# A backtesting tool teaches you not to trust backtests, but its own stars don't hold up: SimpleQuant went from 26 to 128 in five days while downloads fell · GitHub open source project

> Source: https://stringers.io/en/github/gh_AllenPan-Git__SimpleQuant.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> It never really took off: in early October it organically climbed to 26 stars off the author's daily Bilibili videos on counterintuitive backtests, then stalled. From 11-02, five days took it to 128, in sync with a batch of freshly registered fake accounts forking in pairs. Over the same period, downloads per release dropped from 100 to 45–85, which is basically manufactured.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

AllenPan-Git/SimpleQuant · AllenPan-Git

# A backtesting tool teaches you not to trust backtests, but its own stars don't hold up: SimpleQuant went from 26 to 128 in five days while downloads fell

**What it is** A free, open-source, low-code desktop backtesting app for China A-shares. Pick options or describe a strategy in one sentence, and the backtest report proactively warns you when the results may be fake.

**Why it blew up** It didn't really: in early October the author's daily reverse-intuition backtest videos on Bilibili organically pushed it to 26 stars, then it stalled. Starting 11-02, it jumped to 128 over five days in sync with a batch of freshly registered fake accounts forking it in pairs. Over the same period, downloads per version fell from 100 to 45–85 times, multiple times. It's largely manufactured.

**C** **How you could approach it** Don't build another A-share backtesting app (there are several similar ones with thousands of stars). If you have a vertical tool aimed at non-programmers, copy its playbook: post a daily video about a "counterintuitive conclusion tested with the tool," demo one feature in each, and then do one dedicated push in developer communities.

Grade C: this surge is most likely manufactured, and real interest is only about 20–30 stars. The category is already held by projects with thousands of stars like InStock, khQuant, and easy-stock. What's worth learning is the content-plus-tool acquisition approach. Few people know about it yet

Data (updated daily) · Repo created 2026-10-01 · 128 stars · 7 days +102

GitHub · stars

Data 2026-10-09 → 2026-11-08 (31 days, 30 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

SimpleQuant is a free A-share quant backtesting desktop app built by one person. The product is solid, and the author posts a "debunking backtest myths" video on Bilibili almost daily. But the star jump in early November lines up to the day with a batch of newly registered accounts forking it in pairs, and downloads per version also fell over the same period. The real traction is the 26 stars the videos organically earned in early October. Building a copy makes little sense, but its "tool plus daily content" playbook is worth learning.

### One video, opening with a +37.58% headline

On Oct 5, 2026, the Bilibili account "SimpleQuant简单量化" posted its first video: "You can do quant trading without coding? Backtest returns this high, so why the warning?" In 14 minutes, using an RSI strategy on the CSI 300 ETF, it walks through downloading data, building the strategy, backtesting, tuning parameters, and running paper trading, all inside the app. The +37.58% backtest return in the opening got a warning from the software itself, and the author wrote "6:12 reveal" in the pinned comment.

By Nov 6 the video had 3,432 views and 145 favorites, over three times its likes (41) (Bilibili, 11-06). The comments section has only a few entries, like "big shot" and "bookmarking." Favorites far outnumber likes, which suggests viewers are saving it as a tutorial to install the app and follow along later.

The account posts almost daily. As of Nov 6 it had posted 32 videos, with about 18.5k total views across the collection and 165 followers.

### What it is

A low-code backtesting app for A-shares that covers stocks, ETFs, and convertible bonds. It ships installers for Windows, macOS, and Linux, uses free data from AKShare, BaoStock, and local TDX data, runs on the Backtrader engine, and has a NiceGUI interface. There are four ways to write a strategy: pick a template, assemble condition blocks ("buy when MACD crosses up and RSI is below 70"), describe it in plain language for AI to convert into rules, or write Python directly. It also offers multi-factor stock selection, AI factor mining, and risk-rated asset allocation, and you can export to JoinQuant, Myquant, or QMT to double-check.

What sets it apart from similar tools is that every report keeps reminding you the result might be fake: is the trade count big enough, is the profit concentrated in a few trades, how much better is it than random trades of the same number, is it overfitted, and how does the best parameter set rank in each individual year. The A-share rules are also detailed, with trading in lots of 100 shares, T+1 and T+0 depending on the product, stamp duty, and minimum commission all built in.

Licensed under GPL-3.0 and completely free, with only an Afdian donation link from the author. The repo was created in Oct 1, and in 36 days it shipped 37 versions (v0.1.0 through v3.1.0), with incremental auto-updates and signing. Chinese users can download via 123 Cloud Drive. The author's account was only registered in February, and their other repos are course assignments with 0 stars, so the author looks like a student or recent hire in finance or data (inference).

### The video topics are themselves product demos

The titles of the 32 videos almost all follow one format: a counterintuitive backtest conclusion:

- "Can MACD actually make money? Backtesting different underlyings with the free app, the results differ by over 40x", 1,615 views

- "Best parameters for a dual moving average? Ranked 1 last year, 143 the next year", 489 views

- "Stop-loss set at 8%, why did you still lose 84.7%? 150 stocks tested with stop-loss", 429 views

- "A stock-picking rule with a 100% win rate, yet 98.6% of stocks are trapped"

- Other titles: "Buy the smallest 10 A-share stocks by market cap every month" about 1,190, "Dividend low-volatility backtest" about 1,130, and most others in the range of two to seven hundred

Each conclusion comes from a built-in check in the app, so watching the video is effectively watching a feature demo. The pinned comment links to the download in each case and tells you which page and button to click to reproduce it ("Condition blocks → Sell conditions → holding return less than −8"), then teases the next episode. Knowledgeable commenters also chime in: under the MACD video, someone suggested "try 30,60," and the author replied they had tried it and "it all improved somewhat."

This is how the early-October organic growth happened: stars reached 26 the week before, with each version getting 150 to 290 downloads (GitHub Releases download counts). An automated Chinese AI-generated GitHub daily digest mentioned it on Oct 6 and 7, calling it "a new A-share quant star." X returns nothing for it, and the author's posts on V2EX and Zhihu weren't found.

### Where the other 100 stars came from

From Oct 7 to Nov 1, stars stayed between 25 and 28, and then in Oct 24 it lost 3 stars. Then this happened (GitHub daily star records):

| Date | star | fork |
|---|---|---|
| 11-01 | 26 | 2 |
| 11-02 | 47 | 4 |
| 11-03 | 69 | 6 |
| 11-04 | 88 | 6 |
| 11-05 | 88 | 8 |
| 11-06 | 128 | 10 |

There were no Bilibili hits during those days, and new videos got three or four hundred to seven hundred views. Looking at the forks: by Nov 6 there were 14 forks in total, 12 of them from new accounts registered in Oct 22 and Oct 30. They showed up daily in pairs, less than a minute apart. Usernames follow the "English name plus random letters" pattern (stackvoncilemaxk, opslolasturgess), their own repos are random word combinations like misty-bridge-lab and humble-otter-mini, and they forked big projects like nginx and Ghost to "age" the accounts. Their bios are made up, like "Graphics engineer @digitalocean." 4 of them have already been banned by GitHub (GitHub, 11-06). Only 2 forks came from genuine older accounts.

The list of stargazers isn't available, so the accounts can't be verified one by one. But the days the fake forks appeared line up exactly with the days stars jumped, and after Nov 2 each version got only 45 to 85 downloads, fewer than the 100 to 165 downloads in mid-late October. Stars went up five-fold, but few people actually used it. The star surge is largely manufactured (inference). It's unclear whether the author bought it, though boosting services that "age" fake accounts often star small projects along the way.

### Could you build one yourself?

**Building a copy of this A-share backtesting app and getting stars would be hard.** The niche is already crowded:

| Project | Form | Launched | star |
|---|---|---|---|
| InStock(myhhub/stock) | Self-hosted web, focused on stock screening and patterns | 2023-03 | 15k |
| easy-stock | Desktop, AI research agent | 2026-08 | 1,760 |
| Kanhai khQuant open-source edition | Free A-share backtesting desktop app | 2025-01 | 1,630 |
| ZiyuQuant | Personal local A-share quant workstation | 2026-08 | 193 |
| SimpleQuant | Low-code backtesting desktop app | 2026-10 | 128 (real figure around 26) |

Above that are frameworks like vn.py, qlib, and Backtrader with thousands to tens of thousands of stars, online platforms like JoinQuant, Ricequant, and QMT, and Tonghuashun already offers "one-sentence backtesting." SimpleQuant's product is more detailed than most competitors, yet its organic traffic only reached about 20–30 stars. This shows that in the Chinese quant space, making a good app isn't enough.

**Other markets are taken too.** There are more backtesting tools for US stocks and crypto, and the English open-source scene has mature projects like backtesting.py, vectorbt, and Lean.

**What's actually valuable is the playbook.** One person builds a tool and posts daily "counterintuitive conclusions tested with this tool," demoing one feature in each video. Making this kind of content in Bilibili's finance section is cheap, search traffic lasts, and videos and the app pull in users for each other. The gap is that no video has yet broken 10,000 views, so it never took off.

### If it were you

**Verdict** : Don't follow SimpleQuant itself. Its five-day star jump is most likely manufactured, and real organic interest is only 20–30 stars. There are already several A-share backtesting apps with thousands of stars. What's worth learning is the combination of "tool plus daily hands-on content," along with the positioning of arguing against the reader's own backtests: "your backtest may be fake."

**Not for you if** : you want to build yet another A-share or US-stock low-code backtesting app and expect the product alone to earn stars.

**Good fit for** : you have a vertical tool aimed at non-programmers and you can keep producing verifiable, counterintuitive findings from it. Examples: backtests, checkup-report readers, renovation budgets, AI-generated-content detection. How to run it:

1. **Each post is one concrete finding, and the finding is the feature demo.** Title leads with a number ("Set a 8% stop-loss and lost 84.7%"). The body runs it through the software, and the pinned comment has the download link and repro steps.

2. **Argue the opposite.** In a category where everyone brags about returns and results, the first tool that warns "this might be fake" is easy to trust.

3. **Even solo, ship like a real release.** Installers for three platforms, auto-update, and a China-accessible cloud-drive mirror, so non-programmer users can actually install it.

**How to get users:** on video platforms with search traffic, like Bilibili and Xiaohongshu, post daily for at least a month. To get GitHub stars, do a dedicated push on the places developers gather, like V2EX, Zhihu, and HN. SimpleQuant skipped this step. Don't buy stars: new accounts forking in pairs is obvious, GitHub bans them over time, and downloads won't rise with them.

**When to move:** there's no rush. This playbook doesn't depend on a time window. It works as long as nobody is already posting daily content in the same niche.

**What would prove you wrong:** if SimpleQuant's next video passes 10,000 views, downloads climb back above 200 per release, and stars keep growing on their own past 500, then there's room in Chinese non-programmer backtesting, and "build a better one" is worth reconsidering.

[GitHub repo ↗](https://github.com/AllenPan-Git/SimpleQuant)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| Kanhai khQuant open-source edition | Open source | 2025-01 | 1,630 star | Also a free desktop backtester for China A-shares. SimpleQuant adds AI natural-language rule writing, credibility and overfitting checks, and multi-factor and asset allocation. |
| InStock(myhhub/stock) | Open source | 2023-03 | 15.0k star | A web system focused on stock picking, indicators, and chart patterns. You self-host it; it isn't a low-code backtester for beginners. |
| easy-stock | Open source | 2026-08 | 1.76k star | Builds an AI research agent. SimpleQuant focuses on rigorous backtesting. |
| ZiyuQuant | Open source | 2026-08 | 193 star | A personal, local A-share quant workbench with almost the same positioning. SimpleQuant leans harder toward non-coders. |
| vn.py / QUANTAXIS / qlib / Backtrader | Open source | 2015-2020 | 6k–50k star | A framework for people who write code. SimpleQuant is a finished product with a graphical interface on top of Backtrader. |
| JoinQuant / RiceQuant / BigQuant / Juejin / QMT | Web | Years | Large commercial platforms | Online platforms that require an account and partly charge. SimpleQuant runs offline and locally, is free, and can export to these platforms for cross-checking. |
| THS BackTest / Guoren | Big-company features | Years | Large | The big players already offer "one-sentence backtests" and point-and-click stock picking. SimpleQuant's edge is transparent reports and open source. |

Where users come from

- **Verified** **Bilibili backtest-test compilation from the account itself** : "SimpleQuant简单量化" posted 32 videos starting 10-05, the collection has 18,527 views and 538 favorites, the intro tutorial got 3,432 views, and the README pins this video.

- **Verified** **Picked up by an AI-generated GitHub daily digest** : agents-quant.com 10-06 and 10-07 daily digests called it a "new A-share quant project" (at the time, 26 stars).

- Inferred **Star and fork farming** : 11-01–11-05, two new accounts registered 10-22/10-30 per day forked in succession within 1 minutes, forking random-word repos. 4 of them have been banned. The dates and the star jumps line up exactly, while downloads at the same time went down. The star list itself can't be verified.

- **Verified** **Domestic cloud-drive mirror distribution** : the README offers downloads via 123 cloud drive (with extraction code). The download count isn't visible.

## Why SimpleQuant (AllenPan-Git/SimpleQuant) seemed to take off

> Data as of 2026-11-06. Sources: GitHub API (repos, releases, forks, fork-account profiles, fetched 2026-11-06), our github database data cards, Bilibili public endpoints (read via r.jina.ai), Jina/Exa search, x-radar.

### Bottom line

It didn't really take off. SimpleQuant is a personal low-code A-share backtesting **desktop app** (Backtrader + NiceGUI, packaged as Win/Mac/Linux installers). The product itself is solid, and its organic traffic came from the author's daily Bilibili "backtest test" collection. But the surge that brought it stars (11-02 to 11-06, stars going from 26 to 128) can largely be attributed to **star and fork farming** : of 14 forks in that period, 12 came from accounts registered 10-22 to 10-30 days earlier, appearing in pairs less than a minute apart each day, and 4 of them have since been banned by GitHub. Meanwhile, real downloads per release didn't rise during that time. They actually fell below late October.

### 1. What the product is

- A low-code backtesting tool for A-shares (stocks, ETFs, convertible bonds): point-and-click through data → strategy → backtest → parameter optimization → paper trading. Four ways to write strategies: templates, condition components, natural language that AI turns into rules, and Python code. Multi-factor stock selection, AI factor mining, and risk-assessed asset allocation. Results can be exported to JoinQuant / Juejin / QMT for a second check. Data comes from the free AKShare / BaoStock / TDX local data. (README)

- Its standout feature is guarding against self-deception: every backtest report comes with built-in credibility checks (is the trade count high enough, are returns concentrated in a few trades, a comparison against random trading, overfitting warnings, and yearly rankings of the best parameters). The A-share rules are also handled in detail: whole-lot trading, T+1 / T+0 settlement by product type, stamp duty, and minimum commission. (README)

- Completely free, GPL-3.0. The author's GitHub bio only links a donation page on Afdian (afdian.com/a/allengit). No paid tier is visible.

- Very fast iteration: 36 days after the repo was created, 37 releases shipped (v0.1.0 is 10-01, v3.1.0 is 11-06), with 124 commits in total, automatic incremental updates, and Ed25519 signatures. (GitHub releases API)

- The author account AllenPan-Git (display name Lamia) was registered only 2026-02 ago. Its other 8 repos are course assignments and small projects with 0 stars each (College-Scorecard, Data-Bootcamp, etc.). It looks like a student or recent grad in finance or data.

### 2. Growth data and authenticity check

#### Star timeline (our github database repo_day + data cards)

- Repo created 10-01; around 10-07 it reached 26 stars (agents-quant.com's 10-06 and 10-07 digests both say "26 stars").

- From 10-07 to 11-01 it sat flat at 25–28,10-24, then lost 3.

- 11-02 47 → 11-03 69 → 11-04 88 → 11-05 88 → 11-06 128; forks rose from 2 to 10 (GitHub now shows 14).

#### Fork accounts: clear signs of batch fake accounts (GitHub API)

| Fork time (UTC) | Account | Registration date | Profile |
|---|---|---|---|
| 11-05 17:33 / 17:36 | tmmarkwong / hacksoliveaxc | 10-22 / 10-22 | 0 followers; the bio is made up: "Graphics engineer @digitalocean", "Security engineer, Seoul" |
| 11-04 17:54 / 17:55 | stackvoncilemaxk / opslolasturgess | 10-30 / 10-30 | 0 followers |
| 11-03 16:48 / 16:49 | uxadriadouglas / kenyattautleyru | — | **Banned (API 404)** |
| 11-03 04:32 / 04:32 | minxiaotc / devfaebsk | 10-30 / 10-30 | Bio: "Systems programmer @carta" |
| 11-02 00:58 / 00:58 | leora-islamscp / luannoldfieldgfs | — / 10-30 | One already banned |
| 11-01 04:36 / 04:36 | lumennellyxmh / aleishapatterson26 | — / 10-30 | One already banned; bio: "Senior Staff Solutions Architect @npmx-dev" |
| 10-28, 10-19 | foxxnuaa / EliomEssaim | 2014 / 2017 | Real, older accounts (e.g. a USTC student) |

- What these accounts have in common: registered only 1–2 weeks ago; usernames are "English first name + random letters"; repo names are random word combos like misty-bridge-lab, spare-isle-kit, humble-otter-mini; they also forked big projects like nginx, neo4j, Ghost, and caveman to "age" the accounts; and they forked in pairs, less than 1 minutes apart, every day. This is the typical playbook of a view-and-star farming service.

- The star-list API returns 404 for every repo, so individual starring accounts can't be checked. However, the fork times and the star-jump dates line up exactly (each of 11-01 / 02 / 03 / 04 / 05 has a matching pair), so it's reasonable to infer the stars came from the same batch of accounts. **The starring accounts themselves are unverified.**

#### Real usage didn't rise with it (GitHub release download counts)

- 37 releases total 4,590 downloads (including auto-update traffic). In early October each release got 150–290 downloads; in mid-to-late October, 100–165. **After 11-02** (v2.3.0 through v3.1.0), each release only got 45–85. Stars grew 5 times over, yet downloads went down, which means the star surge brought in no new users. There is also a domestic download mirror on 123 cloud drive, but its numbers aren't visible.

#### The real organic traffic: the daily Bilibili collection

- The Bilibili account "SimpleQuant简单量化" (mid 3747568221686847) started posting almost daily around 10-05, each a 5–10 minute backtest test, reaching 11-06 videos by 32. The collection totals 18,527 views, 254 likes, 538 favorites, and 82 comments; the account has 165 followers. (Bilibili endpoints, 11-06)

- The most-viewed was the intro tutorial "Can you do quant trading without knowing how to code?" with 3,432 views (145 favorites), followed by "Does MACD actually make money?" with 1,614, "Buying the 10 smallest-cap A-shares every month" with 1,186, and "Backtesting dividend low-volatility" with 1,128. Most others were in the 200–700 range.

- Every topic is a "counterintuitive backtest finding": "The best dual moving-average parameters ranked 1 last year and 143 the year after", "Set a 8% stop-loss and still lost 84.7%", "Can a monkey beat random trading?" Each one promotes the software's credibility-check feature. This was the main source of the 26 stars and the one or two hundred downloads per release in early October.

- Around 11-02, Bilibili had no viral hit (those days' videos drew 316 / 742 / 492 views), so it can't explain the star jump.

#### Other channels

- The Chinese AI-generated GitHub daily digest agents-quant.com mentioned it twice, on 10-06 and 10-07, calling it a "new A-share quant project".

- x-radar searches for SimpleQuant / AllenPan-Git each returned 0 results. Searches on V2EX, linux.do, and Zhihu turned up no posts by the author (Exa / Jina search).

- No media coverage, no retweets from big accounts, no revenue.

### 3. Competitors

| Name | Where | Launched | Scale | How it differs from SimpleQuant |
|---|---|---|---|---|
| Kanmi quant khQuant open-source edition (khscience/OSkhQuant) | Open-source desktop | 2025-01 | 1,630 star | Closest positioning: a free A-share backtesting desktop app, no brokerage account needed. SimpleQuant adds AI that writes rules from plain language, credibility and overfitting checks, multi-factor, and asset allocation |
| InStock(myhhub/stock) | Open-source Web | 2023-03 | 15.0k star | Focused on stock picking, indicators and chart patterns; you have to self-host it, and it's not a low-code backtester for beginners |
| easy-stock(jundizhou/easy-stock) | Open-source desktop | 2026-08 | 1.76k star | Takes the AI research agent route; SimpleQuant's core is rigorous backtesting |
| ZiyuQuant | Open source | 2026-08 | 193 star | Also a personal, local A-share quant workbench with factor research, nearly the same positioning |
| vn.py / QUANTAXIS / wondertrader / qlib | Open-source framework | 2015–2020 | 6k–50k star | A framework for people who write code; SimpleQuant is a finished app for people who don't program |
| Backtrader | Open-source library | 2015 | 23.7k star | SimpleQuant's backtest engine, which the interface sits on top of |
| JoinQuant / RiceQuant / BigQuant / Juejin / QMT | Web and brokerage platforms | Years | Large | Online platforms that require an account and partly charge. SimpleQuant runs offline and locally, is free, and can export to these platforms for cross-checking. |
| THS BackTest (Tonghuashun), Guorn (果仁网) | Big-company features / web | Years | Large | "One-sentence backtests" and click-to-select stock picking are already offered by big players; SimpleQuant's edge is transparent reports and local execution |

Verdict: the A-share low-code / free backtesting niche is already crowded, with both open-source and commercial platforms. SimpleQuant has no unique feature.

### 4. Why it did (not) break out

1. The organic stretch (10-01~10-07,0 → 26 stars) came from the author's own daily Bilibili backtesting series, plus inclusion in an AI daily digest. Evidence: the launch video got 3.4k views; each early-October release was downloaded 150–290 times.

2. Organic growth then stalled: stars stayed at 25–28 from 10-07 to 11-01, and each Bilibili series video had only a few hundred views and 165 followers.

3. The jump starting 11-02 was manufactured: 12/14 forks came from fake accounts registered in batches in late October, less than a minute apart, some since banned. Over the same period, downloads per release actually fell to 45–85.

4. The author likely hoped stars would push the repo onto GitHub Trending or lend credibility to the Bilibili content. But the repo is not yet 60 days old and hasn't reached Trending (data card trending_days_any=0).

### 5. Differentiation

- The real differentiator is product philosophy: it tells you "your backtest may be fake." Credibility checks, random buy/sell baselines, out-of-sample testing, and year-by-year rankings of optimal parameters are built into the default flow, paired with Bilibili content that "debunks backtest myths." Most similar tools advertise returns; this one talks about risk.

- Also, the bundle of "installer + free data + local offline + export to JoinQuant / QMT" lowers the barrier for retail investors who don't code.

- But this differentiation is not what drove the data-card spike. It only brought niche organic attention (26 stars, a few thousand Bilibili views); the spike on the data card was faked.

### 6. What to borrow

- The "tool + daily real-test content" approach: each video presents a counterintuitive backtest result, and the result itself demonstrates the software's features, so content and product feed each other. Making this content in Bilibili's finance section is cheap and stays searchable long-term.

- Making "overfitting prevention / credibility" the product's selling point, in a category where everyone touts returns, builds trust by doing the opposite.

- Releases are engineered well (incremental auto-updates, signing, domestic cloud-drive mirrors, three-platform CI), so one person can sustain a release a day.

- Cautionary lesson: pairs of freshly created accounts forking together are easy to spot, GitHub keeps banning them, and manufactured stars don't convert into downloads.

### 7. Still unclear

- The list of starring accounts can't be retrieved (the stargazers API returns 404), so "the stars were fake" is inferred from the fake forks and download data, not verified one by one. It's also possible a third party padding its own fake accounts added this repo to their count, not necessarily the author buying them.

- Still missing: domestic download numbers for 123 cloud drives, and real user feedback from Bilibili comments.

- Monetization plans: currently only Afdian tips exist, and revenue is not visible.

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