@@ -57,7 +57,7 @@ jobs:
         working-directory: app
         # Every suite except parity_silhouette, which needs the 8 MB mask fixture
         # that is deliberately not committed.
-        run: node parity_test.mjs && node scan_score_test.mjs && node scan_steps_test.mjs && node overlay_test.mjs && node plan_test.mjs && node progress_test.mjs && node boot_test.mjs
+        run: node parity_test.mjs && node scan_score_test.mjs && node scan_steps_test.mjs && node overlay_test.mjs && node plan_test.mjs && node progress_test.mjs && node boot_test.mjs && node workouts_test.mjs
 
       - uses: actions/configure-pages@v6
       - uses: actions/upload-pages-artifact@v5
 
@@ -94,6 +94,7 @@ is how wrong numbers ship:
 | `plan_test.mjs` | plan maths and safety floors | 30 checks pass |
 | `progress_test.mjs` | history, crew board, goals, ranking | 61 checks pass |
 | `boot_test.mjs` | lazy model loading, retry after failure | 10 checks pass |
+| `workouts_test.mjs` | PR engine, sessions, training streak | 65 checks pass |
 
 ### A bug worth keeping
 
@@ -141,6 +142,61 @@ measurements are still worth having — they are what the plan, the ranking and
 trend are built on — but the honest summary is that this fixed Stage 1's weakest
 result, not the headline. The shipped app uses these models.
 
+## A training log, and knowing when to shout
+
+274 exercises, logged as sets, with records tracked per exercise. The part worth
+reading is what counts as a record, because that decides when the app is allowed
+to interrupt someone.
+
+Four records are tracked for a lifted set — heaviest weight, estimated 1RM, most
+reps at a given weight, and biggest single set — but **only two can raise a
+popup**. The reason is the back-off set. After a top set of 100 kg × 5, a light
+60 kg × 15 finisher is genuinely the most reps ever done at 60 kg and genuinely
+the biggest single-set volume. Both true, both worthless as news, and a tracker
+that celebrates them teaches you to ignore it within a week. Heaviest weight and
+estimated 1RM are exactly the records a lighter set *cannot* fake, so those two
+get the screen and the rest are shown quietly on the set itself.
+
+Three more decisions that fell out of testing:
+
+- **Nothing is a record the first time.** Otherwise every first set of every
+  exercise is a personal best — technically true, completely useless.
+- **Reps are only compared at the same weight.** 21 reps at 20 kg is not better
+  than 20 reps at 40 kg, and no arithmetic on those two numbers says otherwise.
+- **Epley is capped at 12 reps**, because above that it measures conditioning
+  and reports it as strength. It also returns the bar weight exactly for a
+  logged single, where the textbook formula claims 3.3% more than was lifted.
+
+Exercises carry one of three metrics — load, reps or time — and the metric
+decides both the input fields and the records. It is what stops the app
+congratulating someone for holding a *heavier* plank.
+
+## Installable, and it works with no signal
+
+`manifest.webmanifest` plus `sw.js` make it a home-screen app: on iOS, Share →
+Add to Home Screen gives a standalone app with its own icon. There is no native
+iOS build, and from a Windows machine there cannot be — an App Store binary
+needs macOS, Xcode and a paid Apple developer account. A PWA is the whole of
+what is buildable here, and for this app it costs nothing: camera, models and
+storage all work the same.
+
+The caching mirrors the loading split, for the same reason — precaching
+everything would mean the 25 MB download the lazy loading just removed:
+
+| Cache | Contents | When |
+|---|---|---|
+| shell | ~1 MB: markup, scripts, the 11 `.bin` models | precached on install |
+| heavy | ~24 MB: MediaPipe wasm and tflite | kept only once actually used |
+
+Verified by stopping the server and reloading: the app booted from cache, the
+models loaded, and the training log was intact.
+
+Cache-first was the obvious strategy and the wrong one. Scripts keep their
+filenames across deploys, so it serves last week's code until someone remembers
+to bump a version constant by hand — which bit during testing, when a stale
+`exercises.js` silently undid a fix I had just made. Everything but the
+immutable MediaPipe assets is stale-while-revalidate instead.
+
 ## Loading in the order the user arrives
 
 Measuring the boot rather than assuming it: the app fetched **24 MB and stood
 
@@ -362,6 +362,31 @@ localhost every one of these read as ~100 ms, which is why they survived so long
 blip must not brick the camera for the rest of the session. `boot_test.mjs`
 pins the sharing and the retry, 10 checks.
 
+## The training log
+
+`js/exercises.js` is a 274-row table — name, muscle, equipment, metric — parsed
+at load. The metric (`load`, `reps`, `time`) decides the input fields and which
+records exist, which is what keeps a plank from competing on weight.
+
+`js/workouts.js` holds the records engine and is pure apart from an injectable
+store, so all 65 checks run under Node. Sets are stored flat with their own
+timestamps and sessions are derived by calendar day: a stored "current session"
+is state that can be left open, double-opened or orphaned by a refresh, and none
+of those bugs can exist if the session is only a view.
+
+Only `weight` and `e1rm` raise a popup. A back-off set can top the rep and
+volume records without being news, and neither of those two can be reached by
+lifting something lighter. The first set of any exercise is never a record.
+
+### A bug the log found in the scanner
+
+`readCrew()` invents a profile when the device is new, with a fresh random id —
+and never wrote it down. Scan history hid this for months because `saveScan`
+persists the crew as a side effect of saving. The training log reads the active
+profile id without ever saving a scan, so every set was filed under a different
+person and no record ever had any history to beat. `ensureCrew()` now persists
+the invented profile, and `progress_test.mjs` pins the id across calls.
+
 ## Camera framing
 
 `zoomRange` reports what the hardware actually exposes and drives a slider;
 
@@ -4,7 +4,7 @@
   "private": true,
   "type": "module",
   "scripts": {
-    "test": "node parity_test.mjs && node parity_silhouette.mjs && node scan_score_test.mjs && node scan_steps_test.mjs && node overlay_test.mjs && node plan_test.mjs && node progress_test.mjs && node boot_test.mjs",
+    "test": "node parity_test.mjs && node parity_silhouette.mjs && node scan_score_test.mjs && node scan_steps_test.mjs && node overlay_test.mjs && node plan_test.mjs && node progress_test.mjs && node boot_test.mjs && node workouts_test.mjs",
     "start": "node server.mjs"
   },
   "dependencies": {
 
@@ -5,6 +5,7 @@ import { readFileSync } from "node:fs";
 import {
   loadHistory, saveScan, clearHistory, streak, deltas, badges, trendPoints, KEY,
   readCrew, profiles, activeProfile, setActiveProfile, addProfile, removeProfile, crewBoard,
+  CREW_KEY,
 } from "./public/js/progress.js";
 import { setRanks, rank, bandFor, percentileOf, rankLabel } from "./public/js/rank.js";
 import { bodyFatAge, nextMilestone, weeklyRate, etaTo, etaLabel } from "./public/js/goals.js";
@@ -281,5 +282,23 @@ console.log("\ngoals");
   check("one scan gives no rate", weeklyRate([falling[0]]) === null);
 }
 
+/* ---------------------------------------------------- a stable identity --- */
+// Found by building the training log on top of this: anything keyed by profile
+// id needs that id to still be there on the next call. readCrew invents one on
+// a new device, and for a long time only saveScan happened to write it back -
+// so a feature that read the id without ever saving a scan silently filed
+// every entry under a different person.
+console.log("\nThe invented first profile");
+{
+  const store = makeStore();
+  const a = activeProfile(store);
+  const b = activeProfile(store);
+  check("a fresh device gets a profile", !!a && !!a.id);
+  check("and the very same one next call", a.id === b.id, `${a?.id} vs ${b?.id}`);
+  check("because reading it wrote it down", store.getItem(CREW_KEY) !== null);
+  check("listing profiles agrees", profiles(store).length === 1);
+  check("and agrees on the id", profiles(store)[0].id === a.id);
+}
+
 if (failures) { console.error(`\nFAIL: ${failures} check(s) failed`); process.exit(1); }
 console.log("\nPASS - history, crew, goals and population ranking all behave");
@@ -11,6 +11,16 @@
 <link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=Bricolage+Grotesque:opsz,wght@12..96,600;12..96,700;12..96,800&family=Figtree:wght@400;500;600;700;800&display=swap">
 <link rel="stylesheet" href="styles.css">
 
+<!-- Installable: on Android this comes from the manifest, on iOS it does not -
+     Safari ignores the manifest's icons and name and reads these instead. -->
+<link rel="manifest" href="manifest.webmanifest">
+<link rel="apple-touch-icon" href="icons/apple-touch-icon.png">
+<meta name="apple-mobile-web-app-capable" content="yes">
+<meta name="mobile-web-app-capable" content="yes">
+<meta name="apple-mobile-web-app-status-bar-style" content="black-translucent">
+<meta name="apple-mobile-web-app-title" content="bodycomp">
+<meta name="description" content="Body fat and lean mass from two photos, validated against DXA, plus a training log with personal records. Everything runs on your device.">
+
 <!-- The module graph is three levels deep, so the browser discovers it one
      level per round trip: app.js, then its nine imports, then theirs. Naming
      the leaves here starts them all in the first wave. The manifest is the
@@ -101,6 +111,10 @@ <h1>Know your body fat.<br><span class="hl">Not just your weight.</span></h1>
         <svg viewBox="0 0 24 24"><path d="M5 12h14M13 6l6 6-6 6"/></svg>
       </button>
       <button class="btn secondary js-needs-models" type="button" id="sample" disabled>See a demo result</button>
+      <button class="btn ghost" type="button" data-go="train">
+        <svg viewBox="0 0 24 24"><path d="M4 9v6M7 7v10M17 7v10M20 9v6M7 12h10"/></svg>
+        Log a workout
+      </button>
     </div>
     <p class="fine">A fitness estimate, not a medical device.</p>
   </section>
@@ -439,6 +453,17 @@ <h2>Your body, by the numbers</h2>
 
     <!-- progress -->
     <div data-panel="progress" hidden>
+      <div class="card traincard">
+        <div class="card-head"><h3>Training</h3><span class="pill" id="tc-streak"></span></div>
+        <div class="tc-stats">
+          <div><b id="tc-sets">0</b><span>sets this week</span></div>
+          <div><b id="tc-vol">0</b><span>kg moved</span></div>
+          <div><b id="tc-ex">0</b><span>exercises tracked</span></div>
+        </div>
+        <p class="sub" id="tc-note"></p>
+        <button class="btn ghost" type="button" data-go="train">Open my training log</button>
+      </div>
+
       <div class="card">
         <div class="card-head"><h3>Crew board</h3><span class="pill" id="crew-count">--</span></div>
         <ol class="crew" id="crew-list"></ol>
@@ -514,6 +539,112 @@ <h2>Your body, by the numbers</h2>
     </div>
   </section>
 
+  <!-- ================================================================== train -->
+  <section class="screen" data-screen="train" hidden>
+    <div class="screen-head">
+      <p class="kicker">Training log</p>
+      <h2>What did you lift?</h2>
+    </div>
+
+    <div class="trainbar">
+      <div class="statlet"><b id="t-streak">0</b><span>day streak</span></div>
+      <div class="statlet"><b id="t-week-sets">0</b><span>sets this week</span></div>
+      <div class="statlet"><b id="t-week-vol">0</b><span>kg moved</span></div>
+    </div>
+
+    <button class="btn primary big" type="button" id="t-add">
+      <svg viewBox="0 0 24 24"><path d="M12 5v14M5 12h14"/></svg>
+      Log a set
+    </button>
+
+    <div id="t-empty" class="card empty" hidden>
+      <h3>Nothing logged yet</h3>
+      <p class="sub">Add your first set and the app starts tracking records for it. Nothing
+      counts as a personal best until there's something to beat &mdash; so your first set of an
+      exercise is just a baseline.</p>
+    </div>
+
+    <div id="t-sessions"></div>
+
+    <div class="card" id="t-records-card" hidden>
+      <div class="card-head"><h3>Your records</h3><span class="pill" id="t-records-count"></span></div>
+      <div id="t-records"></div>
+    </div>
+
+    <p class="fine">Logged on this device only, like everything else here.
+      <button class="linkish" type="button" id="t-clear">Delete my training log</button>
+    </p>
+  </section>
+
+</div>
+
+<!-- ====================================================== exercise picker -->
+<div class="sheet" id="picker" hidden>
+  <div class="sheet-back" data-close-picker></div>
+  <div class="sheet-body" role="dialog" aria-modal="true" aria-label="Choose an exercise">
+    <div class="sheet-grab"></div>
+    <div class="sheet-head">
+      <h3>Choose an exercise</h3>
+      <button class="iconbtn" type="button" data-close-picker aria-label="Close">
+        <svg viewBox="0 0 24 24"><path d="M6 6l12 12M18 6L6 18"/></svg>
+      </button>
+    </div>
+    <input id="pick-search" class="search" type="search" placeholder="Search 274 exercises" autocomplete="off">
+    <div class="chiprow" id="pick-muscles"></div>
+    <div class="picklist" id="pick-list"></div>
+  </div>
+</div>
+
+<!-- ============================================================ set entry -->
+<div class="sheet" id="setsheet" hidden>
+  <div class="sheet-back" data-close-set></div>
+  <div class="sheet-body" role="dialog" aria-modal="true" aria-labelledby="set-name">
+    <div class="sheet-grab"></div>
+    <div class="sheet-head">
+      <div>
+        <h3 id="set-name">Exercise</h3>
+        <p class="sub" id="set-sub"></p>
+      </div>
+      <button class="iconbtn" type="button" data-close-set aria-label="Close">
+        <svg viewBox="0 0 24 24"><path d="M6 6l12 12M18 6L6 18"/></svg>
+      </button>
+    </div>
+
+    <p class="lastset" id="set-last" hidden></p>
+
+    <div class="setfields" id="set-fields">
+      <label class="num" id="f-kg"><span>Weight</span>
+        <input id="in-kg" type="number" inputmode="decimal" step="0.5" min="0" placeholder="0"><i>kg</i>
+      </label>
+      <label class="num" id="f-reps"><span>Reps</span>
+        <input id="in-reps" type="number" inputmode="numeric" step="1" min="0" placeholder="0"><i></i>
+      </label>
+      <label class="num" id="f-time"><span>Time</span>
+        <input id="in-time" type="number" inputmode="numeric" step="1" min="0" placeholder="0"><i>sec</i>
+      </label>
+      <label class="num" id="f-dist"><span>Distance</span>
+        <input id="in-dist" type="number" inputmode="numeric" step="1" min="0" placeholder="0"><i>m</i>
+      </label>
+    </div>
+
+    <div class="setsteps" id="set-steps"></div>
+
+    <button class="btn primary big" type="button" id="set-save">Save set</button>
+    <div class="todayset" id="set-today"></div>
+  </div>
+</div>
+
+<!-- ================================================================== a PR -->
+<div class="prpop" id="prpop" hidden aria-live="polite">
+  <div class="prcard" role="status">
+    <div class="prburst" aria-hidden="true"></div>
+    <p class="prkicker" id="pr-kicker">New personal record</p>
+    <h2 id="pr-title">Heaviest ever</h2>
+    <p class="prvalue" id="pr-value"></p>
+    <p class="prwas" id="pr-was"></p>
+    <p class="prex" id="pr-ex"></p>
+    <button class="btn primary" type="button" id="pr-ok">Let's go</button>
+  </div>
 </div>
 
 <script type="module" src="js/app.js"></script>