EDMW's Front Page Turns Over Every 6 Minutes: How to Actually Monitor Singapore's Biggest Forum

In one line: Singapore’s loudest public conversation runs on a forum with no API — on 10 September 2026 at 12:49 SGT every one of the 20 non-pinned threads on EDMW’s front page had been replied to inside the previous 6.4 minutes, and that one board runs to 32,456 pages — so the question is not whether to monitor it but how often, and which parts you can afford to expand.

Singapore’s public opinion doesn’t live where your tool is looking

Social listening suites are built on platform APIs. X, Facebook, Instagram, YouTube, news wires — each one has an endpoint, a rate limit, and a vendor relationship behind it. That architecture decides what the product can see.

HardwareZone’s forums have none of that. It is a XenForo board — HTML pages, no public API, no partner programme. Its index lists 100 boards, and the busiest of them, board id 16, is called Eat-Drink-Man-Woman. The name is a joke inherited from the early 2000s; in practice EDMW is the site’s general chat board, and it is where Singaporeans argue about the exchange rate, hiring, politics, and which phone to buy.

So a brand can be quiet in every dashboard it pays for and still be the subject of a 200-reply thread. That is not a gap in anyone’s coverage list — it is a structural consequence of building on APIs when the conversation is on a forum.

What “busy” actually means here

Numbers make this concrete. Pulling EDMW’s first page on 10 September 2026 at 12:49 SGT:

Read the third row again. The twenty threads visible on page one had all been replied to within about six minutes of each other. Page one is not a daily digest; it is a live tail that empties several times an hour.

Two consequences follow, and they set the whole design of a monitor:

1. A once-a-day scrape of page one is not sampling — it is a snapshot of six arbitrary minutes. You will miss the overwhelming majority of the day.
2. Paging deeper is the wrong fix. At 32,456 pages that board alone is several hundred thousand threads. Depth costs money linearly and still misses anything posted after you started.

The right lever is frequency plus a cutoff, which is what the actor’s dateFrom is for.

Pull the board index first

forum mode returns the thread listing for one or more boards. This is the cheap pass, and it should carry your filtering:

{
  "mode": "forum",
  "forumIds": [16, 210, 17],
  "keywords": ["singtel", "starhub", "dbs"],
  "dateFrom": "2h",
  "maxItems": 50
}

keywords filters on thread title, dateFrom accepts a relative value like 2h or an absolute timestamp, and maxItems applies per board. The actor stops paging a board early once a whole page of non-pinned threads falls before your cutoff — so a run scheduled every few hours reads two or three pages, not thirty thousand.

Each listing row comes back as:

post_id, board, title, author, reply_count, view_count, created_at, last_reply_at, post_url, source_mode, source_query, scraped_at

One detail worth knowing: HardwareZone displays view counts rounded, as 14K or 1.2M. The actor normalises those to integers (14000, 1200000), so view_count is always a number you can sort and threshold on — but it inherits the site’s rounding, so treat it as a magnitude, not a precise count.

Here are two real rows from that same pull, to show the shape of what you get:

Neither is a tech-support post. One is purchase intent for a product that has not launched; the other is a currency thread that runs for days and tells you how Singaporeans are feeling about cross-border spending. That is the signal you are paying for.

Then expand only the threads that matter

thread mode fetches a thread’s main post plus every reply, across pages:

{
  "mode": "thread",
  "threadUrls": [
    "https://forums.hardwarezone.com.sg/threads/7220513/"
  ],
  "maxItems": 500
}

You can pass threadIds instead — bare numeric ids work, and the site redirects them to the slug form. Replies come back as reply_id, post_id, floor, author, content, created_at, scraped_at. The floor field is the #12-style post number, which is what you need if you want to quote a specific reply or reconstruct who answered whom.

Expand deliberately. The next section explains why.

The boards that actually carry commercial signal

The forum index has 100 boards, but a monitor does not need all of them. These are the ids worth starting from:

Pass them as forumIds, or paste a list into bulkInput one per line.

Same fields as Hong Kong and Taiwan

hardwarezone-scraper-en shares its field layout with the other forum actors — LIHKG-scraper for Hong Kong, ptt-scraper and mobile01-scraper for Taiwan, hk-forums-scraper for the smaller Hong Kong boards. post_id, board, author, reply_count, created_at and last_reply_at mean the same thing in each.

That matters more than it sounds. It means a sentiment or alerting job downstream reads one schema and covers three markets, with no per-site branching — the only thing that changes is which actor filled the dataset. If you want the plumbing for that, the n8n forum sentiment pipeline walks through the schedule → scrape → score → alert loop end to end, and it works the same with a Singapore board as a Taiwanese one.

It also means the evidence is comparable across markets. A worked example of turning forum threads into something you can hand to somebody else is in the LIHKG deepfake endorsement write-up — the collection method there transfers directly.

What it actually costs

Three events, and the gap between them is the whole budgeting story:

Worked through, watching three boards at 50 new threads a run, four runs a day:

• Listings: 600 rows/day × $0.002 = $1.20/day
• Expanding 20 of those threads: 20 × $0.008 = $0.16/day
• Their replies, at ~50 each: 1,000 × $0.003 = $3.00/day

The replies alone cost more than twice the listing pass and the main posts put together. This is the single most important thing to internalise: the listing pass is cheap, so do your filtering there. Use keywords and reply_count to decide which threads deserve expansion, and expand tens of threads a day, not hundreds. A monitor built that way runs a few dollars a day; one that expands everything on page one runs an order of magnitude more for very little extra signal.

Three ways to start

From least effort to most freedom

I run it for you Managed, daily to your sheet You do nothing. From US$299/mo. Use it here Run the free tool Instant, no setup, on this site. Run my actor On Apify, your account Same data, you run it and pay per result. Build your own Sign up for Apify Free account — build the pipeline yourself.

This post contains Apify affiliate links (no extra cost to you). Costs and comparisons are stated honestly, including where Apify is not the best fit.

FAQ

What is EDMW, and why does it matter more than the tech boards? EDMW stands for Eat-Drink-Man-Woman, board id 16 on HardwareZone’s forums. Despite the name, it is not a food board — it is the site’s general chat board and its highest-traffic one, where Singaporeans argue about currency rates, employment, politics, phones and everything else. Measured on 10 September 2026 at 12:49 SGT, its front page held 24 rows, 4 of them pinned, and the 20 live threads had all received a reply within the previous 6.4 minutes. The forum index lists 100 boards in total, but EDMW is where general consumer and public sentiment concentrates.

Why can’t I just use a social listening tool for Singapore? Mainstream listening suites are built on platform APIs — X, Facebook, Instagram, YouTube, news wires. HardwareZone is a XenForo forum with no such API, so it falls outside that model structurally rather than by choice. When we built this actor in August 2026 the Apify Store had no HardwareZone actor at all. The practical consequence is that a brand can look quiet in every dashboard it pays for while a 200-reply thread about it sits on EDMW.

How often do I need to run it to not miss things? Front-page turnover sets the floor. Two measurements eleven minutes apart on 10 September 2026 put the span of the 20 live threads at 5.8 and 6.4 minutes, so page one is completely replaced several times an hour. Do not try to keep up by paging deeper — use dateFrom instead. Give it a relative value like 2h and run on a schedule; the actor stops early once an entire page of non-pinned threads is older than your cutoff, so a frequent run costs less than a deep one.

What does a realistic month actually cost? Listings are cheap and replies are not. Board listings bill as search-listing at $0.002 an item, a thread’s main post bills as product-detail at $0.008, and every reply bills as review-item at $0.003. Watching three boards at 50 new threads per run, four runs a day, is 600 listings a day — about $1.20. Expanding 20 of those threads with roughly 50 replies each adds $0.16 for the main posts and $3.00 for the replies. The replies alone cost two and a half times the listing pass, which is why the keyword filter belongs in the first pass, not the second.

Does this give me Hong Kong and Taiwan too? Not from this actor, but from the same field shape. hardwarezone-scraper-en emits the same listing and reply fields as LIHKG-scraper for Hong Kong, ptt-scraper and mobile01-scraper for Taiwan, and hk-forums-scraper for the smaller Hong Kong boards. Because post_id, board, author, reply_count, created_at and last_reply_at line up across all of them, one downstream sentiment job can read all three markets without per-site branching.

Chad runs 60+ published Apify actors, including the hardwarezone-scraper-en used here. Every measurement above — the 100 boards, the 6.4-minute front-page span, the 32,456-page count — was pulled live from HardwareZone on 10 September 2026 and can be re-run against the same board.