Guide

For a page on one of your sites — a city, a neighborhood, a subdivision, a property type, or a single building — this tool reads the market numbers Sierra already shows on that page (how many listings, the median price, average days on market, average price per square foot) and writes a short paragraph about them. You review that paragraph here, then paste it into Sierra's own page editor yourself. From then on, the tool keeps checking those numbers and keeps the paragraph accurate — quietly fixing small drift on its own, and flagging the pages where something bigger changed and it's worth a fresh look.

Everything it works from is already public on the page itself — the same stats block and listing addresses any visitor (or Google) can already see. It never logs into Sierra and can't reach anything private.

It also automatically tells apart a for-sale page from a for-rent one (and a single building that lists both), and writes the right vocabulary for each — “rental listings” and a monthly “asking rent” on a rentals page, never “homes listed” or a “list price” that reads like a purchase price. See Templates below.

Add a site. Go to Sites in the left menu, and fill in the site's name, its domain, and the most often its pages should be checked (12 hours, the default, is the normal choice — each page then adapts from there on its own). You can change that for the whole site later, or set an exact interval for one page at a time.

Add a page. There are two ways:

• Paste one URL into Quick workspace and it runs immediately — good for adding a single page or checking one you're curious about.
• Open a site and use Import URLs to paste a whole list (or upload a CSV) — the right way to add many pages at once. See Importing a list of pages below.

What happens the first time a page runs: the tool fetches the page, reads its stats block, and writes the first paragraph — all in one step, generally a few seconds. If the page can't be read for any reason, nothing gets guessed or invented — see When something looks wrong.

Open any tracked page and you'll see four number tiles (listings, median price, days on market, price per square foot) above the written paragraph. Every figure inside the paragraph itself is underlined — hover or tap one and a small note pops up showing the raw number, exactly which line on the source page it came from (e.g. the stats block's “Med. List Price” row), and when it was captured. That's worth doing any time a figure looks off — it's the fastest way to check whether the tool got something wrong or the site itself is showing something unexpected.

Refresh vs. Regenerate — these do different things and it's worth keeping them straight:

• Refresh data goes back to the live page, re-reads the numbers, and writes new wording if anything meaningful changed. Use this when you think the numbers themselves might be out of date.
• Regenerate does not re-check the page — it just asks for different wording from the same numbers already on file. Use this if the numbers are right but you just want the paragraph phrased differently.

A missing figure (a blank tile with a dash) is never a zero standing in for a real one — it means the source page simply doesn't publish that number for this kind of listing (common for property types where price-per-square-foot isn't meaningful, like multi-unit buildings). Hover the ⓘ on the tile for the exact reason.

Four templates cover the shapes a page's content can take. The tool picks automatically:

• Standard — the normal case: a city, neighborhood, subdivision, or property-type page with real listings.
• Building — picked automatically when every active listing on the page turns out to be on the same street (the signature of a single-building page). With a handful of units, the wording adjusts to match: a very small count (one or two active listings) is described plainly — no “median,” “average,” or pace framing, since there isn't really a sample to characterize — while a slightly larger small count keeps a caveat that one sale can still move the numbers noticeably.
• Zero inventory — picked automatically when nothing currently matches the page's search. A short, plain note that nothing's listed right now, with no invented numbers.
• Needs review — picked automatically when the page's saved search is pulling in listings from outside its own area (see Statuses). No figures are used at all here, since none of them can be trusted to describe the area the page claims to.

You can override the choice yourself from the Template dropdown above the paragraph — picking a different one immediately writes new text in that style, from the same numbers. This is a one-off override for that page; it doesn't change how future automatic runs classify it.

For sale, for rent, or mixed is a separate thing the tool detects on top of the template above — a rentals page is still a “Standard” page, just written with rental vocabulary. It looks at the page's own URL and title, and — more reliably, since it reflects what's actually listed right now — whether the individual listings themselves are priced “per month.” You can see which way it read a given page in that page's Extraction details panel, under Listing kind.

A single building that lists both for-sale and for-rent units at the same time is the one case with no single price to report: a median mixing sale prices and monthly rents wouldn't mean anything, so the paragraph mentions the listing count and how quickly things are moving without a blended price figure.

On any tracked page, next to the refresh-interval control, there's a Scope setting: Auto-detect, Municipal, or Regional. This controls one specific check: whether the tool compares each listing's own city against the page's declared city and flags the page as Needs review when too many don't match.

• Municipal means that check runs — right for a page about one specific city or neighborhood, where every listing really should be there.
• Regional turns that check off — right for a page that's intentionally county-wide or spans several cities by design, where listings from more than one town are expected, not a problem.

Left on Auto-detect (the default), the tool guesses from the page's own title and URL — and it can guess wrong in either direction:

• A page that's genuinely county- or region-wide keeps landing in Needs review even though nothing is actually wrong — set it to Regional to stop the false flag.
• A page that's genuinely about one city has an unusual title the tool can't confidently read, so it gets treated as regional by default — which means a real misconfigured search on that page wouldn't get caught. Set it to Municipal to turn the check back on.

Publishing in the left menu shows two separate lists, and they answer different questions:

• Not yet published — pages with a written paragraph that has never been pasted into Sierra at all.
• Published, but changed since — pages you already published, where what's live on the actual site no longer matches what's current here. Each row says which of two reasons applies, because they're different jobs: the figures have moved, or the module's whole format changed and the text reads differently now.

Expect that second reason on every already-published page, once, right after the new module format ships — the wording changed for all of them, so all of them need re-pasting. After that one pass it goes back to being rare.

They're kept separate on purpose: a page can be perfectly up to date in this tool while never having been published yet, or it can be published and still showing exactly what you last pasted even though its data is old — those are two different questions (“is the data current” vs. “does the live site match what this tool has”), and conflating them would hide real work in one or the other.

From a site's page, click Import URLs. Paste one URL per line, or upload a CSV (the first column is used; a header row is detected and skipped automatically). A few things worth knowing before you start a large import:

• A URL that's already tracked, or repeated twice in the same paste, is skipped rather than run twice — you'll see it marked Skipped, not silently dropped.
• Partial success is the normal outcome on a large batch — a handful of failures doesn't mean the whole import failed. Failed rows can be retried on their own once you see why they failed.
• The import keeps running if you close the tab. It runs on the server, one page at a time with a short pause between each — the same politeness any large batch against real, live sites needs. Come back to the same import link any time to see where it stands.
• You can export the batch's results as a CSV once it's done, or at any point along the way.

From your browser (the default). Instead of this tool reading each new page, your own browser can: create the list, drag the Capture pages bookmarklet to your bookmarks bar, open any page of that site, and click it. Your browser reads one page on the list every 15 seconds, with no cookies and a header saying what it is, and sends each page here to be read exactly as a fetch would be. This tool sends no request.

• It stops by itself if the site pushes back, and never retries around it. Only allow it again once you know why it stopped.
• A page is turned down, with the reason shown, if it isn't the page on the list (it redirects, or says it's a different page), if its figures can't be read, or if they don't look real.
• Whatever isn't captured within 48 hours — or when you press Hand the rest to the server fetch — is fetched by this tool instead, whenever fetching is allowed.
• The bookmarklet is shown once and works only for its list. Treat it like a password; make a new one from the list's page if you lose it (the old one stops working).
• It reads from your internet connection. If the site objects, it's your address it may slow down for a while — and whether your agreements with the site owners cover it is yours to check.

Any page list has a checkbox on each row. Tick one or several — or use the checkbox in the header to take everything on the current page — then click Delete selected. Filter first (by status, or by Duplicate URLs only) and the selection follows the filter, which is how you clear a whole category without working down the list one row at a time.

There is no undo. Deleting a page also destroys its snapshot history — every set of figures ever captured for it, which is what a year-over-year comparison would eventually have been built from — along with every generated paragraph and its publish record. Import history is kept: the batch a page arrived in stays intact and still reports the same totals, it just no longer links to the page.

If you only want to stop refreshing a page, deactivate it instead — that keeps everything and can be undone. Deleting is for pages you don't want tracked at all. A deleted URL can be re-added later, by import or through the quick workspace, and comes back as a clean page with no history attached.

Duplicates. A page added once with a trailing slash and once without used to create two separate rows, and both are still tracked and both are still fetched on every refresh. When that happens, the list says so: each copy is marked Duplicate URL and shows its full address underneath, so you can tell the two apart, and the Duplicate URLs only filter narrows the list to just them. Delete whichever copy isn't the published one.

Two things worth expecting. If a refresh happens to be running on a page at the moment you delete it, that one page is refused with a note saying so — the refresh would have re-created it seconds later — and everything else in your selection still goes; try that page again in a minute or two. And the counts update immediately: Needs attention and the per-site totals stop counting deleted pages straight away, rather than showing a number that no longer has rows behind it.

Every tracked page gets checked automatically, with no one involved. The sweep runs twice a day:

• 11am Central, every day — after the morning wave of new MLS listings, which agents typically enter at the start of the business day.
• 5pm Central, weekdays only — catches what came in during business hours. Weekends skip this one; far less hits the feed and it keeps the load down.

A run doesn't visit every page at once. Its visits are spread across the four hours after those times, starting at a random moment in the first twenty minutes, with uneven gaps between pages on the same site and the sites taken in a different order each run. Pages on one site are still visited one at a time. A run that is still going mid-afternoon is working as intended.

Most runs, most pages get checked and nothing happens — and that silence is the point, not a bug. Here's the actual rule for what counts as worth acting on:

• Ordinary day-to-day drift doesn't trigger a rewrite. Days-on-market ticking up by one, or a price moving by a small amount, gets quietly corrected in the existing sentence — the number updates, the wording around it doesn't change, and nothing gets rewritten from scratch.
• A real change does. The listing count changing at all, a price moving by a meaningful amount, the page going from having listings to having none (or the reverse), or the paragraph naming a street that's no longer listed — any of those gets a genuinely new paragraph written.
• So does a change to the module format itself. If a page's paragraph is still in an older format, it gets rewritten on the next run even when not a single figure has moved — otherwise a page whose numbers happen to be stable would sit in the old wording forever.

The Refresh log in the left menu shows every automatic run and what it did to each due page — unchanged, a quiet value update, a real rewrite, or a failure. It also tells you plainly if the job hasn't run recently enough to be firing at all, rather than leaving that to guesswork — and it will say so if one of the two daily runs stops happening while the other keeps going, which is the kind of half-failure that otherwise looks completely normal.

The schedule above is when the tool looks. Whether a given page is refreshed on a given run depends on that page's interval, in hours — and that interval now adapts on its own to how often the page's figures actually change.

• Steady pages are checked less often. After two automatic runs in a row where nothing material changed, a page's interval widens one step: 12 hours → 1 day → 2 days → 4 days. Only the automatic runs count: refreshing a page yourself, however often, never changes its interval.
• Changing pages are checked more often again, straight away. When a refresh finds a real change, the interval narrows two steps at once.
• Never longer than 4 days. However steady a page is, it's looked at again within about four days, so nothing is left to go quietly out of date.
• A failed refresh changes nothing. A page the tool couldn't read has taught it nothing about how often that page changes, so its interval stays exactly where it was.

For this purpose a real change is: the listing count moving by 5% or more (at least one listing), a page going to or from zero listings, the median price or price per square foot moving 5% or more, the page switching between sale and rental listings, its results starting or stopping spilling outside the area it covers, or a street the paragraph names leaving the market. Smaller movements are still picked up whenever the page is looked at — they just don't make it looked at more often.

The site's number is the most often its pages are checked. Pages adapt from there, never below it. 12 hours (the default) lets a page be refreshed at every run; 24 means never more than once a day, on the 11am run; 96 or more fixes the site's pages at that interval, since there's nothing left to adapt.

A number you type on a page always wins. Enter hours in that page's own Refresh every box and that exact interval is used, whatever the page's figures are doing. Clear the box and the page adapts again, starting afresh from its site's number. The page list shows every page's interval and whether it's adapting or set by you, and the page itself says why.

This used to be measured in days. Existing settings were carried over unchanged: a site or page that was set to 2 days is now 48 hours, 7 days is 168 hours, and so on. The one deliberate exception is sites that were on the old “1 day” default — they moved to the new 12-hour default, because leaving them at 24 would have meant the new 5pm run had nothing to do on any of them. If a site was genuinely meant to be once-daily rather than just sitting on the default, set it back to 24.

This tool has no direct connection into Sierra — publishing means copying the text here and pasting it into Sierra's own page editor yourself, then telling this tool you did.

Open a site's Publish flow and it shows you one page at a time, in priority order. For each one:

• Copy plain text or Copy HTML — plain text for a plain-text field, HTML if Sierra's editor expects a wrapped paragraph. Use HTML if you want the bold first sentence: plain text is characters only, so the bold can't travel with it. The HTML copy and the embed block both carry it.
• Paste it into Sierra, then come back and click Mark published & next — this records that you published it and immediately moves you to the next one in the queue.
• If you mark something published by mistake, an Undo appears for a few seconds right after — use it right away, since it won't stay on screen.

Keyboard shortcuts once you're in the flow: C copies plain text, H copies HTML, Enter marks published and advances, U undoes the last one.

Instead of (or alongside) pasting plain text, you can generate a small self-contained block that includes the written paragraph plus a tiny bit of code.

Open any tracked page's own workspace (from its site's page list, not the ad-hoc /workspace tool) and click Copy block, next to Copy paragraph. It copies the complete paste-ready block straight to your clipboard — nothing else to build or configure first.

Two things worth knowing about what that block actually does:

• The paragraph itself is real, complete text from the moment it's pasted in — a search engine reading the page sees the full sentence immediately, with nothing waiting on a script to run.
• The small script only checks in afterward to quietly update the figures in that paragraph if they've moved since it was pasted — the wording around them stays exactly as written.

Blocks copied now also report what the page shows. When someone views the page, the same check tells this tool what Sierra's statistics box on that page says — the figures, the date, and the street names on the listing cards — and nothing about the visitor. That report never changes the figures on the page: those still only come from this tool reading the page itself. What it changes is when the tool reads it. If the visitor's page matches what we have, the next read can wait (up to a week). If it shows something different, the tool reads the page sooner to check.

• Blocks pasted before this still work, but don't report. The Publishing page lists exactly which published pages still have one, once a visitor has loaded them — copy those pages' blocks again and paste over the old ones. The paragraph is identical; only the script line changes, and future changes to the script won't need another re-paste.
• Before re-pasting everything, paste one page and check it: open the page, then look at it in this tool — its Freshness panel shows Seen by visitors once the report arrives.
• A page's list row shows both: Fetched by us (when this tool read the page) and Seen by visitors (what visitors' pages last reported). They aren't the same thing, and a page waiting on visitors shows as Fresh on purpose.
• Under the latest report, each page says how its reports count. Quorum: busy enough that three different visitors' networks reported within half an hour this week — when three agree the page matches, the next read can wait up to two weeks. Single reports: quieter pages, which work as described above. Telling networks apart happens only in memory, for counting, and is forgotten every half hour; nothing about any visitor is stored.

• It can't edit or publish anything on Sierra directly — every publish is a manual copy/paste.
• It can't fix a mis-scoped saved search (a “Needs review” page) — that's a setting on Sierra's side and has to be corrected there.
• It can't write free-form content — no newsletters, no blog posts, no anything outside a short market-summary paragraph built from the four templates above.
• It can't show per-listing detail (individual prices, beds, baths) — only street names, because that's all the source pages themselves publish for it to read.
• It can't discover pages on its own — a page has to be added, one at a time or by list, before it's tracked.
• It can't get past a site that's actively blocking it — that needs the site's host to allow this tool through; no setting here fixes it.

• One page shows Failed. Open it and read the message — it says plainly whether the site is blocking automated requests, the page moved, or the site had a server error. Nothing was overwritten; the last good paragraph is still showing underneath.
• Several pages across different sites fail at once, especially with a “server error” message. Check status.sierrainteractive.com before assuming this tool is broken — that's Sierra Interactive's own status page, and a platform-wide outage there looks exactly like this tool failing.
• Everything looks stale and nothing's updating. Check the Refresh log — it states outright if the automatic job hasn't run recently enough, rather than leaving you to guess from a wall of stale badges.
• A brand-new page shows nothing. It just hasn't been run yet — open it and click Extract listings.