11:50 — Phase 409: /split takes a fixed buy, a floor of one, and case packs

11:50 — Phase 409: /split takes a fixed buy, a floor of one, and case packs

Why: the free split page is where every Sizecurve post sends people (?via=community, ?via=devto). The 10-12 article and Reply 6 (drafted 10-14) both teach the largest-remainder method and then say the page does not do it. A merchant who follows the link to round a real buy finds the page cannot. Merchant installs: 0; this is the page that has to earn the next step.

What: one Opus 5.5 agent changes app/sizecurve/public/split.js, split.html and test/split.test.mjs only. The existing order and its engine-equality test do not change. A new optional section, "My buy is fixed": a total in units or packs, a pack size (default 1), and "at least one of every size". It splits the fixed buy so that after it lands each size is as near its curve share as whole units allow, by largest remainder, and shows it beside the computed order. Tests carry the article's figures (60, 61, 71, 6 with and without the floor, 72 in packs of 6). npm test and npm run browser must pass. No deploy and no commit by the agent; I check the diff, commit, and ask the boss "may I run the release".

Done when the diff is checked, the tests pass, and the release question is asked.

---

Phase 410: the caption bench for Close of Day's screencast (planned 2026-10-14)

The boss asked for the listing screencast (4.5.3). They record it; I caption and host it. The bench for that is gone: ffmpeg was in ~/.local/bin, and the home folder is reset when the container is rebuilt (the third time: 09-18, 09-23, now). So:

1. One Opus agent downloads the same static ffmpeg/ffprobe build used 09-18 into cider2/.tools/bin/, which survives a rebuild, and .tools/ goes in .gitignore.
2. It builds a synthetic 4:30 take of nine labelled steps (no store data), runs sheet.py, retime.py and burn-subtitles.sh on it with a nine-cue draft caption file taken from app/daybook/listing/SCREENCAST.md, and reports what failed.
3. SCREENCAST.md for Close of Day gets the order route the boss can do today: orders made in the admin with Mark as paid are not test orders, so the page and the Finances summary have figures; film Today instead of Yesterday.

Done when a captioned synthetic take plays, ffmpeg survives in .tools/, and the script names the order route. Not done: hosting (it waits for the real take).

---

Phase 411: four X posts for the fixed buy (planned 2026-10-14, 21:00)

Sizecurve has 42 engaged views and needs 50 by 10-20. Among referred visits, X has sent the most (24); its queue has been empty since x8 went out on 10-09. Four text posts, x11 to x14, go out 10-15 to 10-18 at 15:00 UTC, each linking /split?via=x. They cover the largest remainder, the floor of one, case packs and on-hand stock. They quote no figure of two digits or more, because split.html carries none of the worked example's figures and the gate would refuse them. One Opus agent adds them to X.md, X_DUE and PAGES and runs the dry runs and the gates. Then I check the diff, commit, and queue them with schedule.py (Buffer, mine).

Done when Buffer holds the four. Read: /split engaged views and the x source on 10-19.

00:40 — A measurement, no phase: what search sees of Sizecurve

Read with my own Search Console tool (.search/, FACTS 10-02), which I had stopped using; my open ask to the boss for the sitemap was out of date, and is withdrawn.

• Google, 09-01 to 10-14, fresh days included: no rows. Not one impression.
• Sitemap: 13 URLs, downloaded by Google 10-12, 0 errors.
• Index, by URL Inspection (13 of 200 a day): - Indexed (5): / (crawled 10-11), /split, /dataset, /privacy, /terms. - Crawled, not indexed (1): /how-many-of-each-size. - Unknown to Google (7): /check, /broken-size-runs, /which-size-goes-first, /one-star, /returns-arithmetic, /demo, /canary. These are the pages built to be found, and they are unknown four weeks after the sitemap listed them.
• Bing (via DuckDuckGo's results): 3 pages, /, /check, /demo. Submitted the other 9 public pages through bing-submit (9 of 20 a day). /canary left out.

What it means for 10-20: search adds nothing before the date. For later: the indexed pages are the legal ones and the one tool; the articles are not, so a page's being in the sitemap is not enough. What Google needs is a link from a page it already trusts.

---

01:15 — Why seven pages are unknown: not the site

Checked read-only. / (indexed, crawled 10-11) links to all seven unknown pages; /split and /dataset link to most of them too. The unknown pages answer 200, carry their own canonical, and set no noindex header or meta. robots.txt allows everything but /api/. So Google has had a path to them since 10-11 and has not taken it. That is crawl priority for a new subdomain with no outside links it counts, not a fault I can fix in the pages. No change made.

---

01:45 — dev.to links do not get pages found, either

Checked before editing anything: dev.to's article links to our site are followed (rel="noopener noreferrer", no nofollow). But my articles already link /check and /broken-size-runs (4777995, published in September) and /which-size-goes-first (4784351), and all three are still unknown to Google. Adding links to the other unknown pages would do no better. No article edited.

---

10:20 — The thread 690145 follow-up, saved where the boss can copy it

The follow-up drafted for the boss on 10-13 was only in my reply, never in a file. It is now Reply 6 in app/sizecurve/marketing/SHOPIFY-COMMUNITY.md, rewritten (1,633 characters). It covers whole units from a size curve: the largest remainder, a floor of one, and case packs. Every figure comes from the 10-12 dev.to article, and it says what the free page does not do. It ends with a question to merchants. The thread was read once this hour: 8 posts, the last on 10-03.

---

11:30 — The one install is ours; reported as merchant installs from now on

The boss asked. /api/installs lists one shop, installed 09-14, before approval on 09-29: the dev store. Its log has one uninstall, on 10-01: the App Review store, after 51 hours, never opened, no plan. FACTS 09-30 (the Partner Dashboard installs list) agrees. Merchant installs since listing: 0. Every report from now on says "merchant installs: 0", not "installed: 1".

---

12:20 — Phase 409: the split page takes a fixed buy

One Opus 5.5 agent changed public/split.js, split.html and the two split tests. I ran tools/sitemap.mjs for the new /split lastmod.

• New: fixedBuy splits a fixed buy (in units or packs) by largest remainder, toward each size's share after what is on hand, with an optional floor of one. The page has a "My buy is fixed" fieldset and a second table with its own copy button.
• Checked: the article's figures come out exactly: 60, 61, 71, 6 with and without the floor, and 12 packs of 6. With the buy blank, the page's output is byte for byte what it was. The floor cannot run out of a size to take from: if every size gets one, some size holds two.
• npm test 667 pass, 0 fail; npm run browser 64 pass, 0 fail.
• Reply 6 now says the page does it, and is marked to post only after the release. After the release I correct the last paragraph of dev.to 4841120, which says the page cannot.

Not released: release needs the boss's yes.

---

Phase 410: the caption bench for Close of Day's screencast (actual, 2026-10-14)

One Opus agent; I checked its report, the files and a still.

• ffmpeg 7.0.2 static is now in cider2/.tools/bin/, which survives a rebuild. .tools/ is git-ignored, and git status shows nothing from it. ~/.local/bin has been wiped three times.
• app/daybook/listing/retime.py: Sizecurve's tool set to nine beats (PLANNED = 0, 10, 40, 65, 100, 135, 160, 220, 255, 270). Nothing else changed.
• SUBTITLES-draft.srt: the script's nine captions as 13 cues, with no line over 60 characters. Beats 4, 7, 8 and 9 are split in two at a clause break.
• Dry run on a synthetic 4:40 take with deliberately off-plan starts: sheet.py, then retime.py, then burn-subtitles.sh, all exit 0. - Every mid-step still shows that beat's caption, readable, on a dark band inside the frame. I looked at step 7 myself. - Starts read off the sheets come up to 2 s late, because the sheets show one frame every 2 s. - This build has no drawtext, so the agent drew the synthetic labels with PIL. The real pipeline doesn't need drawtext.
• SCREENCAST.md now gives the order route the boss can do today: create the orders in the admin with Mark as paid, which makes them non-test orders, and film Today. The reseed (ASKS 6) is no longer needed for the take.
• Still open: - Beat 3's draft caption says "Yesterday". The final captions are written from the take's frames anyway. - Hosting waits for the real take. - The take itself waits on the boss and on the $9 plan being confirmed.

---

Between phases, 20:00 to 20:20: one sitting for the boss (2026-10-14)

• Push works again. FACTS records it.
• Sizecurve engaged views read 42, the same as this morning: / 31, /split 5, /check 3, /demo 2, /broken-size-runs 1. It needs 8 more by 10-20. The consequence written for 10-20 stands.
• app/daybook/listing/SUBMIT.md: every remaining Close of Day item is the boss's (the 15 marked boss in REQUIREMENTS.md). It lists them in the order they can be done: dashboards, dev store, screencast, Submit. The plan comes first, because six checks wait on it. - It includes the protected-data reason as text, since Close of Day has no PROTECTED-DATA.md. - The fields filled 10-07/08 are marked as done.

---

Phase 409 released (2026-10-14, about 20:30 Chicago)

The boss said yes to "may I run the release".

• First run stopped at the browser tests, before deploy: Playwright's headless shell was gone from ~/.cache after the home reset. I reinstalled it with npx playwright install chromium-headless-shell.
• Second run: npm test 667/667 and npm run browser 64/64 passed, then deploy and frontdoor ran.
• Live: /split serves the "My buy is fixed" fieldset (#buy, #pack, #floor) and the "counted out" paragraph. /split.js exports fixedBuy.
• dev.to 4841120: the closing paragraph now describes the fixed buy instead of "It takes no fixed total, floor or case pack". /api/articles/4841120 confirms the new text is there and the old text is gone. marketing/dev-rounding.md matches.
• Reply 6 in SHOPIFY-COMMUNITY.md is marked postable. Posting it in 690145 is the boss's.
• Learned again: the home reset takes Playwright's browser as well as ffmpeg. The release needs the browser, so before any release run ls ~/.cache/ms-playwright/chromium_headless_shell-* and reinstall if it is missing.

---

Phase 411: four X posts for the fixed buy (actual, 2026-10-14, 21:15)

• The agent's changes: blocks 11 to 14 added to X.md as written, and x11 to x14 added to X_DUE and PAGES (split.html). - All four dry runs passed the figure and reach gates. Lengths with the link counted as 23: 277, 253, 250, 263. - test_figures.py: drafts plus ten more modes ran, all 0 failed. queue was not run, since it calls Buffer. npm test 667/667. - It broke the brief once: it wrote ten tf-*.txt files into the scratchpad. They hold no repo content, and none of them is in the repo.
• Queued by me with schedule.py: Buffer answered "scheduled" for each.

Post	Goes out (UTC)
x11	10-15 15:00
x12	10-16 15:00
x13	10-17 15:00
x14	10-18 15:00

• Read on 10-19: the x source (24 now) and engaged views (42 now, 50 needed by 10-20).