diff --git a/AGENTS.md b/AGENTS.md
index 1f1e172..a4df616 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -10,7 +10,8 @@ debugging, because most entries describe something that looked like one bug and
 turned out to be another. Then
 [`docs/NEXT_STEPS.md`](docs/NEXT_STEPS.md) for current recommendations and known
 deferred fixes. Mobile interaction work also has its own detailed plan in
-[`docs/mobile_enhance.md`](docs/mobile_enhance.md).
+[`docs/mobile_enhance.md`](docs/mobile_enhance.md). The CDN-to-Vite move is
+[`docs/vite_build.md`](docs/vite_build.md).
 
 ## Tests earn their place by catching a real bug
 
@@ -122,6 +123,9 @@ implemented in one go:
   authenticated by OIDC. Read it before adding state to `Runtime`: new state is
   either genuinely shared (market data) or belongs to a user, and knowing which
   now is much cheaper than untangling it later.
+- `docs/vite_build.md` — pin and hash the frontend, stay on Coolify, do not
+  split components on the way. A production Dockerfile first, then Vite;
+  never a root `package.json` while nixpacks is still the builder.
 
 Do not build local user accounts. The destination is OIDC, so password storage
 would be written and then deleted.
diff --git a/docs/NEXT_STEPS.md b/docs/NEXT_STEPS.md
index 2c51a10..70fd504 100644
--- a/docs/NEXT_STEPS.md
+++ b/docs/NEXT_STEPS.md
@@ -16,6 +16,13 @@ and transport failure by carrying exact trade time, the last settled minute,
 source receipt time and an application heartbeat over the existing WebSocket.
 Any stale threshold must account for Yahoo's declared delay.
 
+## Frontend build
+
+The CDN-to-Vite plan is in `docs/vite_build.md`. First tranche is a production
+`Dockerfile` that reproduces today's nixpacks image, so a later `package.json`
+cannot make Coolify treat this as a Node app. Do not start the file move until
+that deploy has been seen live.
+
 ## Mobile authoring
 
 The detailed plan is in `docs/mobile_enhance.md`. Recommended first tranche:
diff --git a/docs/color_refactor.md b/docs/archived/color_refactor.md
similarity index 100%
rename from docs/color_refactor.md
rename to docs/archived/color_refactor.md
diff --git a/docs/esquotes.md b/docs/archived/esquotes.md
similarity index 100%
rename from docs/esquotes.md
rename to docs/archived/esquotes.md
diff --git a/docs/implementation.md b/docs/implementation.md
index 642c2e8..31aaff7 100644
--- a/docs/implementation.md
+++ b/docs/implementation.md
@@ -1042,3 +1042,18 @@ at `x=1082`, producing the long segment. The first fallback still failed because
 interpolate between the surrounding displayed candles' own pixel coordinates,
 which puts `:30` halfway between the hourly candles and remains correct even
 when another series has inserted foreign timestamps into the shared scale.
+
+### 2026-08-14 — a crease at the live edge was two renderers disagreeing
+
+The canvas `LineSeries` holds the historical portion of a manual trendline; an
+SVG segment projects it through future whitespace. `updateBar()` updated the
+candle but did not extend the canvas line when a new displayed bar opened. The
+SVG correctly started at that new bar, producing a visible crease at the live
+edge even after its slope was corrected.
+
+New displayed bars now append one source-space sample to each visible sloped
+manual series. This is deliberately not a full `syncLevels()` call: the source
+geometry and all previous line points are unchanged, and a per-line
+`series.update()` keeps the canvas and SVG endpoints identical without adding
+per-tick redraw work. Browser coverage opens a new 5m candle and asserts that
+the canvas line reaches the same time and price as the projection.
diff --git a/docs/plan.md b/docs/plan.md
index 3544fd9..a0f3f78 100644
--- a/docs/plan.md
+++ b/docs/plan.md
@@ -36,8 +36,8 @@ See [§14](#14-why-execution-is-out-of-scope) for why, and for the seam left beh
 > **For current work, start with `AGENTS.md`**, which every agent loads
 > automatically. It points at the live planning documents:
 > `docs/NEXT_STEPS.md` for the short list, `docs/async_refactor.md`,
-> `docs/multi_user.md`, `docs/feature_undo.md` and `docs/mobile_enhance.md` for
-> designs not yet built.
+> `docs/multi_user.md`, `docs/feature_undo.md`, `docs/mobile_enhance.md` and
+> `docs/vite_build.md` for designs not yet built.
 >
 > **`main` deploys to production.** A push triggers a Forgejo webhook and
 > Coolify rebuild of <https://chart.amow.com>. That is the intended workflow now,
@@ -90,7 +90,7 @@ if your parser doesn't filter those, that fixture will catch it.
 | Decision | Choice | Why |
 |---|---|---|
 | Backend | FastAPI (already scaffolded) | Repo already runs it; native WebSocket support |
-| Frontend | Vue 3 from CDN, **no build step** | Matches existing `static/` setup; keeps deploy trivial |
+| Frontend | Vue 3 from CDN, **no build step** | Matches existing `static/` setup; keeps deploy trivial. Destination is Vite — see `docs/vite_build.md`. Do not treat this row as a reason to reject that move. |
 | Charting | TradingView Lightweight Charts **v5.2.0**, standalone build | Apache-2.0, canvas, built for incremental realtime updates |
 | Data source | **Pluggable `MarketDataSource`.** Yahoo first, Schwab later | Yahoo needs no API key *and* has the history Schwab lacks — see §2.1 |
 | Persistence | **In-memory first**, behind a `BarStore` interface | User confirmed deferring persistence is fine for v1 |
diff --git a/docs/vite_build.md b/docs/vite_build.md
new file mode 100644
index 0000000..ce7034f
--- /dev/null
+++ b/docs/vite_build.md
@@ -0,0 +1,311 @@
+# Vite build — from CDN script tags to a real frontend
+
+**Status: tracked, not started.** A direction to refactor toward, not a project
+with a date. Each phase below is worth shipping on its own; none of it is
+speculative scaffolding for a component rewrite.
+
+Today `static/index.html` loads Vue 3, Lightweight Charts 5.2.0 and Font Awesome
+7.3.1 from unpkg, then two plain scripts. FastAPI serves those files and stamps
+`?v=` onto every `/static/` URL. Production is Coolify + nixpacks + a Python
+`Procfile`. That is the setup this document replaces.
+
+## The goal is not "more Vue"
+
+The target is **a pinned, hashed, same-origin frontend** that we can grow
+without unpkg and without a Python hasher. It is not a component split, not
+TypeScript, not a router, and not leaving Coolify.
+
+- `chart.js` stays a plain class. Vue still must not wrap chart or series
+  objects in `ref()` / `reactive()`.
+- `window.__chart` stays. E2E and diagnostic work depend on it.
+- One `App.vue` holding today's template and `setup()`. Do not extract the
+  color picker or the tool panels in the same change.
+- Stay on Coolify. The friction is nixpacks autodetection, not the platform.
+
+## What is already right
+
+- Vue 3 Composition API in `static/app.js` (`createApp`, `ref`, `computed`,
+  `watch`, `onMounted`). That maps 1:1 onto `vue` from npm.
+- `ConfluenceChart` is already framework-free. It only needs `export` instead
+  of `window.ConfluenceChart`, and ESM named imports instead of the
+  `LightweightCharts` global.
+- FastAPI already owns `/`, `/api`, `/ws`. The built SPA still comes from
+  that origin. Do not put a Vite server in production.
+- Asset hashing exists because a tab left open kept running yesterday's JS
+  (`main.asset_version`, `tests/test_asset_versioning.py`). Vite's content
+  hashes replace that rewriter; the *reason* does not go away.
+
+## Constraints this repo will punish you for forgetting
+
+- **The agent is on a different machine from the user's browser.** Local Vite
+  on `localhost:5173` is invisible to them. Whatever serves the UI in dev must
+  still be reachable as `hera.local:8010` (or whatever host port compose
+  publishes). HMR has to work across that hop, or we do not use HMR.
+- **`--reload` plus an 82-second seed.** Never put a scratch `.py` in the repo
+  root. Frontend files are safe; uvicorn watches Python. Do not "help" by
+  adding a Python build helper at the root.
+- **Every push to `main` is a production deploy**, and a deploy restarts the
+  market stream. The Vite cutover is one of those deploys. Land the production
+  Dockerfile *before* a root `package.json` exists, or nixpacks may decide
+  this is a Node app and the site goes dark.
+- **E2E hits `http://api:8000`**, waits on `window.__chart.bars`, and uses
+  `--lang=en-US`. None of that changes. A blank canvas after the move is
+  still the locale bug until proven otherwise.
+- **Pin what unpkg currently pins.** Lightweight Charts **5.2.0** and Font
+  Awesome **7.3.1**. Vue's CDN tag is `vue@3` (floating). Pin a current Vue
+  3.x on the way in; do not upgrade LWC in this work. v5 series creation is
+  `chart.addSeries(CandlestickSeries, opts)` — the v4 helpers do not exist.
+- **One uvicorn worker, forever**, until the streamer is a separate process.
+  The Dockerfile `CMD` is the `Procfile` line. Do not add `--workers`.
+
+## Target layout
+
+```
+frontend/
+  package.json
+  package-lock.json          committed
+  vite.config.js
+  index.html                 Vite entry; empty #app
+  src/
+    main.js                  createApp(App).mount('#app')
+    App.vue                  today's markup + today's setup()
+    chart.js                 export class ConfluenceChart
+    style.css                moved from static/
+dist/                        gitignored; Vite outDir, served by FastAPI
+Dockerfile                   production; Coolify prefers this over nixpacks
+```
+
+`static/` goes away when FastAPI is serving `dist/` and the e2e suite is green.
+Do not keep both as a fallback — a missed build would silently serve the CDN
+app.
+
+Suggested `frontend/src/main.js`:
+
+```js
+import { createApp } from 'vue';
+import '@fortawesome/fontawesome-free/css/all.min.css';
+import './style.css';
+import App from './App.vue';
+
+createApp(App).mount('#app');
+```
+
+Suggested chart import (names used today):
+
+```js
+import {
+  createChart,
+  CandlestickSeries,
+  HistogramSeries,
+  LineSeries,
+  LineStyle,
+  LineType,
+  CrosshairMode,
+  TickMarkType,
+} from 'lightweight-charts';
+```
+
+Keep `export default { setup() { ... return { ... }; } }` in `App.vue`.
+`<script setup>` is a rewrite of the return bag for no gain.
+
+## Dev: same origin, same port
+
+A Vite dev server on 5173 is the usual tutorial and the wrong default here.
+The user's browser already has one URL. Adding a second public port, plus an
+HMR websocket that has to reach a remote host, is how this loses a day.
+
+**Default:** a Node sidecar runs `vite build --watch` into `dist/`. The
+existing `api` service serves that directory at `/` exactly as production
+will. Compose still publishes one port. Edits to `.vue` / `.js` / `.css`
+rebuild hashed assets; the next refresh picks them up. No HMR, no second
+origin, no proxy for `/ws`.
+
+```yaml
+frontend:
+  image: node:22-alpine
+  working_dir: /app/frontend
+  volumes:
+    - .:/app
+  command: sh -c "npm ci && npm run build -- --watch"
+```
+
+`Dockerfile.dev` stays Python-only. Do not install Node in the API image.
+
+Optional later, not part of the move: `vite` with `server.host: true` and
+`server.hmr` pointed at the machine the *browser* can see. Only worth it if
+the watch-and-refresh loop is actually painful.
+
+`vite.config.js` needs nothing clever for the default path — `base: '/'`,
+`build.outDir` set so FastAPI and the watcher agree (repo-root `dist/` or
+`frontend/dist/`, pick one and use it everywhere). No `/api` proxy until
+someone runs the Vite dev server.
+
+## Production: Dockerfile, not nixpacks
+
+Coolify builds a Dockerfile if one exists, and ignores the `Procfile`. Land
+that switch as its own deploy, *reproducing today's image*, before the
+frontend exists:
+
+```dockerfile
+FROM python:3.12-slim
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8000
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
+```
+
+Then, when `frontend/` exists, make it multi-stage:
+
+```dockerfile
+FROM node:22-alpine AS frontend
+WORKDIR /src
+COPY frontend/package.json frontend/package-lock.json ./
+RUN npm ci
+COPY frontend/ ./
+RUN npm run build
+
+FROM python:3.12-slim
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+COPY --from=frontend /src/dist /app/dist
+EXPOSE 8000
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
+```
+
+Add a `.dockerignore` so `data/`, `.venv`, `node_modules`, `artifacts/` and
+`.env` never enter the build context. A missing ignore is how the Schwab
+token or the drawing store gets baked into an image.
+
+Unchanged, and not Coolify's problem:
+
+- env vars (`CHART_PASSWORD`, `LIVE_SOURCE`, Schwab keys, ntfy)
+- the persistent volume at `/app/data`
+- `SOURCE_COMMIT` → `/api/version` → `bin/wait-deploy`
+- the domain registered as `chart.amow.com:8000` (Traefik target port)
+
+The first multi-stage deploy will be a **cold** build (Node layer is new).
+Expect the ~90s end of the current range, plus `npm ci`. The old container
+keeps serving until the swap; ignore the usual one-minute 502.
+
+## FastAPI after the cutover
+
+`GET /` reads `dist/index.html` and still sends `Cache-Control: no-store`.
+The document must never be cached, or the hashed filenames inside it are the
+stale thing instead — same reason as today.
+
+Mount Vite's hashed directory, not a rewrite pass:
+
+```python
+app.mount("/assets", StaticFiles(directory=DIST_DIR / "assets"), name="assets")
+```
+
+Those files can be cached for a long time (`immutable`, or a one-year
+`max-age`). Vite changes the filename when the content changes.
+
+Delete `asset_version()` and the `ASSET_REF` rewrite. They hash `static/*`
+and would either no-op or stamp `?v=` onto URLs Vite already uniquely named.
+
+`tests/test_asset_versioning.py` keeps its purpose, changes its evidence:
+
+- `/` is `no-store` and references `/assets/…` with a content hash
+- hashed asset URLs do not need `?v=`
+- a rebuild after editing a frontend source file changes the hash in the
+  HTML (this one needs the built `dist/` in the test fixture, or a tiny
+  committed stub `dist/` used only by that test — do not hit `npm` from
+  pytest)
+
+## Phases
+
+Each is independently deployable. Do not fold 2–4 into the Dockerfile PR.
+
+### Phase 1 — Production Dockerfile, still CDN
+
+Add `Dockerfile` + `.dockerignore`. Confirm `git push && bin/wait-deploy`
+and `/api/version`. nixpacks is gone; the site is byte-identical.
+
+This is the phase that makes a later `package.json` safe.
+
+### Phase 2 — Scaffold `frontend/`, no cutover
+
+`npm create vite@latest` (Vue, JS, no TS). Pin `vue`, `lightweight-charts@5.2.0`,
+`@fortawesome/fontawesome-free@7.3.1`. Commit `package-lock.json`. Add
+`node_modules/` and `dist/` to `.gitignore`.
+
+Do not add `package.json` at the repo root. nixpacks is already gone after
+phase 1; keep Node metadata under `frontend/` anyway so a future builder
+cannot mis-detect the app.
+
+### Phase 3 — Move the two files, same behaviour
+
+- `static/chart.js` → `frontend/src/chart.js` with ESM imports and `export`.
+  Drop `window.ConfluenceChart`.
+- `static/app.js` `setup()` + the `#app` inner HTML → `frontend/src/App.vue`.
+  `import { ConfluenceChart } from './chart.js'`. Keep assigning
+  `window.__chart = chartApi` in `onMounted`.
+- `static/style.css` → `frontend/src/style.css`.
+- Font Awesome via the npm CSS import, not the unpkg `<link>`.
+
+The global Vue build includes the compiler. Vite's Vue plugin compiles SFCs
+and ships the runtime-only build. That is why the markup has to live in
+`App.vue` (or another compiled module), not as HTML children of `#app`.
+
+`npm run build` locally. Open the `dist/` preview against a running API only
+if you need a sanity check; the real proof is phase 4.
+
+### Phase 4 — FastAPI serves `dist/`, delete `static/`
+
+Point `index()` and the static mount at `dist/`. Add the compose `frontend`
+watcher. Rewrite `test_asset_versioning.py`. Run pytest and `./bin/e2e`.
+Delete `static/`. Update the Dockerfile to the multi-stage form. Update
+README / `docs/plan.md` §1 and §9 so they no longer describe unpkg.
+
+After this, a frontend change that is not rebuilt is not deployed. The
+multi-stage `Dockerfile` is what builds it on Coolify. Locally the watcher
+is what builds it. There is no third path.
+
+## What not to do in this work
+
+- Do not extract Vue components, add Pinia, Vue Router, or TypeScript.
+- Do not upgrade Lightweight Charts.
+- Do not put a Vite origin in production, or a second public port in compose.
+- Do not leave Coolify, add workers, or move env/volume/TLS anywhere else.
+- Do not keep `static/` as a fallback once `dist/` is the source of truth.
+- Do not add a root `package.json` before phase 1 is live.
+- Do not run `npm` from pytest or from the API container.
+
+## Verify
+
+Same commands as today, plus a frontend build:
+
+```bash
+docker exec chart-api-1 sh -c "cd /app && python -m pytest -q"
+./bin/e2e
+```
+
+E2E still waits on `window.__chart.bars`. If the canvas is blank, check
+`--lang=en-US` before the bundler. If icons are missing, the FA CSS import
+did not land. If drawings or the socket die, the page origin changed and
+`/ws` is not on the same host.
+
+After the first multi-stage deploy:
+
+```bash
+git push && bin/wait-deploy
+curl -fsS https://chart.amow.com/api/health
+curl -fsS https://chart.amow.com/api/version
+```
+
+View-source on `/` should show `/assets/…` with a hash and no unpkg script
+tags. A hard refresh on a tab that was open across the deploy should pick
+up the new JS without a `?v=` rewriter.
+
+## When this is done
+
+`docs/plan.md` §1 currently says "Vue 3 from CDN, **no build step**". That
+row becomes the lie the day phase 4 ships — change it in the same commit,
+along with §9's script-tag snippet and the README layout line for `static/`.
+This file then becomes history, like M0–M10 in the plan.
diff --git a/static/chart.js b/static/chart.js
index 64a59c6..4c83622 100644
--- a/static/chart.js
+++ b/static/chart.js
@@ -533,17 +533,30 @@ class ConfluenceChart {
     // the source: a bar behind the last one has nothing to contribute.
     const last = this.bars[this.bars.length - 1];
     if (last && bar.t < last.t) return;
+    const advanced = !last || bar.t > last.t;
     this.keepLivePriceVisible(bar.c);
     this.candles.update(this.toCandle(bar));
     this.updateCurrentPrice(bar.c);
     this.volume.update(ConfluenceChart.toVolume(bar));
     if (this.bars.length && this.bars[this.bars.length - 1].t === bar.t) this.bars[this.bars.length - 1] = bar;
     else this.bars.push(bar);
+    if (advanced) this.appendManualLineSamples(bar);
     this.renderAnchorHandles();
     this.renderContextLabels();
     this.renderTrendlineProjections();
   }
 
+  appendManualLineSamples(bar) {
+    const index = this.bars.length - 1;
+    for (const level of this.levels) {
+      if (level.kind !== 'manual' || level.hidden || ConfluenceChart.isFlat(level)
+          || bar.t < level.anchor_t || bar.t > (level.cutoff_t ?? Infinity)) continue;
+      const entry = this.levelSeries.get(level.id);
+      const value = this.linePriceAtIndex(level, index);
+      if (entry && value != null) entry.series.update({ time: bar.t, value });
+    }
+  }
+
   setComments(comments) {
     this.comments = comments;
     this.renderComments();
diff --git a/tests/e2e/trendline.test.mjs b/tests/e2e/trendline.test.mjs
index a38b866..1f09101 100644
--- a/tests/e2e/trendline.test.mjs
+++ b/tests/e2e/trendline.test.mjs
@@ -158,6 +158,46 @@ test('future 5m projections keep their slope after the settlement break',
       assert.ok(Math.abs(result.futureChange - result.historicalChange) < 1e-9,
         `future slope changed from ${result.historicalChange} to ${result.futureChange}`);
       assertNoPageErrors(page, assert);
+  });
+});
+
+test('a new 5m candle extends the line before its future projection',
+  { timeout: 180000 }, async () => {
+    await withChart(async page => {
+      const result = await page.evaluate(() => {
+        const c = window.__chart;
+        const step = 5 * 60;
+        const bars = Array.from({ length: 20 }, (_, index) => ({
+          tf: '5m', t: 100000 + index * step, o: 100, h: 101, l: 99, c: 100,
+          v: 1, closed: true,
+        }));
+        const line = {
+          id: 'handoff-probe', kind: 'manual', tf: '5m', side: 'support', label: 'probe',
+          anchor_t: bars[0].t, anchor_p: 100, last_t: bars[4].t,
+          slope: 1 / (4 * step), cutoff_t: null, hidden: false, color: '#65b7cf',
+          line_width: 2, provisional: false,
+        };
+        c.setBars(bars);
+        c.setTrendlineGeometry({
+          mode: 'source_tf', series: { '5m': { times: bars.map(bar => bar.t), duration: step } },
+        });
+        c.syncLevels([line]);
+        const next = { ...bars.at(-1), t: bars.at(-1).t + step };
+        c.updateBar(next);
+        const data = c.levelSeries.get(line.id).series.data();
+        return {
+          seriesLastTime: Number(data.at(-1).time),
+          seriesLastPrice: data.at(-1).value,
+          expectedTime: next.t,
+          expectedPrice: c.linePriceAtIndex(line, c.bars.length - 1),
+        };
+      });
+
+      assert.equal(result.seriesLastTime, result.expectedTime,
+        'the canvas line stopped before the projection began');
+      assert.equal(result.seriesLastPrice, result.expectedPrice,
+        'the canvas line and projection disagreed at the live edge');
+      assertNoPageErrors(page, assert);
     });
   });