What happened here

Nothing written since Tuesday

The clubs still run every afternoon, 2–6 PM. See the week.

⛔⛔ 1. This page was outside the design system for an hour

Its component lived in a <style> block inside the page — so the responsive rules, the measure caps, the kitnote cap and the li exception list, all of which live in _kit.css, reached every kit except this one.

⭐⭐ And the part nobody spotted: assertion 13 could never have caught anything here, because the check knows classes and this page was built from classes the system had never seen. A page outside the system passes the system's checks by being invisible to them. ⛔ A clean measurement would have been meaningless and would have read as a pass — the "gate that cannot fail" pattern arriving through a door nobody had shut. Moved into _kit.css.

⛔ 2. Every entry has its own URL, and it is nested

Owner ruled: /activities/{slug}/, not root-level. ⭐ The second URL shape this project has chosen rather than inherited. At ~30 entries a month, deciding this a year from now means redirecting ~360 URLs; deciding it today costs a routes.yaml collection. ⚠️ Which is the Ghost Settings wall — a human in admin, not a script — and joins the paste-in package already pending.

⛔ The individual entry page does not exist as a kit and is now the gap. ⭐ Shareability is its primary requirement, not a property it happens to have: direct traffic is 60.8% — 5,449 visitors — against 20.7% search. This is already a shared-link site whose pages contain nothing shareable. A heading inside a 12,628px guide cannot be sent to anybody. So the entry page is designed for the moment it arrives in a WhatsApp thread, which makes og:image, og:title and the description product decisions, not metadata hygiene.

⚠️ The floor is adequate and should be stated as such: every post already has og_image, og_title, og_description, meta_* and canonical_url fields — all currently empty, so Ghost falls back to the title and feature image. ⛔ That fallback is fine. Nobody is blocked on hand-writing 360 descriptions; a hand-written one earns its keep only where the headline alone would mislead.

⚠️ 3. The photographs now carry width and height — a defect I had already solved

All three lacked both attributes, so the feed reflowed as each image arrived — content jumping under a reader's thumb, on the page whose whole purpose is scrolling. ⛔ I got this right on /team/ one page earlier — “width/height on every one so the grid does not reflow as thirteen images arrive” — and did not carry it here. ⚠️ These are unresized WhatsApp originals, the worst case for it.

⚠️ 4. What the import must guarantee — not mine to build

The monthly guides stay, and the duplicate-content worry does not apply: ⭐ a past record cannot drift. “On 7 August we finished fourteen bags” is true forever. The one real cost is Google splitting signals, solved by setting each entry's canonical_url to the guide it came from.

⛔⛔ CORRECTED 10 Aug — March and April DO have day headings, in a different tag. This note previously said they had none. They use h3 where May onward uses h2, and a parser keyed on h2 finds zero days in them and exits clean. ⭐ April holds 30 days, 41 activities and 30 photographed ones — the second-best-photographed month in the corpus — and it is precisely the month a working script would silently drop. ⚠️ “No day headings” and “day headings I did not look for” produce the identical output; only reading the markup separates them. Accept both tags, and assert a per-month day count.

⛔ Two more will bite the splitter: Matching each photograph to the right day is the fuzzy-match trap in a new place, and unlike a redirect map a wrong answer here is invisible. And ⛔ the newsletter: 796 subscribers. Seven posts on this site already have status sent — prove, do not assume, that API-created posts do not send. A hundred creations with the wrong flag is a hundred emails.

⚠️ Newsletter is publish-only by an existing team habit, not by a setting — Ghost's default is the other way, so the next person to join needs to know it is a habit that can be broken by accident.

⛔⛔ 4b. THE MOCKUP COMMITTED THE EXACT ERROR NOTE 4 WARNS ABOUT

Note 4 says “matching each photograph to the right day is the fuzzy-match trap in a new place, and unlike a redirect map a wrong answer here is invisible.” ⛔⛔ Three cards on this page and four on the homepage carried the WRONG photograph — the images were shifted one place against their headlines. A dance rehearsal sat under “Chapter leaders met for a quarterly district meeting”, and children sitting at a science session sat under “The community dance class practised together again”.

⭐ It was found by looking at the rendered page, not by any check. Every photograph loaded, every card measured correctly, every date was right. ⚠️ A wrong picture is a well-formed page.

⛔⛔ RETRACTED, and the retraction is the more useful half. This note said “4 July's five real activities are…none of them is stitching…that entry did not happen that day.” It did. 4 July has SIX activities and one of them is the Stitching Lab, 2–6 PM, Mary Gracy — “The stitching team continued working on new products while improving quality and exploring better production techniques.”

⭐⭐ The reason I could not see it is the whole lesson: the stitching paragraph is the only one of the six with NO photograph after it. I enumerated the day from a photo-filtered extraction and then stated a fact about the day. ⛔ I read “absent from my filtered view” as “did not occur” — and the filter was the very thing under test.

⛔ This is my own rule from the homepage notes, broken by me: “out of view, not out of existence.” ⚠️ And it matters at scale, not just here: 141 of the corpus's 250 activities have no photograph. 56%. Any rule that treats an unphotographed entry as doubtful doubts the majority of the record. ⭐ The replacement card stands on its own merits — Monday 6 July, a real 6 July photograph — so no page is wrong; only this note was.

⛔⛔ 4c. AND THE SOURCE ITSELF MISPAIRS. Verified by rendering all twelve photographs on one page and reading the headline under each. The 6 July stitching entry's picture — placed by the guide, correct by DOM order, correct by day — is a person dragging a cut branch of she-oak foliage along a paved lane. ⛔ Not fabric. Not stitching. Not indoors. Card removed rather than given an invented headline.

⭐⭐ So “the photograph follows the paragraph” is a rule about the DOCUMENT, not about reality. ⛔⛔ A pairing audit built from the guides can only prove conformance to the guides — a source-level mispairing is invisible to it and audits as correct. The only instrument that finds these is a person looking at the picture beside the headline.

⚠️ Three softer cases in the same sweep, all kept: the bags-on-sale entry shows someone sewing rather than bags on a shop shelf; the Reading Corner entry shows the science session it shares a paragraph with; the fruit-seed entry shows the room, no seeds. ⭐ Each is a picture from that activity's session rather than of the thing the headline names — acceptable for a record, and worth knowing the corpus is like this. ⛔ The hard failure is the one where the picture is of a different activity entirely.

⭐ Two things the full 4 July block shows that the filtered one could not. The district meet-up entry contains a participant quotation — “The home-like feeling. Jasentha Chechi who cooked us breakfast at short notice…” — so testimonials live inside the activity guides, not only in posts, which widens the testimonial inventory. And a non-paragraph element can sit between an activity and its photograph, so the DOM-order rule must skip blockquotes rather than reset on them.

⭐⭐ For the importer this raises the bar on note 4: the pairing must come from DOM order inside the day block — each photograph belongs to the activity paragraph it follows — and the result must be spot-checked by a human against the pictures. ⛔ There is no automated check for “this is a photograph of a different thing.” ⚠️ Only the entries with no prose are ambiguous to pair — so those are the ones a human checks first. ⛔ Not a reason to leave them out.

⛔⛔ 5. Every July and August weekday in these mockups was WRONG

Against the real 2026 calendar: 4 July is a Saturday, not a Friday. 7 August is a Friday, 6 August a Thursday, 4 August a Tuesday. Corrected here and on the homepage. ⚠️ The May dates were right, so it was not a uniform slip — a wrong stretch inside a correct list looks like data.

⛔ This is not a layout defect and no layout check could see it. The whole claim of this page is this happened, on this day; naming the wrong weekday is a different statement, not a smaller one. Someone who was in the room on Saturday reads “Friday” and stops believing everything else on the page.

⭐⭐ The source guides carry no weekday at all — they say “July 4”. So every weekday on this site is computed, and it must be computed in one place, from the date, at Asia/Kolkata. ⛔ A date rendered in UTC lands on the previous evening and shifts the weekday back by one — which is this exact error, mechanised and applied to all 250 entries.

⛔ 6. The no-photo form is out of view, and must not be deleted

The owner ruled the mockups pull only photographed entries for now, so the two exemplars that were on this page are gone from it: 1 August, a rain closure — nobody photographs a shut room, so that one can never have a picture — and 2 August, a dance group meeting that simply was not photographed. ⭐ 92 of 157 days are in the same position.

⛔⛔ The CSS for that form stays in _kit.css and stays documented. The rule behind it is unchanged and is the reason the grid has no holes: no slot without a guaranteed filler — an entry with a photograph is a card, an entry without one has no container at all. The failure to guard against is somebody later reading “we never see this” as “this cannot happen”, deleting it as dead code, and rediscovering it as an empty 4:3 well on the day a rain closure ships.

⚠️ The site still has no surface for “something is coming up”, and that is deliberate — future entries were cut. ⛔ It must not creep back in here. The record says what happened, and nothing else.

⚠️ 7. The source has a typo the importer will trip on

July 11 reads Faclilitated by Olimalar Foundation. ⛔ A schedule-prefix stripper matching Facilitated by exactly will leave that string in the body of a published entry, where it reads as our error rather than the source's. ⭐ Match loosely, then assert that no published body still contains a time range or the word lab-prefixed room name — a check on the output, not on the pattern.