feat(home): Rebuild the homepage on a declarative motion system

.gitignore

@@ -159,3 +159,10 @@ playwright/.cache/
 # `astro check` (and therefore `npm run build`) runs first, so nothing needs it
 # committed — tracked, it just produced a dirty tree on every dev server start.
 .astro/
+
+# Superpowers brainstorm scratch (visual companion)
+.superpowers/
+
+# macOS Finder metadata. Browsing public/ in Finder drops these into the
+# deploy directory, where they would be served as real files.
+.DS_Store

docs/adding-content.md

@@ -115,6 +115,58 @@ The card on `/blog` will link directly to your content. PhilaCon Valley amplifie
 
 Same as above but without `externalUrl` and `platform`. Write the full content in the Markdown body and it will be hosted on the site.
 
+## Adding Event Photos
+
+The homepage's "The room" section renders a five-photo mosaic under the headline
+"Every photo is a real Thursday." Until there are photos, the section renders no
+mosaic at all and the headline reads "Come see the room." instead — the claim and
+the evidence ship together or not at all.
+
+Two steps. Put the image file in `public/images/gallery/`:
+
+```
+public/images/gallery/patch-002-pairing.webp
+```
+
+Then add one JSON file per photo in `src/content/gallery/`:
+
+```json
+// src/content/gallery/patch-002-pairing.json
+{
+  "image": "/images/gallery/patch-002-pairing.webp",
+  "alt": "Two people pairing on a laptop at a long table",
+  "event": "PATCH 002: Tap In NFC Lab",
+  "date": "2026-07-30"
+}
+```
+
+### Field reference
+
+| Field     | Required | Notes                                                                           |
+| --------- | -------- | ------------------------------------------------------------------------------- |
+| `image`   | Yes      | Site-absolute path into `public/`, starting with `/images/`. Not a file import. |
+| `alt`     | Yes      | What is happening in the photo, for screen readers. Not "event photo".          |
+| `event`   | Yes      | Which night it was taken at.                                                    |
+| `date`    | Yes      | `YYYY-MM-DD`. The five most recent photos are the ones the homepage shows.      |
+| `caption` | No       | Optional visible caption.                                                       |
+
+The homepage takes the five newest by `date`, so adding a sixth retires the
+oldest rather than growing the grid. YAML (`.yaml` / `.yml`) works too if you
+prefer it — the collection accepts both.
+
+**The first photo gets a double-height cell.** The mosaic's leftmost tile spans
+two rows, so whichever entry sorts first should be a **portrait** photo. A
+landscape shot there gets cropped to a tall slot, which cuts people off at both
+edges — a wide group photo is the worst thing to put in it. Give the group shot
+one of the four wide cells instead.
+
+Sorting is newest `date` first; entries sharing a date fall back to filename
+order. So among photos from the same night, `patch-001-pairing.json` comes
+before `patch-001-together.json`.
+
+**Before you add a photo:** make sure the people in it are okay with being on a
+public homepage. There is no way to un-publish a face someone finds later.
+
 ## What About GitHub Repos?
 
 The `/projects` page automatically displays all repositories from the [philaconvalley GitHub organization](https://github.com/philaconvalley). If your project is in the org, it shows up with no file needed.

docs/perf-baseline.md

@@ -0,0 +1,45 @@
+# Performance baseline
+
+Captured on `waskar/site-redesign` at commit `de52793`, immediately
+before the motion system landed. Spec §5.5 gates the redesign on staying within 3 performance
+points of these numbers.
+
+Command: `npx --yes @lhci/cli@0.15.x autorun --config=.lighthouserc.json --collect.numberOfRuns=3` (resolved to `@lhci/cli@0.15.1`)
+Machine: local dev (numbers are not comparable to CI — re-baseline there if CI is ever the gate).
+
+| URL          | Performance | Accessibility | Best practices | SEO   |
+| ------------ | ----------- | ------------- | -------------- | ----- |
+| `/`          | `79`        | `95`          | `93`           | `100` |
+| `/about/`    | `84`        | `95`          | `93`           | `100` |
+| `/events/`   | `92`        | `96`          | `75`           | `100` |
+| `/projects/` | `81`        | `93`          | `96`           | `100` |
+
+## Read the performance column as a median, not a measurement
+
+The table records the median of three runs. The runs were not tight. `/about/` scored
+`0.77 / 0.84 / 0.95` — an 18-point spread on one machine, in one sitting, against one build.
+The other pages were sampled the same way and there is no reason to think they are steadier.
+
+That matters because spec §5.5's gate is ±3 points, and the noise here is six times the gate.
+Three local runs cannot resolve a 3-point change: a page that genuinely regressed by 3 points
+will often score _higher_ than this table, and an unchanged page will often score lower. Taken
+at face value, `84` sends the next implementer chasing a regression that does not exist, or
+waving through one that does.
+
+So evaluate the gate like this:
+
+1. **CI is the enforcement surface.** If the numbers here are ever used to block a change, they
+   must first be re-taken in CI, where the machine is at least consistent between runs. The
+   figures above are local and indicative — useful for spotting a page that halved, not for
+   adjudicating three points.
+2. **Locally, compare medians of at least 5 runs, taken back-to-back in one sitting**, with the
+   same command and nothing else running. Compare median against median, never a single run
+   against this table.
+3. **Record the spread, not just the median**, whenever this file is updated. A median with no
+   spread beside it reads as precision the measurement does not have.
+4. **Treat a sub-5-point local move as no signal at all.** Investigate the trace (LCP, TBT, the
+   actual bytes shipped) rather than the score, because those are stable where the composite
+   score is not.
+
+If any page later drops more than 3 performance points against this table, measured that way,
+spec §5.5 applies: depth drops to two planes site-wide before anything else is cut.