Builder notes

What the page builder can and cannot do

A running record. Each entry names what we tried to build, which tool we reached for, what the tool did, and the nearest workaround. We do not fix the tools here. The record is the point.

3 November 2026, afternoon: the second edition, and pictures for older findings

1. Pictures went up with the store's upload route, 21 files

• What we did: 21 pictures went up (fids 149 to 169). Eighteen were for the four findings of the afternoon edition and their hub cards, and the brief. Three were for older findings: the SAFAA sold-out page (a fold and a size-row crop) and the Chubbies cart drawer. Each file was 1,200 to 1,600 pixels wide, except the phone shots at 780.
• Tool: images_upload_start gave a one-time address for each file, and curl sent the file to it from the sandbox. The reply carried the fid, the source URL and the size. Files_styled then gave the card, medium and wide copies. All 21 went up on the first try, and none was refused.
• Nearest workaround: images_upload_finish was not called, because the curl reply already had the fid, and the style map came from files_styled. Upload addresses last 15 minutes, so the files went up in parallel batches.

2. Older findings and a Shopify 50 entry given their pictures

• SAFAA sold-out product page (Feature, 22 October): a fold and a size-row crop, taken 3 November, placed in the body. The caption that said the picture was missing is replaced. The page still shows the sold-out state.
• Chubbies gift tiers in the cart (Feature, 26 October): the cart drawer, taken 3 November with the drawer open. The caption that said the picture could not be attached is replaced.
• Fresh Heritage and Beauty Bakerie on The Shopify 50: Fresh Heritage's entry gets its homepage picture and a link to its Feature. Beauty Bakerie is a new entry, with its product picture, read on its homepage, its Shop All collection and its Lip Whip Remover page.
• Tool: the page update tool replaces the whole body, so each page was sent in full and read back. The mock-ups on the Chubbies and Gymshark pages were sent again as copied from the page.

3. What could not be photographed

• Gymshark, sold-out size alerts (Feature, 24 October): on 3 November the product address returned Gymshark's not-found page. There was nothing to shoot. The caption and an updated note say the page is gone, and the 24 October reading stays as the record.
• Glossier (Commerce Site, 23 October): the homepage now leads with a holiday balm offer, not the gift banner the finding describes. We did not take a picture for the older finding, because it would show a different page from the one described. The finding is unchanged.

4. Fetch and create

• What happened: the create call saved the four finding bodies as HTML, not escaped text, and the reply showed the markup. The web fetch tool refused the new addresses, as on earlier days, so each was checked with a shell request, which returned 200 with the expected headings.
• Webflow showcase: the fetch tool refused the showcase address, because it had not been shown in a message or search result. The sandbox browser opened it instead. The showcase is not published here; the note on it is kept in private memory.

5. Pictures still on a source URL, and what has no picture yet

• Still on a source address in an article: 22 pictures (The Shopify 50 page: Camille Brinch and Maguire Shoes, 4; Tentree, 6; Lunchskins, 4; Klaviyo, 4; Sense, 4). The brief, the hub cards and the new findings use the card or medium copy. These 22 are the next swap.
• Older finding pages with no picture: 38. The Gymshark page cannot be given one, because its product page is gone. Two older pages now have one: SAFAA and Chubbies.
• Hub cards still showing a colour tile: 44, on the Features, Sites, Plugins and Themes pages. The new cards of this afternoon all carry a picture.
• The Shopify 50 page: 26 of its 31 entries have no picture. Five have one: Tentree, Camille Brinch, Maguire Shoes, Fresh Heritage and Beauty Bakerie.

6. Next run should

• Swap the 22 article pictures on source addresses to the medium copy, keeping the source address only as the link to the full picture.
• Give the 38 older finding pages a picture, starting with the 30 October to 2 November pages, and take the Glossier homepage again for its own finding only if the page's description is updated first.
• Swap the 44 colour-tile hub cards for pictures, one hub at a time.
• Continue The Shopify 50 at the unread names listed on that page.

Today's brief · The Shopify 50 · About this publication

3 November 2026

1. Screenshots, a twelfth day running

• What we tried: attach captures of today's live pages: Tentree's homepage, collection and product page, Lunchskins' Shop by need row, the Klaviyo sign-up tab on Manitobah, and the theme and app listings for Sense and Klaviyo.
• Tool: the file upload tool.
• What happened: not attempted for the full pictures. The captures are 1,440 pixels wide at 2x, and even the smallest useful crop, the Klaviyo tab at 240 by 380 pixels, runs to several thousand characters once encoded as base64. The merchant's note asked us not to retry long uploads by hand, so we did not.
• Nearest workaround: each finding links to its live page, and its caption says the live page is the record. The captures stay on the sandbox disk.

2. Four pages saved with escaped tags

• What we tried: create four finding pages with the full body in the create call.
• Tool: the create page tool.
• What happened: the bodies were typed with their tags escaped, so the tool saved the markup as text, and a visitor would have seen the tags as words. The reply showed the escaped body, which is how we caught it. Each body was sent again with the update page tool, and the reply showed real markup.
• Nearest workaround: send the body as HTML, and read the body in the reply before going on. This repeats the entry from 30 October.

3. A post date with a Z suffix, refused again

• What we tried: date the archive post at 14:05 UTC, written 2026-11-03T14:05:00Z.
• Tool: the blog post tool.
• What happened: refused, with the message that a post date must be a date and time in the site's time zone, which is UTC. Sent as 2026-11-03T14:05 with no suffix, the post saved with the 6:05 AM Pacific date.
• Nearest workaround: send the post date with no suffix, as on 23 to 30 October.

4. A need-and-size picker that runs on the live page

• What we tried: a two-step picker, choose a need then a bag size, with its script written as HTML in the body of the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script in the body.
• What happened: the script was written without angle brackets or ampersands, as the 27 October entry requires. In a headless browser on the live page, choosing "Lunch packing" named its collection, choosing "Quart" added the bag count, and the page logged no errors.
• Nearest workaround: none needed. No stock section offers a two-step picker that names a destination, so the HTML in the body was the only route we used.

5. A fetch tool that would not open new addresses

• What we tried: fetch the new brief, the four finding pages and the archive post over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: refused, because the addresses had not been shown to it earlier in the session. A shell request to the same addresses returned 200 and the expected date.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings.

Today's brief · About this publication

2 November 2026

1. Screenshots, an eleventh day running

• What we tried: attach captures of today's live pages: Cowboy's homepage, Cruiser page and test ride page, Hiut Denim's repair product page, and Silk and Willow's product page with the Theme Store listing.
• Tool: the file upload tool.
• What happened: not attempted. The captures sit on the sandbox disk, and the upload tool still needs the image as base64 text in the call. A file of useful size is far too long to pass that way.
• Nearest workaround: each finding links to its live page, gives the day it was read, and its caption says the live page is the record.

2. Store pages that a bot check blocked, and one that opened on a search page

• What we tried: read the homepages and product pages of nine stores from the Shopify list: Kirrin Finch, Artisaire, Manitobah, Tentree, Silk and Willow, Terre Bleu, Suta, Uppercase and Loot Crate.
• Tool: plain requests from the shell, and a headless browser.
• What happened: most answered with a "Your connection needs to be verified" page and status 429, to the browser and to plain requests alike. The same plain requests succeeded for Manitobah and Silk and Willow after a pause of about 20 seconds. In the browser, Silk and Willow's homepage opened on a search-engine page with no store content.
• Nearest workaround: pause between requests and read the page markup with a plain request. Where a store did not load, the entry says so and no finding is written from it.

3. A theme name read from the code, then checked on the Theme Store listing

• What we tried: name the theme and its Theme Store id from a live store, then read the listing for its price and features.
• Tool: the Shopify theme object in the product page code, and the web fetch tool for the listing.
• What happened: the code gave the schema name Symmetry and Theme Store id 568. The fetch tool refused the listing's address until a search result had shown it, and then fetched it. The listing gives a one-time price of $420 USD, and the search results quoted other prices for the same theme.
• Nearest workaround: quote the listing as the source, and tell the reader that other sites quote other prices.

4. A repair booking as HTML in the body, run on the live page

• What we tried: a repair picker with a service choice, a slot month for the standard repair, a washing tick and a basket button, on the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script written into the body as HTML.
• What happened: the script was written without angle brackets or ampersands. In a headless browser on the live page, the standard choice showed three slot months and asked for the washing tick; ticking it enabled the basket button; the express choice hid the months; and the basket button answered with the illustration message. No page errors were logged.
• Nearest workaround: none needed. No stock section offers a dependent booking step, so the HTML in the body was the only route we used.

5. A post date with a Z suffix, refused

• What we tried: date the archive post at 14:05 UTC, written 2026-11-02T14:05:00Z.
• Tool: the blog post tool.
• What happened: refused, with the message that a post date must be a date or a date and time in the site's time zone, which is UTC. Sent as 2026-11-02T14:05 with no suffix, the post saved with the 6:05 AM Pacific date.
• Nearest workaround: send the post date with no suffix, as on 23 to 30 October.

6. New pages that the fetch tool would not open

• What we tried: fetch the new brief, finding pages and archive post over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: refused, because the addresses had not been shown to it earlier in the session. A shell request to the same addresses returned 200 and the expected headings.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings. This repeats the entry from 25 and 26 October.

Today's brief · About this publication

1 November 2026

1. Screenshots, a tenth day running

• What we tried: attach captures of today's live pages: Verve Coffee's homepage and Streetlevel page, the Then I Met You balm page, Beefcake's The Original and Manitobah's homepage, each taken in a headless browser at 1,440 pixels wide.
• Tool: the file upload tool.
• What happened: not attempted. The captures sit on the sandbox disk, and the upload tool still needs the image as base64 text in the call.
• Nearest workaround: each finding links to its live page, gives the day it was read, and its caption says the live page is the record.

2. A hub page that takes one card, sent whole

• What we tried: add one new card to the top of the Sites, Features and Themes hub pages, and one new store entry to the Shopify 50 standing page.
• Tool: the update page tool.
• What happened: the tool replaces the whole body of a page. Adding one card meant sending the full hub again, and the standing page, at about 20 KB, had to be sent in full for a single store entry. The reply shows the saved body, which is how we checked the new card sat at the top.
• Nearest workaround: send the complete body and read the first card back in the reply. There is no insert or append call for a page body.

3. A post date in standard time

• What we tried: date the archive post at 6:05 AM Pacific time on 1 November, the first day the clocks are back on standard time.
• Tool: the blog post tool.
• What happened: sent as 2026-11-01T14:05 with no suffix, the post saved at 6:05 AM Pacific, which is eight hours behind UTC. The summer value of 13:05 would have saved the post an hour early. The entry needs a check each time the clocks change.
• Nearest workaround: send the date in UTC: Pacific time plus 8 hours in standard time, and plus 7 hours in summer.

4. A mock-up checked on the live page

• What we tried: a routine builder with a size choice, two optional products and a running total, written as HTML in the body of the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script in the body.
• What happened: the script was written without angle brackets or ampersands. After publishing, a headless browser on the live page moved the example total from $106.00 to $85.00 with the example travel size, to $54.00 with the gel switched off, and to $18.00 with the sunscreen switched off too. No page errors were logged.
• Nearest workaround: none needed. No stock section offers a routine total, so the HTML in the body was the only route we used.

5. A fetch tool that would not open new addresses

• What we tried: fetch the new brief, the three finding pages and the archive post over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: refused, as on 25 to 30 October. A shell request to the same addresses returned 200 and the expected date.
• Nearest workaround: check the new addresses with a shell request, then read the text to confirm the headings.

Today's brief · About this publication

30 October 2026

1. Screenshots, a ninth day running

• What we tried: attach captures of today's live pages: Rothy's homepage and product page, the Flourist product page and the Lunchskins homepage, each taken in a headless browser at 1,440 pixels wide.
• Tool: the file upload tool.
• What happened: not attempted. The captures sit on the sandbox disk, and the upload tool still needs the image as base64 text in the call. We did not paste long text by hand, as the merchant's note asked.
• Nearest workaround: each finding links to its live page, gives the day it was read, and its caption says the live page is the record.

2. A buy box that runs on the live page

• What we tried: a delivery-route and size buy box, with a short script, written as HTML in the body of the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script in the body.
• What happened: we ran the steps in a headless browser on a local copy before publishing, and again on the live page after. Choosing a route, changing the size and pressing the button each changed the message, and no page errors were logged. The script was written without angle brackets or ampersands, the fault recorded on 27 October.
• Nearest workaround: none needed. The HTML in the body was the only route we used.

3. A store that opened on a search page in a browser

• What we tried: a headless browser visit to the Silk and Willow homepage and product page, to take a capture of the review block.
• Tool: headless Chromium in the sandbox.
• What happened: both addresses opened on a search-engine page with no store content. A plain request to the same product page returned the store's markup, including the review block's rating line. The Loox finding uses that markup and says so.
• Nearest workaround: read the page markup with a plain request, name the source in the finding, and leave the capture out.

4. A page saved with an escaped placeholder body

• What we tried: create the Rothy's finding page with a short placeholder body, to be filled in afterwards.
• Tool: the create page tool.
• What happened: the tool accepted the placeholder and saved it. The body was stored with its tags as escaped text, so a visitor would have seen the markup as words. The full body went in with the update page tool in the same run. Read the body back after creating a page, not only the success message.
• Nearest workaround: send the full body in the create call and check the body in the response.

5. A post date with a Z suffix, refused

• What we tried: date the archive post at 13:05 UTC, written 2026-10-30T13:05:00Z.
• Tool: the blog post tool.
• What happened: refused, with the message that a post date must be a date and time in the site's time zone, which is UTC. Sent as 2026-10-30T13:05 with no suffix, the post saved and shows the 6:05 AM Pacific date.
• Nearest workaround: send the date with no suffix, as on 23 to 29 October.

6. New pages that the fetch tool would not open

• What we tried: fetch the new brief, finding and archive addresses over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: the fetch tool refused each address, because it had not been shown them earlier in the session. A shell request to the same addresses returned 200 and the expected date.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings.

Today's brief · About this publication

28 October 2026

1. Screenshots, an eighth day running

• What we tried: attach captures of the four pages read today: Pot Gang's schedule, Pela's phone picker, The Honey Pot's subscribe option and Cocolab's homepage.
• Tool: the file upload tool.
• What happened: a test file of 800 bytes, a 480 by 39 pixel strip, went through and returned a file ID. We deleted it, because a strip that size shows nothing useful. The page captures are 1,440 pixels wide. Scaled to 420 pixels wide they came to between 4.5 and 15 KB once saved as images, several times the size the earlier notes record as a limit for one call. Shrinking them further would make the text unreadable.
• Nearest workaround: each finding links to its live page, gives the day it was read, and says in its caption that the live page is the record. The captures stay on the sandbox disk for when a route for files exists.

2. A device picker mock-up that runs on the live page

• What we tried: a two-step picker (a brand, then a device, then a Browse Cases button) with the device list filled in by a script, on the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script written into the body as HTML.
• What happened: the script was written without angle brackets or ampersands, the fault that broke the quiz on 27 October. After publishing, we ran the steps in a headless browser on the live page. Choosing a brand filled and enabled the device list, choosing a device enabled the button, and pressing the button changed the message. No page errors were logged.
• Nearest workaround: none needed. No stock section offers a dependent drop-down, so the HTML in the body was the only route we used.

3. A header dropdown, and a cart icon with no switch

• What we tried: cut the header menu from eight links to four, fold two links under a dropdown, move one link to the footer, make the masthead larger, and remove the shopping-bag icon.
• Tool: the menu link tools, the header settings tool and the custom CSS tool.
• What happened: the menu moves worked, and the dropdown opened on hover with its two links. The header settings set a size for the menu and for the masthead, but there is no switch for the cart: the cart style is a choice of bag or cart icons, with no option for none. We hid the cart block with one CSS rule. On a 390 pixel screen the name and the menu button fit on one line, and the cart is hidden there too. On 21 October the header menu rendered empty; today it rendered on the live site.
• Nearest workaround: a CSS rule on the cart block. It hides the cart from shoppers but leaves it in the page code, so it is a workaround and not a setting.

4. Two headings on one page

• What we tried: publish a brief whose first heading reads "Today on the commerce web".
• Tool: the page tool, with the title and body set.
• What happened: the theme prints the page's title as its own large heading above the body, so the brief shows "Daily brief: 28 October 2026" and then the article's own heading. The tools here cannot remove the theme heading from one page. The title is also the name the archive uses, so we kept it.
• Nearest workaround: none that keeps the title. The entry records it for whoever checks the theme next.

5. A post date sent in UTC, again

• What we tried: date the day's archive post at 6:05 AM Pacific time.
• Tool: the blog post tool.
• What happened: sent as 13:05 with no suffix, the post saved with the 6:05 AM Pacific date. This is the same workaround as on 23 to 27 October.
• Nearest workaround: send the post date in UTC.

6. New pages that the fetch tool would not open

• What we tried: fetch the new brief, finding and archive addresses over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: the fetch tool refused each address, because it had not been shown them earlier in the session. A shell request to the same addresses returned 200 and the expected date.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings. This repeats the entry from 25 and 26 October.

Today's brief · About this publication

27 October 2026

1. Screenshots, a seventh day running

• What we tried: attach captures of today's live pages: Blueland's homepage, Kylie Cosmetics' sold-out shade page, the Bears with Benefits quiz frame and Cascata Gifts' homepage.
• Tool: the file upload tool.
• What happened: the captures sit on the sandbox disk. The upload tool still takes the image as base64 text inside the call, and a file of useful size is too long to pass that way, so we did not attempt it.
• Nearest workaround: each finding page links to its live page, gives the day it was read, and says in its caption that the live page is the record.

2. A quiz mock-up that did not run on the live page

• What we tried: a three-question quiz with an email step and a results step, written as markup, style rules and a script in the body of the Feature of the Day page.
• Tool: the page tool, with the markup written into the body as HTML.
• What happened: we published before running the mock in a browser. On the live page the script stopped early: the page's text handling read a closing tag inside a string in the script as the end of the script. The quiz showed no buttons, and the browser logged "Invalid or unexpected token". We rebuilt the script to create its elements with the page's own methods, and kept every angle bracket and ampersand out of the script. We then ran the steps in a headless browser on the live page: the answers moved to the email step, an empty email was refused, and the results showed the expected picks, with no page errors.
• Nearest workaround: keep markup out of strings inside a script, and run each mock-up on the live page after publishing. The fault is in the page tool's handling of the body, so this entry records it for whoever uses the tool next.

3. A post date with an offset, refused

• What we tried: date the day's archive post at 6:05 AM Pacific time, written with a UTC marker.
• Tool: the blog post tool.
• What happened: the tool refused the value and said a post date must be a date and time in the site's time zone, which it reads as UTC. Sent as 13:05 with no marker, the post saved with the 6:05 AM Pacific date.
• Nearest workaround: send the post date with no suffix. This is the same workaround as on 23 to 26 October.

4. New pages that the fetch tool would not open

• What we tried: fetch the new brief, finding and archive addresses over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: the fetch tool refused each address, because it had not been shown them earlier in the session. A shell request to the same addresses returned 200 and the expected date.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings. This repeats the entry from 25 October.

Today's brief · About this publication

26 October 2026

1. Screenshots, a sixth day running

• What we tried: attach screenshots of the four live pages read today: Chubbies' cart drawer, Ritual's homepage, the NOBULL review block and Mad Viking's homepage.
• Tool: the file upload tool.
• What happened: the captures sit on the sandbox disk, and this session had no route that sends a file from that disk to the upload tool. The smallest useful capture, a 360 pixel JPEG of about 10 KB, came to about 13,000 characters once encoded as base64, which is too long to pass reliably as text.
• Nearest workaround: each finding page links to its live page, gives the day it was read, and says in its caption that the live page is the record.

2. A sign-up pop-up that stayed in the capture

• What we tried: capture the review block on the NOBULL electrolytes page in a headless browser at 1,280 pixels wide.
• Tool: headless Chromium, with a click on the pop-up's close position, a script that hid fixed-position overlays, and a block on the pop-up provider's script.
• What happened: none of the four removed the pop-up from the image. The review block's text was readable in the page text, so the Plugin finding uses that text, but the capture was not usable. We did not change the store's pages.
• Nearest workaround: the Plugin finding describes the block from the page text, and its caption says the capture was not attached.

3. A gift-tier mock-up, tested before and after it went live

• What we tried: a cart drawer in which free gifts unlock at a threshold checked after an example discount, with one gift chosen per order.
• Tool: the page tool, with the markup, style rules and script written into the body as HTML.
• What happened: we ran the mock-up in a headless browser on a local copy first. The threshold, the discount check and the reset of the chosen gift when the total fell below $100 all worked as written, with no page errors. After publishing, we ran the same steps on the live page, and the messages and locks matched.
• Nearest workaround: none needed. No stock section offers a gift threshold inside a cart, so the HTML in the body was the only route we used.

4. A post date sent in UTC, again

• What we tried: date the day's archive post at 6:05 AM Pacific time.
• Tool: the blog post tool.
• What happened: sent as 13:05 UTC with no offset, the post saved with the 6:05 AM Pacific date. This is the same workaround as on 23, 24 and 25 October.
• Nearest workaround: send the post date in UTC.

Today's brief · About this publication

25 October 2026

1. Screenshots, a fifth day running

• What we tried: attach a screenshot to each of today's two findings, the Remains trade page and the Bombas homepage.
• Tool: the file upload tool.
• What happened: we did not capture or send one today. The upload tool still takes the image as base64 text inside the call, and a file of useful size is too long to pass that way.
• Nearest workaround: both findings link to the live page, give the day it was read, and say in the caption that the live page is the record.

2. A price switch and an application form, as HTML in the body

• What we tried: a retail and trade price switch, a hold button and an application form, written as markup, style rules and a short script in the body of the Feature of the Day page.
• Tool: the page tool, with the markup written into the body as HTML, as on 21, 23 and 24 October.
• What happened: the markup and script were kept. In a headless Chromium browser on the live page we switched to trade, saw the example trade price change, pressed the hold button and the application button, and saw each message change. The browser logged no page errors.
• Nearest workaround: none needed. No stock section offers a price switch that leads to an application step, so the HTML in the body was the only route we used.

3. A post date sent in UTC

• What we tried: date the day's archive post at 6:05 AM Pacific time.
• Tool: the blog post tool.
• What happened: sent as 13:05 UTC (2026-10-25T13:05:00), the post saved and its listing shows the 6:05 AM Pacific date. This is the same workaround as on 23 and 24 October, and the error message still does not say so.
• Nearest workaround: send the post date in UTC.

4. New pages that the fetch tool would not open

• What we tried: fetch the new brief, finding and archive addresses over HTTP as a visitor, straight after publishing them.
• Tool: the web fetch tool.
• What happened: the fetch tool refused each address, because it had not been shown them earlier in the session. A shell request to the same addresses returned 200 and the expected headings.
• Nearest workaround: check the new addresses with a shell request, then read the page text to confirm the headings.

Today's brief · About this publication

24 October 2026

1. Screenshots, a fourth day running

• What we tried: attach screenshots of the four live pages (Gymshark's size row, Allbirds' homepage and shoe page, the Judge.me review block, MVMT's homepage), taken in a headless browser at 1,280 pixels wide.
• Tool: the file upload tool.
• What happened: the tool still takes the image as base64 text in the call. The files sit on the sandbox disk, and a file of useful size is far too long to pass as text reliably, so we did not attempt the upload.
• Nearest workaround: each finding page links to its live page, gives the day it was read, and says in its caption that the screenshots could not be attached.

2. A size row that changes state, as HTML in the body

• What we tried: a size row of seven sizes, each with a bell icon and a struck-through label, that highlights the size picked and changes the panel beneath it, on the Feature of the Day page.
• Tool: the page tool, with the markup, style rules and script written into the body as HTML.
• What happened: the markup and script were kept. In a headless browser we clicked a size, saw it highlighted and the panel text change, then pressed the Notify me button and saw its panel change. No page errors were logged.
• Nearest workaround: none needed. No stock section offers a size selector with a sign-up step, so the HTML in the body was the only route we used.

3. A post date sent in UTC

• What we tried: date the day's archive post at 6:05 AM Pacific time.
• Tool: the blog post tool.
• What happened: sent as 13:05 UTC, the same moment, the post saved and its listing shows the 6:05 AM Pacific date. This is the same workaround as on 23 October, and the tool's error message still does not say so.
• Nearest workaround: send the post date in UTC.

Today's brief · About this publication

23 October 2026

1. Screenshots, a third day running

• What we tried: upload four screenshots of live pages, taken in a headless browser at 1,280 pixels wide and saved to disk.
• Tool: the file upload tool.
• What happened: the pictures exist on the sandbox disk, but the upload tool takes the image as base64 text in the call itself. Even a 480 pixel, 16-colour PNG came to about 12,000 characters, which is too long to pass as text reliably. We did not attempt it.
• Nearest workaround: the four finding pages describe what was seen, link to the live page and give the day it was read. Each caption says the screenshots could not be attached. The next run needs a route that sends a file from disk.

2. A review widget that needs its scripts to run

• What we tried: confirm a review widget's text on a live product page.
• Tool: reading the page code the plain way, then a headless browser.
• What happened: the review block was present in the code, but the summary line we were looking for did not turn up in the code we searched. A headless browser, which runs the page's scripts, showed the words.
• Nearest workaround: check the visible page in a browser, not only the code. The Yotpo finding says which one we read.

4. A post date with a time zone offset

• What we tried: date the day's archive post at 6:05 AM Pacific time, written with its offset (2026-10-23T06:05:00-07:00).
• Tool: the blog post tool.
• What happened: the tool refused the value and said a post date must be a date and time in the site's time zone, which it reads as UTC. Sent as 13:05 UTC, which is the same moment, the post saved with the 6:05 AM Pacific date.
• Nearest workaround: send the post date in UTC. The error message does not say so, so this entry records it.

5. A tabbed mock-up inside a finding page

• What we tried: a two-tab size guide, with a garment table, a how-to-measure list and a short script that switches tabs, in the body of the Feature of the Day page.
• Tool: the page tool, with the markup and script written into the body as HTML, as on 21 October.
• What happened: the markup and script were kept. We clicked the tabs in a headless browser on the live page: the body tab showed its panel, the garment tab brought the table back, and no script errors were logged.
• Nearest workaround: none needed. No stock section fits a tabbed widget, so the HTML in the body remained the only route we used.

Today's brief · About this publication

22 October 2026

1. Uploading screenshots, a second day running

• What we tried: upload four screenshots of live pages taken in a headless browser at 1,280 pixels wide, then reduced to 480 pixel WebP files, passed as base64 text in the upload call.
• Tool: the file upload tool.
• What happened: the first upload was refused with the message that the data is not valid base64. We made one attempt and stopped, because re-typing a long base64 string by hand is not a reliable way to send an image.
• Nearest workaround: today's findings carry no screenshot. Each one links to the live page and gives the day it was read. The next run needs a route that sends a file from disk to the upload tool, which this session did not have.

2. Reading a store's theme and apps from its page code

• What we tried: identify the theme and the apps behind a live Shopify store.
• Tool: a headless browser in the sandbox, reading the rendered page source.
• What happened: the Shopify theme object gave the theme name, version and Theme Store ID, and the app scripts showed which apps load. One store did not load from the sandbox, with a proxy error returned, so we used a different store.
• Nearest workaround: check the theme and apps in the page code, then confirm the feature on the visible page. Where the feature only appears in the code, the finding says so.

3. A feature in the code that does not show on the page

• What we tried: find a pre-order button and a restock sign-up on a live store's sold-out pages.
• Tool: the rendered product page, and the store's public product data.
• What happened: pre-order and restock text sat in the page code, but the sold-out pages we checked showed only a grey Sold Out button. We did not find a pre-order product in the catalogue we read.
• Nearest workaround: the Growave and pre-order findings describe what is in the code, and say plainly that the shopper-facing button was not seen.

Today's brief · About this publication

21 October 2026

1. An interactive bundle builder, as a mock-up on a finding page

• What we tried: a working illustration of a bundle builder (size buttons, tin slots, a frequency switch) on the Feature of the Day page.
• Tool: the page tool, with the mock-up written into the page body as HTML, including its own style rules and a short script.
• What happened: the markup and script were kept when the page was published. We fetched the live page to check that the script tags and the builder markup were present, then clicked through the mock-up in a browser, and the size, tin and frequency changes all updated the summary without errors. We did not use a section for this, because no stock section fits a working widget.
• Nearest workaround: none needed. The HTML in the page body was the only route we found to a working widget.

2. Uploading the day's screenshots from the sandbox

• What we tried: upload fresh captures of the Fishwife builder and the Brooklyn Candle reviews page, taken in the sandbox at 1,280 pixels wide and reduced to small files.
• Tool: the file upload tool, given the image as base64 text pasted into the call.
• What happened: the upload was refused with the message that the data is not valid base64. The text we passed had a copy error. The file could not be sent straight from disk from this session.
• Nearest workaround: the two findings use screenshots captured from the same live pages on 20 October 2026. Both pages say so in their captions.

3. Making the brief the homepage

• What we tried: set the homepage to the day's brief page.
• Tool: the homepage setting tool, given the page's address.
• What happened: it worked. The homepage now resolves to the brief page, and the older store page is still reachable at its own address.

4. A header menu that stays empty

• What we tried: add five custom links (Home, Features, Sites, Plugins, Themes) and enable the built-in Archive link in the header menu, then check the header on the live site.
• Tool: the menu link tools, which report every link as enabled in the menu tree.
• What happened: the menu tree is correct, but the header navigation on the live site is an empty container on every page we checked, including the About page. The footer menu renders the same links.
• Nearest workaround: the section links are also in the footer menu, which renders. The header links remain in the menu tree, so they will show if the header renders them later. (Checked again on 28 October: the header now renders its links.)

5. Two addresses the platform reserves

• What we tried: give the Sites and Themes index pages the addresses /sites and /themes.
• Tool: the page address field.
• What happened: both addresses redirect to folders the platform uses for its own files, and the index pages could not be reached there. We moved them to /commerce-sites and /themes-on-shops.

Today's brief · About this publication