Splash catches up: augment-it, a Studies landing page, nested-ps…

Why Care?

The splash is supposed to be the truthful surface for what's happening across ai-labs. The minute it stops reflecting reality — a new child arrives, an app inside an existing child starts shipping its own ship-notes, a study gets pinned — readers lose the ability to trust it as a current read. This round closes four such gaps at once: a new child (augment-it) is now first-class on every count and card; the rollup now recurses into pseudomonorepo children's own apps/* so memopop-ai's three internal apps (memopop-native, memopop-orchestrator, memopop-site) contribute their changelog and context-v files; the three studies finally have a dedicated landing page that reads itself from the pinned-repo metadata rather than hand-maintained card copy; and OpenPanel analytics is wired the same way the other splash sites do it so we'll actually know whether anyone reads any of this.

What's New?

`augment-it` is now visible everywhere

Three small edits make the new child first-class:

childSlugs in splash/src/pages/index.astro gains 'augment-it' — the hero meta-strip now reads 4 active.
CHILDREN in splash/scripts/rollup-sync.ts gains the matching entry — augment-it's changelog/ and context-v/ now roll up alongside the others (12 changelog entries on first sync — augment-it has been shipping).
A new splash/src/content/product-highlights/04-augment-it.md adds the bench-card in ch.02 · active builds. The lede notes it as the outlier in the labs — React + Next.js + Module Federation + Turbo, where the rest is Astro/Svelte — "because Module Federation pays off where the architecture warrants it."

Rollup learns about pseudomonorepo children

memopop-ai is itself a pseudomonorepo with apps/memopop-native, apps/memopop-orchestrator, and apps/memopop-site as nested submodules, each carrying its own changelog/ and context-v/. The previous rollup only walked each child's root, so memopop-ai showed 2 changelog entries when there were really ~80 across its apps.

rollup-sync.ts now auto-discovers sub-children:

A new Source type decouples where files live on disk from the provenance slug stamped on each rolled-up file.
discoverSubChildren(child) walks <child>/apps/, returns every directory inside it that has at least one of changelog/ or context-v/, with slug <child>/<app> and dir <child>/apps/<app>.
resolveSources() flattens the top-level CHILDREN into the full source list — each top-level child immediately followed by its discovered sub-children, preserving presentation order.
syncCollection was changed to take a sources: Source[] argument instead of reading the constant directly — same loop body, broader iteration.

Result on the first run:

changelog : 107 files (was 26)
           · context-vigilance-kit: 4
           · memopop-ai: 2
           · memopop-ai/memopop-native: 6
           · memopop-ai/memopop-orchestrator: 72
           · memopop-ai/memopop-site: 3
           · dididecks-ai: 8
           · augment-it: 12

context-v : 125 files (was 54)
           · memopop-ai/memopop-orchestrator: 70  (the big jump)

The from provenance field on each rolled file is now <child>/<app> for nested entries, which the existing changelog and context-v list pages render as a proper provenance pill (memopop-ai/memopop-native).

The convention generalizes: any future top-level child that turns into a pseudomonorepo (apps + changelog inside each app) gets its inner apps surfaced automatically, no rollup-sync edit required.

`/studies/` — a real landing page for the pinned reading

The ch.05 · studies row on the homepage was the only one without a destination. It now has one at splash/src/pages/studies/index.astro, which:

Reads itself from the source. At build time it walks each study's .gitmodules for the authoritative pinned-repo list and lifts the first > blockquote from each README as the working question. Bump a study's submodules and the next splash build shows the change without touching this page.
Has a hero with the same eyebrow/gradient/lede shape as the homepage, plus a meta-strip with study count and total pinned-repo count.
Renders one section per study, each with: numbered eyebrow, h2 title, blockquote of the question (accent-bar left border), meta row (pinned-count pill · study repo → external link · #slug permalink anchor), and a responsive grid of pinned-repo cards. Each card shows the local slug (mono, larger) above owner/name (mono, dim) and links to the upstream GitHub URL.
Has an outro — ch · why pin — explaining the discipline in three bench-cards: grounded (read source, not folklore), stable (SHA-pinned, doesn't rot), scoped (one study, one question).

The homepage's ch.05 section now has a browse all studies → link in the section header, and each of its three cards is a real <a> deep-linking to /studies/#<slug> with a lift-on-hover affordance. Studies also joined the top nav in Header.astro, slotted between Products and Changelog.

OpenPanel analytics, finally

The other splash sites (astro-knots, content-farm, lfm, context-vigilance-kit) have been wired for OpenPanel for over a week. ai-labs/splash was the holdout. Same pattern, no surprises:

splash/src/components/Analytics.astro — OpenPanel-only (no Fathom history here to layer in), gated on import.meta.env.PROD && openPanelClientId so localhost can't pollute the dashboard and a misconfigured CI can't ship a broken <script> tag. The familiar queue-then-flush op shim, then https://openpanel.dev/op1.js deferred.
splash/src/layouts/BaseLayout.astro — imports and renders <Analytics /> at the tail of <head> so every page emits the script tags.
splash/.env.example — new, documents OPENPANEL_CLIENT_ID with the two operational gotchas (repo Variable not Secret; OpenPanel project must have https://lossless-group.github.io/ on its supported-domains list, otherwise events 401 server-side — the bug that bit the other three splashes for a full day).
.github/workflows/pages.yml — env: OPENPANEL_CLIENT_ID: ${{ vars.OPENPANEL_CLIENT_ID }} on the Build step.
context-v/prompts/Setup-Analytics-Across-Sites.md (at the monorepo level) — appended a row to the GitHub Pages rollout matrix capturing the wiring + the three remaining manual steps.

Three manual steps remain (dashboard work, not commit-able): create the OpenPanel project (suggested name splash_ai-labs), set the OPENPANEL_CLIENT_ID repo Variable, add https://lossless-group.github.io/ to the project's supported domains.

Status-line refresh on `context-vigilance-kit`

The product card's status: v0 · 583 files indexed was stale — the kit now powers four ChromaDB collections, not just the original context-v indexer. The card now reads v0 · four collections live with a lede that names all four (context-vigilance-corpus, lossless-changelog, claude-code-sessions, claude-code-tool-traces) and the MCP wiring.

One graduated card removed

The "graduated" strip on the homepage used to include investment-memo-orchestrator with the note "now MemoPop backend." But the orchestrator no longer stands as its own root — it lives nested inside memopop-ai at apps/memopop-orchestrator/. Calling it "graduated to its own home" reads as the opposite of where it ended up. Dropped the card; the memopop-ai product card's lede already preserves the lineage: "the backend is the former investment-memo-orchestrator, now nested."

Parent-level exploration: shared auth for the applied-AI-labs

Not on the splash, but related to the shape of where ai-labs is going: a new exploration landed at context-v/explorations/Shared-Auth-for-Applied-AI-Labs.md (versioned to 0.0.0.3). Three of the active children (memopop-ai, dididecks-ai, augment-it) all need an auth system soon, at a scale that doesn't justify Auth0 / Clerk costs but does justify shared patterns. The exploration captures the recommended shape (TS @lossless/auth-core + Python validator companion, per-app libSQL via Turso, lossless_id UUID + auth_events_outbox as the roll-up seam, org-as-firm dual vocabulary, two-surface viewer auth for published memos on client subdomains), the six forks considered (with rejected paths recorded), and three structured sections — user stories, hard/soft/out requirements, open ideas — ready for Mike to riff into. Next step is a sibling spec.

How to Test

Local rebuild + dev preview:
BASH
```
cd ai-labs/splash
pnpm rollup:sync    # regenerates src/rollup/* including the new sub-children
pnpm dev
```
Open http://localhost:4321/lossless-ai-labs/ — hero meta-strip should read 4 active. Click Studies in the top nav — /studies/ should render with three sections, each lifting its question from the upstream READMEs.
Verify nested rollup picked up the new memopop-native entry: splash/src/rollup/changelog/memopop-ai/memopop-native/2026-05-17_01.md should exist with from: "memopop-ai/memopop-native" in its frontmatter.
OpenPanel (after the three manual steps): push to main or master, wait for the Actions deploy, then view-source:https://lossless-group.github.io/lossless-ai-labs/ should show the inline op queue + the deferred openpanel.dev/op1.js script tag near the end of <head>. Verify with DevTools Network in an incognito window (extensions block openpanel.dev by default) — look for a POST api.openpanel.dev/track returning 200 (a 401 means the supported-domains step on OpenPanel got skipped).

What's Next

The three OpenPanel manual steps to finish the analytics rollout.
The Trustworthy Citations Phase 1 work (orchestrator-side) is still the original thread we were on before this splash sweep — apps/memopop-orchestrator/context-v/plans/Trustworthy-Citations-Source-Harvester-Rollout.md. Phase 1 is the no-architecture fixes: the self-disabling citation_validator.py:302 clause, the HEAD-only urllib probe replaced with a real GET, the verdict ladder, the reputable-gated publisher allow-list.
The shared-auth exploration converges next into a spec at context-v/specs/Shared-Auth-Core-Package.md, with per-app implementation prompts following.
An optional follow-up on the splash itself: a small "cross-cutting at the parent" section that surfaces ai-labs's own context-v/explorations/ and context-v/plans/ separately from the child-rolled feed. Would have surfaced today's auth exploration automatically.