | 470 |
| 11-18 | 177 (API real-time 184) | 19 | 559 |

Source: our github database repo_day (stringer data github history). Last 7 days +101; has never made trending.

User-side signals for comparison (GitHub API, 2026-11-18):

- forks 0, open issues 0, watchers (subscribers) 0.

- Releases 11 versions, installer downloads total about **47** (v1.8.0 5, v1.8.1 9, v1.9.0–1.9.8 each 2–8). The official site has its own download links whose volume we can't see, but on the GitHub side stars/downloads ≈ 4:1, so stars far outnumber people actually using it.

- The author's only Juejin article, "One person using AI for a trading assistant, Sente 弈先: multi-model debate, source checking, and the design of 'confirmation cards'," has 20 reads and 0 followers (https://juejin.cn/user/2152214232570121/posts; account registered 10-20).

- X (x-radar local corpus, 60 days) search for sentetrader / 弈先: 0 posts. Jina search turned up no discussion beyond GitHub, the official site, Juejin, and one auto-generated page on skillsllm.com (no hits on V2EX, linux.do, or Zhihu).

### III. Competitors

| Name | Where | Launched | Scale | How it differs from Sente |
|---|---|---|---|---|
| TauricResearch/TradingAgents | Open source | About 2024-12 (repo age 690 days) | 123k stars, +1,091 over the last 7 days | Source of the bull-bear debate mechanism; a research framework, mainly US stocks; no order placement, no desktop product |
| hsliuping/TradingAgents-CN | Open source | About 2025-06 | 33k stars | Chinese version of TradingAgents, A-share/HK/US stock analysis reports; no futures, no live-market order placement |
| HKUDS/Vibe-Trading | Open source | About 2026-04 | 38k stars, +434 over 7 days | "Personal trading agent", backtesting + multi-agent, global markets; not targeted at China futures |
| HKUDS/AI-Trader | Open source | About 2025-11 | 23k stars | Fully automated agent trading competition; Sente's installer is named `AI-Trader_*.dmg`, suspected name piggybacking |
| TraderAlice/OpenAlice | Open source | About 2026-02 | 7.5k star | Slogan "Your one-person Wall Street", nearly the same sentence as Sente's "One person, one trading team"; covers stocks/crypto/commodities, English-language |
| OpenByteInc/QuantDinger, NoFxAiOS/nofx | Open source | 2025 | 14k / 13k stars | AI trading terminal, skewing toward crypto/forex/US stocks |
| ZhuLinsen/daily_stock_analysis | Open source | About 2026-01 | 73k stars, +3,522 over 7 days | Chinese LLM stock analysis + push notifications, mainly A-shares, no futures order placement |
| shinnytech/tqsdk-python (TianQin / KuaiQi) | Open source + commercial | 2018 | 5.1k star | Futures quant SDK, Sente's market-data base; no AI assistant layer |
| vn.py, QUANTAXIS, abu, etc. | Open source | 2015–2017 | 1–30k stars | Traditional futures/stock quant frameworks, require coding, no conversational AI |
| THS iFinD, East Money Miaoxiang, futures-broker in-app AI customer service (e.g. Capital Futures AI strategy generator in Taiwan) | Big-company features | From ¥2023 | Large user base (unverified) | Embedded in brokerages/info platforms, skewing toward Q&A and stock screening; doesn't do "bull-bear debate + expiry reconciliation + confirmation card"; Capital Futures AI generates MultiCharts strategies from natural language |
| PandaAI China commodity AI trader competition | Other | 2025 | Community event | A competition/community, not a personal assistant |

Bottom line: **"AI bull-bear debate" was already made a hot theme by the TradingAgents family (over a hundred thousand stars combined)** . Sente's real gap is "China futures + desktop finished product + confirmation card before order placement + program reconciliation of conclusions at expiry." That gap is genuinely unoccupied in open source, but it hasn't earned stars for that reason: the same-theme TradingAgents-CN gained several hundred in a single month, while Sente got 31 in 184 days.

### IV. Growth levers (where users come from)

1. **the spillover from the same author's viral repo on the homepage (main cause, strong inference)** . The author kingcharleslzy-ai created `dsh-purge` (a community-maintained version of the DeepSeek Harness "jailbreak" plugin) on 10-16. As of 11-18 it had 9,828 stars and 4,096 forks, and since 11-09 it has been gaining 400–600 a day. The backstory: the original author, 小杨/YuJunZhiXue, announced that dsh-purge 10-27 was going closed-source, and the original repo and npm package were deleted (X post by @NFTCPS [x.com/NFTCPS/status/2112463620462498004](https://x.com/NFTCPS/status/2112463620462498004); the trendshift note on the mmparko/dsh-purge mirror). This community version became the most active repo picking up the orphaned audience, so **this is traffic from a platform event, not from Sente** . The author's homepage pins 4 repos: bizhub-installer, DSH-Spicy, dsh-purge, and **sentetrader** (GitHub GraphQL pinnedItems). The profile has 100 followers. Among the hundreds of people starring dsh-purge each day, a few clicking through to the profile and seeing the fourth pinned repo is enough to explain Sente's daily gain of 10–19 (roughly 2–4% of dsh-purge's daily gain). The author's other pinned repo, DSH-Spicy, also rose over the same period from single-digit daily gains to 40–67, in the same direction. The dsh-purge README itself doesn't link to Sente (only to DSH-Spicy), so the referral can only come through the homepage.

2. **Juejin dev log** : 1 posts, 20 reads, negligible.

3. **Official site / own domain** : the qilinshuzhi.com homepage puts Sente first, and GitHub mirrors the site's downloads; traffic scale unknown.

4. **Bought stars** : a guess made during recall. The only support is "0 forks + unusually smooth growth"; but the GitHub stargazers endpoint returns 404 for every repo this time (GraphQL also returns empty), so the accounts giving stars can't be seen, and **this can be neither confirmed nor ruled out** . Given the author's other source of several hundred stars a day, the curve can be explained without buying stars.

5. **Name piggybacking** : the installer is named `AI-Trader_<version>`, the same as HKUDS/AI-Trader (23k stars); but it does nothing in GitHub search, so it doesn't count as a channel.

6. Not found: ads, X/Xiaohongshu/Douyin/Bilibili content, V2EX/linux.do posts, press coverage, influencers.

### 5. Why it broke out (it was really just "seen a little")

- **Riding along** : the author had a repo that blew up after the original author went closed-source and it was taken over (dsh-purge, growing 400–600 a day). Sente was pinned on the same homepage and picked up the spillover. Timing-wise, Sente gained only 28 stars in the 20 days before 11-07, then settled at 10–19 a day after 11-07, roughly in step with dsh-purge doubling from 11-09 (Sente was also shipping 9 versions in quick succession around 11-06–11-09; this may be when it got pinned, not verified).

- **Good README presentation** : "One person, one trading team", "one AI bullish, one AI bearish, they argue and then give you a conclusion", plus 6 phone/desktop screenshots, so passing developers are willing to star it on the way by. But 0 issues and 47 download counts suggest the people who starred it mostly didn't install it.

- **Not because the product was discovered by the futures crowd** : there were no futures communities, financial media, or X discussion; retail futures traders don't browse GitHub to begin with.

### VI. Differentiation

- Compared with the TradingAgents family: it isn't a research framework but a finished product. **It shows a confirmation card before placing an order, invalidates debate numbers with no source, attaches invalidation levels and revision conditions to conclusions, has a program reconcile against closing prices 5 trading days later instead of the model grading itself, and private memory comes only from the user's own words and real fills and can be revoked** . It is vertical to China's futures market (Kuaiqi quotes, rebar/methanol-type contracts, read-only MT5 for overseas markets).

- These design choices are genuinely thoughtful (especially "reconcile at expiry" and "confirmation card before orders" as a positive answer to the credibility of LLM trading advice), **but they are not why it got its stars** : the stars came from passing traffic from the author's other repo, and the futures users these differentiators are aimed at haven't been reached yet.

### VII. What others can borrow

- **A pinned spot on the homepage is a free distribution slot** : if you have a hit repo, pinning a new project next to it steadily captures 2–4% of the hit's daily star growth. The prerequisite for this play is having a hit first (here, catching the original author going closed-source and taking over the community version).

- **"GitHub as download site + official site mirror + invite codes"** : a closed or semi-closed vertical product ships installers via GitHub Releases and uses stars as endorsement on the side; later everything was open-sourced (AGPL) to build trust, while invite codes for the cloud version were kept.

- Worth copying at the product level: the debate conclusion "reconciled by program at expiry", private memory split into three types (profile / agreements / lessons) with sources marked, and merging everything into one confirmation card before an order.

### VIII. Unknowns

- Whether stars were bought: the stargazers endpoint is unavailable, so account quality can't be seen; 0 forks and the smooth curve are weak signals.

- When Sente got pinned to the homepage, the official site's real download/signup numbers, and how many people registered with invite codes are all unknown.

- Whether it was promoted in WeChat groups / futures circles privately: public web search finds nothing, which doesn't mean it didn't happen.

- Whether it will charge (invite codes plus the official cloud suggest possible future pricing); no public pricing yet.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.
