@@ -831,16 +831,32 @@
          on. The reference styles the glyph rather than the button, so the button stays
          flat when pressed and only the mark changes. */
 
-      .traj-controls {
+      /* One sticky band for both filter axes.
+
+         The type row and the 异常 row decide the same question — which records the ledger shows —
+         so they are pinned as a single block rather than one of them following the other out of
+         view. Only the type row claimed `position: sticky` before, and it did not survive: with
+         the ledger below it and both rows scrolled away, a session filtered down to a few hundred
+         rows gave no sign that anything was filtered at all. The band carries its own background
+         because it now sits over the ledger, which is opaque underneath it.
+
+         Everything pinned below this measures its offset from it rather than assuming it — see
+         syncStickyOffsets(). The anomaly row wraps at narrow widths and the note under it comes
+         and goes, so its height is not a number anyone can write down once. */
+      .filter-sticky {
         position: sticky;
         top: 0;
-        z-index: 5;
+        z-index: 6;
+        background: var(--mcode-bg);
+        border-bottom: 1px solid var(--mcode-border);
+      }
+
+      .traj-controls {
         display: flex;
         flex: none;
         align-items: center;
         padding: 6px 16px;
         background: var(--mcode-bg);
-        border-bottom: 1px solid var(--mcode-border);
       }
 
       .traj-controls-actions {
@@ -941,7 +957,7 @@
 
       .overview {
         position: sticky;
-        top: 32px;
+        top: var(--sticky-band-h, 32px);
         z-index: 4;
         border-bottom: 1px solid var(--mcode-border);
         user-select: none;
@@ -1207,7 +1223,11 @@
 
       .turn-head {
         position: sticky;
-        top: 0;
+        /* Below every band above it, measured. It used to pin at 0 with z-index 2 and simply got
+           drawn over — its turn title and its kind summary landed on the type row's buttons and on
+           the overview's tick labels, and neither could be read. */
+        top: calc(var(--sticky-band-h, 32px) + var(--sticky-overview-h, 51px)
+          + var(--sticky-window-h, 45px));
         z-index: 2;
         display: flex;
         align-items: center;
@@ -1586,11 +1606,12 @@
 
       .window-bar {
         position: sticky;
-        /* Sticky offsets stack here: the type row owns 0–32 and the overview strip owns
-           32–83 (32 + the 50px plot + its 1px border), so a bar pinned at 0 would simply
-           hide behind both of them. z-index 3 keeps it under the overview (4) and over the
-           turn head (2). */
-        top: 83px;
+        /* Sticky offsets stack here, and every one of them below the filter band is measured
+           rather than written down: the band is the two filter rows (whose height changes when
+           异常 wraps or when its note appears) plus the overview strip. The old numbers — 0–32 and
+           32–83 — were correct for a layout where the 异常 row was not in the band, and they are
+           exactly the kind of constant that goes stale silently. syncStickyOffsets() owns them. */
+        top: calc(var(--sticky-band-h, 32px) + var(--sticky-overview-h, 51px));
         z-index: 3;
         padding: 8px 16px;
         background: var(--mcode-bg);
@@ -2058,7 +2079,8 @@ <h2 class="section-title" id="kpi-heading">关键指标</h2>
         <div class="kpi-grid" id="kpi-grid"></div>
       </section>
 
-      <div class="traj-controls" id="kind-controls" role="toolbar" aria-label="按类型筛选轨迹" aria-controls="overview" data-filter-scope>
+      <div class="filter-sticky" id="filter-sticky">
+        <div class="traj-controls" id="kind-controls" role="toolbar" aria-label="按类型筛选轨迹" aria-controls="overview" data-filter-scope>
         <nav class="traj-controls-actions" aria-label="按类型筛选轨迹">
           <button type="button" class="tbtn tbtn-action tbtn-all" id="kind-all" aria-pressed="true"
             aria-label="当前显示全部类型 · 点击只看空白" title="当前显示全部类型 · 点击只看空白">
@@ -2136,6 +2158,7 @@ <h2 class="section-title" id="kpi-heading">关键指标</h2>
           <div class="anomaly-quick" id="anomaly-controls" role="group" aria-labelledby="anomaly-label"></div>
           <span class="anomaly-bar-note" id="anomaly-mode-note" hidden></span>
         </div>
+      </div>
 
       <section class="overview" id="overview" aria-labelledby="overview-heading">
         <h2 class="sr-only" id="overview-heading">轨迹总览</h2>
@@ -3124,6 +3147,7 @@ <h2 class="section-title" id="inspector-heading">检视</h2>
         dom.staleBar = document.getElementById('stale-bar');
         dom.toolbar = document.getElementById('toolbar');
         dom.kindControls = document.getElementById('kind-controls');
+        dom.filterSticky = document.getElementById('filter-sticky');
         dom.sessionSelect = document.getElementById('session-select');
         dom.generationField = document.getElementById('generation-field');
         dom.generationSelect = document.getElementById('generation-select');
@@ -3139,6 +3163,7 @@ <h2 class="section-title" id="inspector-heading">检视</h2>
         dom.anomalyNote = document.getElementById('anomaly-mode-note');
         dom.anomalyMulti = document.getElementById('anomaly-multi');
         dom.windowBar = document.getElementById('window-bar');
+        dom.overview = document.getElementById('overview');
         dom.headerToggle = document.getElementById('header-toggle');
         dom.kpiToggle = document.getElementById('kpi-toggle');
         dom.kpiGrid = document.getElementById('kpi-grid');
@@ -6091,6 +6116,28 @@ <h2 class="section-title" id="inspector-heading">检视</h2>
 
       /* ---------------- top-level render ---------------- */
 
+      /**
+       * Publishes the heights of the pinned bands so everything below them can be positioned from
+       * a measurement instead of a number written down once.
+       *
+       * The alternative was three constants — 32 for the type row, 51 for the overview strip, 45
+       * for the window bar — and they were already wrong the moment the 异常 row joined the band:
+       * that row wraps at narrow widths and grows a line when its note appears, so no constant can
+       * stay right. This runs at the end of `render()`, after the note has been written and the
+       * buttons re-counted, which is the only moment the band's height is final. `offsetHeight`
+       * rather than `getBoundingClientRect().height`, so a fractional device-pixel ratio cannot
+       * make the bands overlap by a third of a pixel per poll.
+       */
+      function syncStickyOffsets() {
+        if (!dom.filterSticky || !dom.overview || !dom.windowBar) return;
+        var root = document.documentElement.style;
+        var band = dom.filterSticky.offsetHeight;
+        var overview = dom.overview.offsetHeight;
+        root.setProperty('--sticky-band-h', band + 'px');
+        root.setProperty('--sticky-overview-h', overview + 'px');
+        root.setProperty('--sticky-window-h', dom.windowBar.offsetHeight + 'px');
+      }
+
       function render() {
         renderBars();
         renderGeneration();
@@ -6123,6 +6170,10 @@ <h2 class="section-title" id="inspector-heading">检视</h2>
         renderInspector();
         renderFooter();
         updateHandoffButton();
+        // Last: it reads the bands the lines above just changed. The 异常 note appears and
+        // disappears here, so measuring before this point would pin the bands to yesterday's
+        // height and leave the window bar overlapping the strip above it.
+        syncStickyOffsets();
       }
 
       function renderGeneration() {
 
@@ -219,7 +219,57 @@ Two guards hold this: the exact call sequence, and that `render()` contains exac
 
 ---
 
-## 8. Say what you did not manage to check
+## 8. Measure the layout; never write it down
+
+A sticky offset is a measurement, not a constant. This file carried `top: 32px` and `top: 83px`
+with a comment explaining them, and both were correct only while the layout had exactly one filter
+row above the strip. Adding a second filter row — one that wraps at narrow widths and grows a line
+when its note appears — invalidated them silently: nothing overlaps at the width the comment was
+written at, and at 735px the window bar sits on top of the strip.
+
+```css
+/* wrong: correct for one layout, and there is no way to tell from reading it */
+.overview  { top: 32px; }
+.window-bar { top: 83px; }
+```
+
+```js
+function syncStickyOffsets() {
+  if (!dom.filterSticky || !dom.overview || !dom.windowBar) return;
+  var root = document.documentElement.style;
+  var band = dom.filterSticky.offsetHeight;
+  var overview = dom.overview.offsetHeight;
+  root.setProperty('--sticky-band-h', band + 'px');
+  root.setProperty('--sticky-overview-h', overview + 'px');
+  root.setProperty('--sticky-window-h', dom.windowBar.offsetHeight + 'px');
+}
+```
+
+`offsetHeight`, not `getBoundingClientRect().height`: this runs on every 2.5s poll, and a
+fractional device-pixel ratio would otherwise accumulate a third of a pixel per poll into an
+overlap nobody can see and everybody inherits.
+
+Three rules make that hold, and each of them needed a mutation before it was believed:
+
+- **Pin what answers one question as one block.** The two filter rows were pinned separately, which
+  allowed the state where the type row is still on screen and the 异常 row is not. Half a filter
+  visible is worse than none: the reader is told nothing is filtered.
+- **The set of variables the stylesheet reads must equal the set the measurement writes.** One
+  direction lets a fallback govern a band that nothing measures; the other lets the measurement
+  keep writing a variable nobody reads, which is what "the offsets went back to being constants"
+  looks like from the outside.
+- **Everything pinned over the ledger paints a background.** The rows underneath are
+  `background: transparent`, so a pinned band that stops covering them shows them through.
+
+And the general form of the third one: **an element named in markup and an element named in a
+selector are two names, and nothing checks that they agree.** A band looked up by an id the markup
+does not ship measures nothing, all three offsets fall back to their constants, and no error is
+raised anywhere. A guard that hands the function a stub cannot catch this — the stub supplies
+exactly what the function asked for — so the binding is checked against the markup instead.
+
+---
+
+## 9. Say what you did not manage to check
 
 Some things cannot be driven through the browser harness. The honest form is the useful one:
 
@@ -245,6 +295,9 @@ A harness limitation reported as a limitation is useful. The same limitation sil
 [ ] prose assertions are produced by running the function, numbers parsed back out of the string
 [ ] no assertion name contains a colon
 [ ] everything one reader can compare came from one snapshot
+[ ] no layout offset is written down; anything pinned is measured, and measured at the end of render
+[ ] every variable the stylesheet reads is written by the measurement, and one written is one read
+[ ] every element the measurement reads is bound to an id the markup actually ships
 [ ] anything not verified is written down as not verified
 ```
 
 
@@ -94,6 +94,7 @@ none of them, and the guards then fail for want of something to find.
 | `verify-prompt-fix.mjs` | injected-block splitting via `canonicalTextRange` |
 | `verify-window-widen.mjs` | `向上加载更早的 N 条` survives the next poll (self-contained: the shipped rule is transcribed) |
 | `verify-anomaly.mjs` | the four 异常 rules, their units, their position on the page, and their exclusivity with the other axis |
+| `verify-sticky.mjs` | both filter rows pin as one band, and every band below it is measured rather than written down |
 | `verify-block-order.mjs` | assistant blocks are ordered per message: thinking → text → toolCall |
 | `verify-response-rail.mjs` | every row carries the rail of the response it came from |
 | `verify-call-fold-coverage.mjs` | folding every tool call really removes them from the rendered build |
@@ -105,9 +106,10 @@ none of them, and the guards then fail for want of something to find.
 | `mutate-turn-breakdown.py` | the turn breakdown parts sum to the headline |
 | `mutate-compaction-turn.py` | the compaction-turn attribution is not a free-text label |
 | `mutate-pr15-fixes.py` | each of the six PR #15 fixes has a guard that can fail |
-| `mutate-anomaly.py` | each of the 50 异常 mutations, in the page, has a guard that can fail |
+| `mutate-anomaly.py` | each of the 70 异常 mutations, in the page, has a guard that can fail |
 | `mutate-steer.py` | the steer split and the turn classification hold |
 | `mutate-window-default.py` | the window defaults, labels, resets and load-more hold |
+| `mutate-sticky.py` | each of the 18 pinned-band rules has a guard that can fail |
 
 ## The 异常 axis, and what it promises
 
@@ -150,6 +152,41 @@ be a field of the search form any more, and the KPI cards must not come between
 to be the last field of the form, four cards above the axis it reads as part of — a layout defect
 that no functional assertion in the suite could see.
 
+## The pinned bands, and why none of them carries a number
+
+Four things pin themselves while the ledger scrolls under them: the filter band (both axes), the
+overview strip, the window bar, and the turn head. Three properties are asserted, and each one was
+false in this file's own history.
+
+1. **The two filter rows pin as one block.** They answer the same question — which records the
+   ledger shows — so a state where the type row is still pinned and the 异常 row has scrolled away
+   is the worst of both: the reader is told nothing is filtered while half the filter is off
+   screen. Only the type row ever claimed `position: sticky` and it did not survive; the guard
+   now asserts that neither row pins itself and that both are inside the band, because a band that
+   does not contain both is the same defect wearing a different hat.
+2. **No pinned offset is written down.** The band is two rows whose height changes when the 异常 row
+   wraps at narrow widths and when its note appears or disappears, so `0–32` and `32–83` were
+   correct only for a layout without the 异常 row. `syncStickyOffsets()` measures the three bands at
+   the end of `render()` and writes them into `--sticky-band-h`, `--sticky-overview-h` and
+   `--sticky-window-h`; the stylesheet only ever reads those. The guard runs that function against
+   stub heights of 137 / 65 / 57 — no constant in the file is any of them — so a version that
+   writes a number instead of a measurement fails, and so does one that measures the band for the
+   strip.
+3. **Anything pinned over the ledger has to paint a background.** The rows underneath are
+   `background: transparent`, so a pinned band that stops covering them shows them through. The
+   strip paints through `.overview-plot` rather than through the sticky section, which is why the
+   guard names that rule and not `.overview`.
+
+Two more assertions exist because the mutation control found them missing: the set of variables the
+stylesheet reads has to be **exactly** the set the measurement writes (one side alone lets a stale
+fallback or a dead measurement through), and every band the function reads has to be bound to an
+element the markup actually ships. A band that is never bound measures nothing, all three offsets
+fall back to their constants, and no error is reported anywhere — the stub the guard measures with
+cannot catch it, so the binding is checked against the markup instead.
+
+The order of `render()` is asserted too: the measurement has to be the last statement, because the
+异常 note is written by `syncAnomalyMode()` and that note is what changes the band's height.
+
 ## Deliberately not in the default run
 
 Shipping a suite that is red teaches reviewers to ignore it, so these are present but skipped,