Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(x-content): write the volume model into the author brief and the schedule docs

nirholas2c70c8e01959
diff --git a/api/_lib/x-content/priority.js b/api/_lib/x-content/priority.js
index f40a6cd66..71c0430f1 100644
--- a/api/_lib/x-content/priority.js
+++ b/api/_lib/x-content/priority.js
@@ -63,10 +63,12 @@ export function loadVolumeModel(root) {
return { ...model, features: model.features.map((row) => ({ ...row, regex: row.pattern ? new RegExp(row.pattern.source, row.pattern.flags) : null })) };
}
-// The attributes that are not a text pattern. Every slot sits inside the hours
-// the model calls `usHours` (schedule.js), and the queue only ever posts from the
-// company account, so both are true for every queued post: they move every score
-// by the same amount and are kept so the reported chance is an honest one.
+// The attributes that are not a text pattern. Every slot in the queue's cadence
+// sits inside the hours the model calls `usHours` (12:00 to 20:00 UTC), and the
+// queue only ever posts from the company account, so both are true for every
+// queued post: they move every score by the same amount and are kept so the
+// reported chance is an honest one. A cadence with a slot outside that window
+// would make `usHours` untrue; keep the slots inside it.
const VOLUME_CHECKS = {
video: (item) => (item.posts?.[0]?.media || []).some((row) => /\.(mp4|mov)$/i.test(row.path || '')),
thread: (item) => (item.posts?.length || 0) >= 2,
diff --git a/docs/x-content-pipeline.md b/docs/x-content-pipeline.md
index 37dc42bc5..1987af5f0 100644
--- a/docs/x-content-pipeline.md
+++ b/docs/x-content-pipeline.md
@@ -34,7 +34,7 @@ pipeline, enforced before anything is sent:
| Silent autoplay | `prepare-video` burns captions into the clip, because most of the feed watches muted. |
| Hype copy | [api/_lib/x-content/quality.js](../api/_lib/x-content/quality.js) rejects launch openers ("Introducing", "We're excited"), hype vocabulary, hashtags, emoji, stacked exclamation marks, all-caps shouting, and en or em dashes. |
| Repeats | A head that reads too much like another queued item, or like anything @trythreews already posted (the scraped archive in `data/archives/` plus everything this pipeline published), is rejected. |
-| Clockwork timing | Three slots a day, eight hours apart, every day, each opening at a jittered minute, with a minimum gap and a daily cap. |
+| Clockwork timing | Three slots a day inside US hours, every day, each opening at a jittered minute, with a minimum gap and a daily cap. |
| A schedule anyone can read off the repository | The slot jitter is an HMAC under `X_CONTENT_SCHEDULE_SEED` (production only), and which post fills a slot is decided at the moment it opens. Unset, the jitter falls back to a public hash, and `plan` says its minutes are placeholders. |
| A link that cannot be checked | Every link is resolved before a post can be approved, and again seconds before it is sent. npm answers 403 to any non-browser client, so a package page is resolved against `registry.npmjs.org`, which is the authoritative record of whether that package exists and answers any client. |
| A broken post stalling the feed | A post that cannot go out (a link is down, a feature probe fails, X rejects the content) is held with the reason and a backoff, and the slot goes to the next-best post in the same run. |
@@ -131,7 +131,7 @@ Cadence lives at the top of the file: `slots` (each `{ tier, at }` in UTC), `win
## How the next post is chosen
-**When:** three slots a day, one per tier, eight hours apart, every day of the week: T3 at 04:00 UTC, T2 at 12:00, T1 at 20:00. That cadence is the owner's (2026-09-20): three a day, weekends included. It replaced a tighter window drawn from the volume study, which measured every original post against 1-minute candles of the $THREE pool and found that a post between 12:00 and 20:00 UTC was followed by a volume response about 1.7 times as often as one outside it. Two of the three slots still sit in that window; the third takes the off-peak turn so the day stays evenly spaced. Each slot opens at a minute only the production seed can reproduce (up to `windowMinutes` late), and stays open for three hours (`slotOpenMinutes`), so a missed run or a deploy still posts, but no slot is ever spent twice. Slots are at least `minimumMinutesApart` apart even at their latest opening, so an earlier post never blocks a later slot. There are no quiet hours: with fixed slots eight hours apart, the slot times already say when the account posts. `flagshipWeekdaysOnly` would withhold the T1 slot on Saturday and Sunday and keep T1 posts out of the lower slots too; the queue no longer sets it, and it is the one lever that would return the account to weekdays-only flagships.
+**When:** three slots a day, one per tier, every day of the week, all inside 12:00 to 20:00 UTC: T2 at 12:00, T1 at 15:30, T3 at 19:00 (8:00, 11:30 and 15:00 in New York). The volume study measured every original post against the $THREE pool's candles and found a post in that window was followed by a volume response about 1.7 times as often as one outside it, one of the two strongest effects in the model. The owner's 2026-09-20 cadence spread the three slots eight hours apart (04:00, 12:00, 20:00), which put two of them outside the window; on 2026-10-01 the owner made pool volume the goal, so all three moved inside it, three and a half hours apart. Each slot opens at a minute only the production seed can reproduce (up to `windowMinutes` late), and stays open for three hours (`slotOpenMinutes`), so a missed run or a deploy still posts, but no slot is ever spent twice. Slots are at least `minimumMinutesApart` apart even at their latest opening, so an earlier post never blocks a later slot. There are no quiet hours: the fixed slots already say when the account posts. `flagshipWeekdaysOnly` would withhold the T1 slot on Saturday and Sunday and keep T1 posts out of the lower slots too; the queue no longer sets it, and it is the one lever that would return the account to weekdays-only flagships.
**Three a day is a quota** (owner, 2026-09-30), so a slot is never left empty while an approved post
could fill it. When the usual order finds nothing, the slot takes, in turn:
diff --git a/docs/x-story-author-brief.md b/docs/x-story-author-brief.md
index a3c72035b..f9287dfb2 100644
--- a/docs/x-story-author-brief.md
+++ b/docs/x-story-author-brief.md
@@ -88,6 +88,29 @@ production and has been filmed, so that `public/x-media/<id>/reel.mp4` and
probe when the feature has a public GET (and one asserting any count the copy states), and
`"covers"` listing the backlog keys the story speaks for.
+## Writing for volume
+
+The queue ranks every post by its chance of being followed by a volume response on the $THREE pool
+(the hour after the post trading at least twice the hour before). The chance comes from
+[`data/x-content/volume-model.json`](../data/x-content/volume-model.json), a model fitted on 313
+original posts and the pool's 15-minute candles. What it found, strongest first:
+
+| Attribute | Odds ratio | What it means for the copy |
+|---|---|---|
+| Announces something shipped | 1.96 | Lead with the launch, in plain words: "is live", "shipped", "now". The voice rules still ban "Introducing" and hype; "is live on three.ws" is neither. |
+| Posted 12:00 to 20:00 UTC | 1.73 | The slots handle this; every slot sits in that window. |
+| Recognition language | 1.63 | When the feature runs on, or is recognised by, a partner programme we are really in, say so: partner, listed, verified, featured, joined. |
+| Names a tier-1 company by its @handle | 1.49 | Tag the company the feature genuinely runs on (@nvidia, @GoogleCloud, @OpenAI, @IBM, @awscloud ...), at most two, each with its reason in `mentions`. A tagged post waits for the owner; that is fine. |
+| Longer than 180 characters | 1.47 | The head carries the news and the mechanism. |
+| A thread | 1.34 | One or two replies with the detail. |
+| Talks about the token | 0.97 | No effect either way. Do not lead with $THREE to move the pool. |
+
+So the strongest story is a feature that shipped, running on a partner we are really in, told as
+news. Our real programmes, as `/partners` states them: OpenAI Select Partner, IBM Business Partner,
+AWS Partner (Software Path; the Marketplace listing is "coming", not live), Google Cloud (production
+runs on it), NVIDIA Inception, Alibaba Cloud. Quote the status exactly as the page words it; never
+upgrade it.
+
## The report
For each finished story: id, the one-sentence story, tier, lane, pattern, reel length and