At 3 p.m., someone in your group chat forwards a screenshot: "Company X just landed a huge order. Revenue will double next month." No source, no date, yet a row of "all in!" replies underneath. That evening you search for the company, and the top result is an upbeat analysis. You don't notice it's three years old. Before bed you ask some AI tool, and it tells you with a straight face that the earnings call is next Thursday and foreign investors have been buying for five days straight.
Quite possibly none of these three pieces of information is true. But they have one thing in common: you will trade on them. Bad information doesn't lose you money by itself; the decisions you make based on it do. Stock Helper treats "where the data came from" as part of the analysis: for every sentence, you know whether it was looked up, inferred, or not found.
30-second overview
| Item | How it's handled |
|---|---|
| Every fact | Sourced on the spot, for example "(Source: Anue, 7/12)", so you can check it yourself |
| Data it can't find | Stated plainly as "Not found," with no gaps filled by "expected," "likely," or "usually" |
| Inferences and judgments | Start with "In my view," "My assessment," or "My guess," never dressed up as fact |
| News and discussions | Dates and years are checked; posts older than a month aren't treated as current market sentiment |
| P&L and returns | Always calculated by a program, using the original cost you registered |
| Big moves in a single stock | For single-day moves beyond ±5%, it checks the broader market and the previous session before naming a cause |
How does bad information turn into losses?
Bad information usually turns into losses in one of three ways:
- Buying on fake good news: An unsourced tip gets you chasing the price. By the time you learn the news doesn't exist, or was priced in long ago, you've already bought near the top.
- Selling on stale bad news: You treat old news as today's and take a loss when there was no need to panic.
- Miscalculating your stop-loss: A wrong P&L number makes you sell before your stop is actually reached, or think you're fine after it has already been breached.
Stock Helper's data discipline blocks these paths one by one.
Eight data rules, each blocking one way to lose money
1. Every fact is sourced, so you can check it yourself
Every fact it states comes with a source right there, for example "TSMC closed at XXX, down 2.03% (Source: XXX)" or "News (Source: Anue, 7/12)." When a paragraph contains several facts, each one gets its own source.
This rule blocks one way of losing money: believing unsourced tips. The biggest problem with a group-chat screenshot isn't what it says; it's that you have no way to check it. With a source, you can verify the claim and judge how much weight the source deserves: an official filing and a forum post are simply not in the same league.
2. If it can't find something, it says "Not found": better to say less than to say something wrong
This is the most counterintuitive rule, and the one that protects you most. When data can't be found, it's tempting to write "the earnings call is expected soon." It looks complete, but it's made up. Stock Helper's rule: if it can't find something, it writes "Not found." No blanks, no inventions, and no filling gaps with words like "expected," "likely," or "usually." What you will see is: "Earnings call: no recent information found."
Why say less? Picture two scenarios (illustrative):
- If it invented "the earnings call is next Thursday": you would buy early expecting good news, only to find there's no such call, and you've bought at a level with no catalyst.
- If it invented "foreign investors net-bought 3,000 lots five days in a row": you would think institutions were propping the stock up and loosen your stop-loss, when foreign investors were actually selling.
A "Not found" costs you at most two minutes of checking yourself. A made-up date or lot count gets you into the wrong trade. The first is a cost in time; the second is a loss in real money.
3. Facts and judgments are kept apart: "In my view" means inference
What it looks up is fact and gets a source. Inferences start with "In my view," "My assessment," or "My guess," and are never dressed up as fact. It won't write "TSMC is going to rise next." It will write "In my view, there's a chance of a rebound, but watch XX."
This rule blocks going all-in on an inference you mistook for fact. A confident "it's going up next" easily tempts people to add to a position. But if you can see at a glance that it's an inference, you'll go back to its reasoning and to the part that says "under what conditions this judgment no longer holds," and then decide how much to bet.
4. Dates, years, and timing are checked: old data can't pose as current
Before citing news or discussions, it checks the date and the year. Forum search results often mix in posts from years ago, and discussions older than a month aren't treated as "current market sentiment." The same time discipline applies to prices:
- The morning briefing goes out at 08:30 by default, and the Taiwan market doesn't open until 09:00, so every Taiwan stock figure in it is the "previous session close," never described as "this morning's trading."
- Pre-open auction indicative prices between 08:30 and 09:00 are never used. They are simulated values produced by orders that can be placed and cancelled at will, not prices anyone actually traded at. A real case (2026-08-05): the indicative price showed an electronics stock at -9.84%, close to limit-down; it actually opened at +3.28%. If that number had scared you into placing a sell order, you would have been driven out by a price that never traded.
- Technical indicators such as RSI are calculated from daily data, and it notes the reference date when quoting them. Intraday prices are labeled "intraday price, subject to change before the close; the closing price is final."
This rule blocks making today's decisions with outdated information: treating a heated three-year-old discussion as a reason to chase, or treating an indicative price as today's market.
5. P&L is always calculated by a program: you use this number to decide your stop-loss
There's a real case behind this rule. On 2026-08-13, a holding with a cost of NT$207 closed at NT$205. The correct P&L was -0.97%, but it was once calculated as -2.4%, as if NT$3 of cost had appeared out of nowhere.
A gap of about 1.4 percentage points doesn't sound like much, but put it into a stop-loss rule (illustrative): if your rule is "trim once I'm down 2%," -0.97% means you're still within plan and don't need to act; -2.4% would make you think your stop had triggered and sell a position you shouldn't have touched. You make decisions with this number. So since that case, P&L and returns are always calculated by a program, never with mental math.
Two more details: cost is always the original cost you registered, and if it also shows a return "including dividends," both numbers are listed separately and clearly. You won't get one vague percentage.
6. Market context first, then the cause
When a stock moves more than ±5% in a day, before offering any explanation it checks how the broader market did that day, and how both the market and the stock did in the previous session. If a stock fell sharply or hit limit-down the day before, another big drop today is treated first as a continuation or a catch-up decline, unless it finds specific bad news about the company. It will not conclude that something is wrong with the company just because "the market fell a little and the stock fell a lot today." Stocks locked at limit-down often catch up on the downside the next day, even as the broader market has already stabilized.
Scenario (illustrative): your stock falls 7% today while the market falls just 1%. Your gut says something must have happened at the company. But if it was locked limit-down yesterday and sellers simply couldn't get out, today's drop may just be the catch-up. Get the cause wrong, and whatever you do next, whether panic-selling at a stop or buying more because you think it was sold off unfairly, will be wrong too.
7. Official and reliable sources come first
It checks these sources first:
- Official: Taiwan Stock Exchange (TWSE), Taipei Exchange (TPEx), Market Observation Post System (MOPS)
- Financial information sites: Yahoo Finance Taiwan, Goodinfo, Anue, MoneyDJ
- Financial media: Economic Daily News, Commercial Times
For figures such as institutional net buying and selling, monthly revenue, ex-dividend dates, and material company announcements, official sources are the final word. When the source is an official site, you can be more confident; when it's a media report, you know it's secondhand, and important numbers are worth confirming once more against an official source.
8. Taiwan's own terminology, so you don't have to translate in your head
It uses the terms Taiwan's financial community uses, such as the Taiwanese names for the Fed and for chips, rather than mainland Chinese terms. It sounds minor, but when an analysis mixes vocabularies, you have to translate as you read and are more likely to misread. Written in the terms you know, your attention stays on the numbers and the judgments.
How to read an analysis: fact, judgment, "Not found"
Below is an illustrative excerpt from a stock analysis (the company, prices, and figures are all fictional):
Stock A closed at 120.5 in the previous session, down 1.6% (Source: TWSE). August revenue rose 12% year over year (Source: MOPS). Foreign investors were net sellers on 4 of the last 5 days (Source: TWSE). RSI(14) is 68 (reference: 9/24 daily), approaching the hot zone. Earnings call: no recent information found. In my view, there's limited room to chase in the short term; it's better to wait for a pullback toward support near 115 and reassess. This judgment no longer holds if the stock breaks above resistance at 125 on rising volume, which would signal stronger buying and call for a fresh assessment.
Here it is broken down:
| Sentence | Type | How you can use it |
|---|---|---|
| "Closed at 120.5, down 1.6% (Source: TWSE)" | Fact | Check the closing price on TWSE |
| "August revenue up 12% year over year (Source: MOPS)" | Fact | A fundamental data point; see the original filing on MOPS |
| "Foreign investors net sellers on 4 of the last 5 days (Source: TWSE)" | Fact | A clue about who is buying and selling; read it together with where the price is |
| "RSI(14) is 68 (reference: 9/24 daily), approaching the hot zone" | Fact + interpretation | The value is fact; "approaching hot" is an interpretation based on the RSI zones. Note that the reference date is the previous session |
| "Earnings call: no recent information found" | Not found | This is a blank, not "nothing is happening." If it matters, check it yourself or treat it as an uncertainty |
| "In my view, there's limited room to chase…" | Judgment | See whether the reasoning convinces you; it's an analytical reference, not an instruction |
| "No longer holds if the stock breaks above 125 on rising volume…" | Judgment (invalidation condition) | Knowing in advance what would overturn the judgment keeps you from clinging to it after conditions change |
Each kind of sentence has its own use: facts are for checking, judgments are for thinking, and "Not found" is a reminder of what you still don't know. Read them all the same way and you'll mistake inference for fact and blanks for "nothing to see."
Three checking habits you can build yourself
- Treat unsourced news as if it doesn't exist. Forwarded messages, screenshots, "I heard": until you find a source, none of it is a reason to buy or sell.
- Check the date before the number. For news, discussions, and quotes, confirm which day they're from first. The year is especially easy to get wrong.
- Recalculate any number that could make you sell. P&L is (current price − cost) ÷ cost. A stop-loss number that could make you hit "sell" is worth ten seconds on a calculator.
If it has a source, can it still be wrong?
Sourcing isn't a 100% guarantee. It exists so you can check, not to promise it's always right: the source itself may be wrong, and the retelling may slip. That's why every reply ends with a reminder that "AI analysis may contain errors; please verify it yourself." The difference is that when a mistake happens, you have a way to catch it, instead of finding out after losing money that the number was made up. The -0.97% case above was caught and turned straight into a rule.
FAQ
Q: Why do some answers contain several "Not found" entries?
It means those items had no reliable data this time, for example no recent earnings call news, or no specific lot counts for institutional buying and selling. It would rather say so than use "expected" or "likely" to patch together an answer that looks complete. For items that matter to you, ask a follow-up or check MOPS or TWSE yourself.
Q: Can I just act on the parts marked "In my view"?
Those are analytical references, not investment instructions. Every full analysis includes the biggest risk, the conditions under which the judgment no longer holds, and the uncertainties, so you know what the judgment rests on. Whether to act on it is your call, based on your own capital and risk profile.
Q: Are the Taiwan stock prices in the morning briefing from today?
No. When the morning briefing goes out at the default 08:30, the Taiwan market hasn't opened yet, so its figures are always the previous session close. Pre-open auction indicative prices from 08:30 to 09:00 are not used, and they never change its stop-loss or take-profit calls. Intraday prices are labeled as intraday, with the closing price as final.
Q: What cost is P&L based on? Are dividends included?
It uses the original cost you registered in the dashboard, calculated by a program. If it also shows a return including dividends, that's listed separately from the return on original cost, with both numbers spelled out.
Q: Can I ask it to check whether a rumor is true?
Yes. Just ask on Telegram, for example: "I heard Company A landed a big order. Is there any news on that?" If it finds something, it cites the source and date; if not, it says "Not found." It won't cobble together an answer just to give you one.
Further reading
- Analysis method: technicals, news, sentiment, and market context
- Read your morning briefing in 5 minutes before the open
- It remembers what it said: investment notes and checking its calls
- Stop-loss discipline: intraday alerts before losses grow
- Stock Helper FAQ
Stock Helper is not an investment advisory service. AI analysis is for reference only and does not constitute investment advice. Investing involves risk; please make your own decisions.
