feat(opal26): all four open cuts settled by ear — and the one nobody had counted

PLN made the four calls in the boundary lab. Merged into `verified`, so the
ear-boundary file now covers #1 and #3-#13 and #15; `unjudged` is empty. The raw
export is committed next to it as provenance rather than being consumed and
thrown away.

  #11  3557.33   takeover was +0.13s  -> the machine had it
  #12  3680.10   takeover was -8.00s  -> "here the right cut is mid :)"
  #13  3887.25   ear-only, no candidate could exist
  #15  4547.57   ear-only, no candidate could exist

Worth recording that #11 and #12 disagree about WHICH fraction is right: on one
the takeover was the cut, on the other the midpoint was, 8 s earlier. That kills
any remaining hope of a single universal fraction and confirms the tool's shape
was the right call — offer the candidates, let the ear pick, do not average.

The unnoticed gap: #13's END. The set is butt-joined, so that edge is the
boundary INTO Desire (nominal 4180.0), which was never verified because Desire
itself is cut from the release — it looked like work that no longer mattered.
It is the opposite: with Desire dropped, that edge is where the last official
track ends, and if it is late then Desire's intro bleeds into the end of the
record. Rendered it as a fifth boundary; the builder now reads
`cut_from_release` and labels such an edge "END of #13 Vague de CRIME" with an
on-screen explanation, so the next gig cannot lose the same edge the same way.

Also made bounds-smoke pick its subject by content instead of position: it broke
the moment #11 became settled and lost its candidate buttons, which is a data
change, not a regression. It now prefers a boundary that still has a machine
candidate and falls back to the nominal marker every boundary carries. 11/11
against the production server.

armada/tide-table/build_bounds_set.py

@@ -113,6 +113,11 @@ def main() -> int:
 
         v = verified.get(str(tno), {})
         c = cands.get(str(tno), {})
+        # A boundary into a track that is CUT from the release is not a start at
+        # all — it is the previous track's END, and it still has to be right or
+        # the dropped track's intro bleeds into the last one that ships. Label it
+        # as what it is, or it reads as work that no longer matters.
+        cut = ear.get("cut_from_release", {}).get(str(tno))
         marks = []
         # order matters: this is the audition order, best-first. `takeover` came
         # within 0.7 s on both boundaries that also had ear answers — n=2, on the
@@ -126,9 +131,12 @@ def main() -> int:
         if v.get("alt_start") is not None:
             marks.append({"key": "alt", "label": "ear alt", "t": float(v["alt_start"])})
 
+        prev_title = clean_title(by_track.get(tno - 1, {}).get("title", "?"))
         out_rows.append({
             "track": tno,
             "title": clean_title(nxt["title"]),
+            "cutFromRelease": bool(cut),
+            "role": (f"END of #{tno - 1} {prev_title}" if cut else "start"),
             "fromTitle": clean_title(by_track.get(tno - 1, {}).get("title", "?")),
             "nominal": nominal,
             "clipStart": t0,
@@ -138,7 +146,7 @@ def main() -> int:
             "marks": marks,
             "settled": v.get("start") is not None,
             "noCandidate": not marks or all(m["key"] in ("ear", "alt") for m in marks),
-            "note": c.get("note") or v.get("note") or "",
+            "note": (cut.get("note", "") + " " if cut else "") + (c.get("note") or v.get("note") or ""),
         })
 
     doc = {

armada/tide-table/judge_specs/opal26_boundaries_ear.json

@@ -28,7 +28,7 @@
     },
     "7": {
       "start": 2248.8,
-      "note": "first 21s are still Take Five Drops — the xfade"
+      "note": "first 21s are still Take Five Drops \u2014 the xfade"
     },
     "9": {
       "start": 2954.4,
@@ -42,11 +42,11 @@
       "start": 2726.4,
       "note": "piment until 0:02, xfade starts, but unfortunate MIDI events make a clean single start only at 0:06.4; PLN also floats xfading it ourselves from 0:10.9 in postprod",
       "alt_start": 2730.9,
-      "alt_note": "0:10.9 — only if we craft our own crossfade"
+      "alt_note": "0:10.9 \u2014 only if we craft our own crossfade"
     },
     "1": {
       "start": 1.45,
-      "note": "GO, but 'trim leading silence tho' (PLN). The amount is now MEASURED, not detected: the head is literal digital silence — every sample is exactly zero up to sample 64038 (= 1.452109 s @ 44.1 kHz), and the first audible frame peaks at -28 dBFS. Cutting at 1.45 keeps a ~2 ms guard before the first transient so the attack cannot be clipped.",
+      "note": "GO, but 'trim leading silence tho' (PLN). The amount is now MEASURED, not detected: the head is literal digital silence \u2014 every sample is exactly zero up to sample 64038 (= 1.452109 s @ 44.1 kHz), and the first audible frame peaks at -28 dBFS. Cutting at 1.45 keeps a ~2 ms guard before the first transient so the attack cannot be clipped.",
       "measured": {
         "first_nonzero_sample": 64038,
         "first_nonzero_s": 1.452109,
@@ -55,54 +55,25 @@
         "method": "sample-exact scan; NOT a threshold and NOT a detector estimate",
         "date": "2026-08-16"
       }
-    }
-  },
-  "unjudged": [
-    11,
-    12,
-    13,
-    15
-  ],
-  "_candidates_comment": [
-    "UNCONFIRMED machine proposals, kept separate from `verified` on purpose —",
-    "nothing here is a boundary until PLN's ear says so. From cut_lens3 run 2",
-    "(cut_candidates_v3run2.json). On the two boundaries that also have ear",
-    "answers, u0.85/`takeover` landed within 0.7 s (#8 2726.25 vs 2726.4;",
-    "#10 3268.3 vs 3269.0) — encouraging, but that is n=2 on the cases that were",
-    "already easy, so it is a reason to audition `takeover` FIRST, not to trust it.",
-    "#13 and #15 are absent by exhaustion: zero voting orbits across three gates",
-    "and two estimators. Those two are ear-only; see cut_lens3.py's docstring."
-  ],
-  "candidates": {
+    },
     "11": {
-      "title": "Ghosts in the Toilets",
-      "nominal": 3560.2,
-      "takeover": 3557.2,
-      "onset": 3515.7,
-      "mid": 3554.2,
-      "voters": 3,
-      "note": "all five fractions inside 3.5 s => SHORT crossfade; nominal looks ~3 s late"
+      "start": 3557.33,
+      "note": "takeover is the right cut time here. \u2014 ear call via boundary lab 2026-08-16 (source: playhead); machine takeover 3557.2 was +0.13s"
     },
     "12": {
-      "title": "Mafia",
-      "nominal": 3681.6,
-      "takeover": 3688.1,
-      "onset": 3667.1,
-      "mid": 3680.1,
-      "voters": 3,
-      "note": "span 21 s => LONG crossfade; nominal sits mid-span, so the takeover is later than the current cut"
+      "start": 3680.1,
+      "note": "here the right cut is mid :) \u2014 ear call via boundary lab 2026-08-16 (source: mid); machine takeover 3688.1 was -8.00s"
     },
     "13": {
-      "title": "Vague de CRIME",
-      "nominal": 3889.1,
-      "note": "NO CANDIDATE — timbral lens structurally blind here (shared sounds across the switch). Ear-only."
+      "start": 3887.25,
+      "note": "ear call via boundary lab 2026-08-16 (source: playhead)"
     },
     "15": {
-      "title": "REVOLUTION",
-      "nominal": 4547.4,
-      "note": "NO CANDIDATE — same as #13. Ear-only."
+      "start": 4547.57,
+      "note": "ear call via boundary lab 2026-08-16 (source: playhead)"
     }
   },
+  "unjudged": [],
   "cut_from_release": {
     "14": {
       "title": "Desire",

armada/tide-table/judge_specs/opal26_ear_calls_2026-08-16.json

+{
+ "gig": "opal-festival-2026",
+ "_comment": [
+  "Ear calls from the boundary lab. Times are absolute MASTER seconds.",
+  "These are PLN's answers — merge them into the `verified` block of",
+  "judge_specs/<gig>_boundaries_ear.json, which is the ground truth."
+ ],
+ "decided": {
+  "11": {
+   "start": 3557.33,
+   "source": "playhead",
+   "note": "takeover is the right cut time here."
+  },
+  "12": {
+   "start": 3680.1,
+   "source": "mid",
+   "note": "here the right cut is mid :)"
+  },
+  "13": {
+   "start": 3887.25,
+   "source": "playhead",
+   "note": ""
+  },
+  "15": {
+   "start": 4547.57,
+   "source": "playhead",
+   "note": ""
+  }
+ },
+ "skipped": []
+}
\ No newline at end of file

armada/ui/scripts/bounds-smoke.mjs

@@ -54,7 +54,15 @@ ok('waveform canvas rendered')
 // 3 — the clip is the RIGHT clip. Two independent checks, because either alone
 //     lies: the mount could 404 (caught by the fetch) or fall through to the
 //     wrong directory and serve *a* file (caught by the displayed duration).
-const first = doc.boundaries[0]
+// Pick the subject by CONTENT, not by position: as boundaries get settled their
+// candidate buttons disappear, and a test pinned to boundaries[0] starts failing
+// on data changes rather than on regressions. Prefer one that still has a
+// machine candidate; fall back to the nominal marker every boundary carries.
+const firstIdx = Math.max(0, doc.boundaries.findIndex((x) => x.marks.length > 0))
+const first = doc.boundaries[firstIdx]
+const probe = first.marks[0] ?? { key: 'nominal', label: 'nominal', t: first.nominal }
+await p.$$eval('nav button', (n, i) => n[i].click(), firstIdx)
+await p.waitForTimeout(600)
 const head = await p.request.get(new global.URL(first.url, PAGE).href)
 const bytes = Number(head.headers()['content-length'] ?? 0)
 head.ok() && bytes > 500_000
@@ -70,23 +78,23 @@ Math.abs(shownS - want) <= 1
   : fail(`transport shows ${shownDur} (${shownS}s), expected ${want}s`)
 
 // 4 — markers are drawn, one per candidate plus the nominal
-const wantMarks = first.marks.length + 1
+const wantMarks = first.marks.length + 1  // candidates/ear + the nominal
 const marks = await p.$$eval('[part="region-content"], .wavesurfer-region', (n) => n.length)
 marks >= wantMarks
   ? ok(`${marks} markers drawn (>= ${wantMarks} expected)`)
   : fail(`expected >= ${wantMarks} markers, saw ${marks}`)
 
 // 5 — clicking a candidate button records a call IN MASTER TIME, inside the clip
-await p.getByRole('button', { name: /^takeover · / }).first().click()
+await p.getByRole('button', { name: new RegExp('^' + probe.label + ' · ') }).first().click()
 await p.waitForTimeout(400)
 const callTxt = await p.textContent('section:has(textarea) .tnum')
 const secs = Number(/\(([\d.]+)s/.exec(callTxt ?? '')?.[1])
 Number.isFinite(secs) && secs >= first.clipStart && secs <= first.clipEnd
   ? ok(`call recorded at ${secs}s master (inside [${first.clipStart}, ${first.clipEnd}])`)
   : fail(`call "${callTxt}" is not a master time inside this clip`)
-Math.abs(secs - first.marks.find((m) => m.key === 'takeover').t) < 0.01
-  ? ok('call equals the takeover candidate exactly — no clip/master drift')
-  : fail(`call ${secs} != takeover ${first.marks.find((m) => m.key === 'takeover').t}`)
+Math.abs(secs - probe.t) < 0.01
+  ? ok(`call equals the ${probe.label} marker exactly — no clip/master drift`)
+  : fail(`call ${secs} != ${probe.label} ${probe.t}`)
 
 // 6 — audio actually PLAYS. The media element is owned by wavesurfer and never
 //     enters the DOM, so querySelector('audio') is null by construction — that

armada/ui/src/bounds/BoundaryLab.tsx

@@ -31,6 +31,8 @@ interface Boundary {
   marks: Mark[]
   settled: boolean
   noCandidate: boolean
+  cutFromRelease?: boolean
+  role?: string
   note: string
 }
 interface BoundsDoc { gig: string; title: string; pad: number; boundaries: Boundary[] }
@@ -173,7 +175,17 @@ export function BoundaryLab({ gig }: { gig: string }) {
         <p className="font-mono text-xs text-ink-faint">
           {b.fromTitle} <span className="text-magenta">→</span> {b.title}
         </p>
-        <h2 className="text-xl font-semibold">#{b.track} {b.title}</h2>
+        <h2 className="text-xl font-semibold">
+          #{b.track} {b.title}
+          {b.settled && <span className="ml-2 align-middle text-xs font-normal text-emerald-300">settled</span>}
+        </h2>
+        {b.cutFromRelease && (
+          <p className="mt-2 rounded border border-sky-500/30 bg-sky-500/10 px-3 py-2 text-xs text-sky-200">
+            <strong>{b.role}.</strong> {b.title} is cut from the release, so this edge is not a
+            track start — it is where the last official track <em>ends</em>. It still has to be
+            right, or {b.title}&rsquo;s intro bleeds into the end of the record.
+          </p>
+        )}
         {b.noCandidate && (
           <p className="mt-2 flex items-start gap-2 rounded border border-amber-500/30 bg-amber-500/10
                         px-3 py-2 text-xs text-amber-200">