docs(last-third): the closing three, read cold before the last monitored run

The KRKs go home after tomorrow's dawn rehearsal, so this is the last run with
real monitors — and desire / REVOLUTION / electric_hammer have not been played
this cycle at all. Read them cold so his ears go to the music.

Two findings worth the read, both the same shape as Sunday's do_it_right bug:

desire declares d4 THREE times in one block (lines 50, 56, 62), under his own
"-- FIXME BASS!!!". Later wins, so the two takes he tuned are dead code. The
winner differs in ways that do not read as a choice: it has no lpf on ^32, so
that knob is dead on the bass, and it carries no gain at all where the other two
say 1.15 — which under SuperDirt's quartic law (amp = gain^4) is 4.8 dB quieter
than the level he dialled.

electric_hammer declares d5 twice (32, 46). The superfreak voice wins, so ^89
and ^33 have nothing behind them, ^57 does something other than what the dead
copy above it describes, and hammer:7 — the sample the closing track is named
after — never sounds.

Also: the five closing seams with their ghost orbits, paste-ready orphan guards
for the two worst, his two vague_de_crime TODOs with the gain converted to dB,
and the gate result (GO, 12/12).

One methodology note recorded in the doc: silent-eval must be run --seeded here.
Unseeded it reports 12 SILENT orbits across these three tracks, every one of them
a midiOn gate sitting at default-off. That is the harness, not the music.

LAST_THIRD.md

+# The last third — cheat sheet for the dawn run
+
+Prepared cold on 2026-08-03, **no edits made to your tracks**. Everything below is
+a read: what the code actually does, so your ears can go to the music instead of
+the debugging. The three calls at the end are yours.
+
+Set tail, in order:
+
+    11. you_my_sunshine  [166]
+    12. vague_de_crime   [120]
+    13. desire           [129]
+    14. the_revolution_will_be_sampled
+    15. electric_hammer      ← closes the show
+
+---
+
+## 1. Three seams, three different problems
+
+    Mafia    -> Sunshine     GHOST  d7 (acidOto3091)
+    Sunshine -> Vague        GHOST  d9, d11 (both no_sunshine)
+    Vague    -> Desire       clean  ✓
+    Desire   -> REVOLUTION   GHOST  d6 (aeolian, superzow)
+    REVOLUTION -> Hammer     GHOST  d7 (FMRhodes1), d8 (jungle_breaks), d12 (like_sugar)
+
+**The last one is the worst seam in the set** — three orbits from REVOLUTION keep
+playing under the closer, and REVOLUTION is the single biggest ghost-producer in
+the whole 15 (3 orphans, tied with gimme_acid).
+
+Under pressure: `hush`. Properly, paste at the TOP of the incoming file as its own
+block — `dN` is `xfade N`, so these *fade* the ghost out rather than cutting it:
+
+```haskell
+-- top of the_revolution_will_be_sampled.tidal
+d6 $ silence
+```
+
+```haskell
+-- top of electric_hammer.tidal
+d6 $ silence
+d7 $ silence
+d8 $ silence
+d12 $ silence
+```
+
+Listen for it first, though — d7 FMRhodes1 bleeding into the hammer may be the
+transition rather than the bug. That is an ear call, and you have the monitors
+tonight and not after.
+
+## 2. Dead code in BOTH closers — buttons that cannot fire
+
+`pvlint` flags 3 warnings across the tail, all PV005 (same orbit declared twice
+in one block). **The last declaration silently wins; everything above it is dead.**
+This is the same failure mode as `do_it_right`'s ^43/^44 on Sunday.
+
+### desire — d4 is declared THREE times (lines 50, 56, 62)
+
+Three bass takes side by side, under your own `-- FIXME BASS!!!`. What actually
+plays is the third one, **"Claude's Gesture"** — and it differs from the two you
+tuned in ways you would not hear as a *choice*:
+
+| | line 50 "the FIXME" | line 56 "Weight" | **line 62 "Gesture" ← WINS** |
+|---|---|---|---|
+| `# lpf (range 300 1800 "^32")` | yes | yes | **absent — knob ^32 is dead** |
+| gain | `# gain 1.15` | `\|* gain (… * 1.15)` | **none → default 1** |
+| crush | `midiOn "^45"` | `midiOn "^45"` | `midiOn "^45"` + `crushbus 41 ^52` |
+
+Two consequences worth knowing before you play it: **^32 does nothing on the
+bass**, and the winning take runs at gain 1 where the other two were set to 1.15
+— which in SuperDirt's quartic law (`amp = gain⁴`) is **4.8 dB quieter** than the
+level you dialled in. If desire's bass feels thin at dawn, that is why, and it is
+one number, not a mix problem.
+
+### electric_hammer — d5 declared twice (lines 32, 46)
+
+Line 32 is the `hammer:7` voice with **three** buttons on it — `^89` (ply 8),
+`^33` (n 3), `^57` (ply 4 + begin/end slice). Line 46 is a `superfreak` voice with
+one button (`^57`). **Line 46 wins**, so on the closing track:
+
+- `^89` and `^33` are dead — nothing behind them at all.
+- `^57` still works, but does something completely different from what the dead
+  copy above it says it does.
+- The `hammer:7` sample — the one the track is named after — never sounds.
+
+Both are A/B variants you left side by side, which is a fine way to write. They
+just need one of them commented out before the room hears it.
+
+## 3. vague_de_crime carries your two TODOs
+
+Committed tonight in `f9ec204`, in your words:
+
+- d1 — *"Vraiment solide? on peut le comparer au reste du set?"* You raised it
+  to `gain 1.3` already. Quartic: that is **+3.4 dB** over `gain 1`, so it is a
+  real move, not a nudge.
+- d3 — *"Less solid than in do it right or drops or ouais maybe? to be
+  improved/replaced"*
+
+Both are ear-work for this run. Not touched.
+
+## 4. What is already proven, so you can skip checking it
+
+`tools/gig-up.sh` ran cold tonight: **GO, 12/12**. Boot helpers, exec bits,
+setlist-vs-backlog (15 tracks), surface grid (0 drift), fader baseline (matches
+the 2026-08-02 capture), all 12 orbits reach master, all 15 tracks compile
+seeded, mute map, pvlint, preload, LCXL present.
+
+One caveat the gate prints itself: it proves the set *boots* correct, not that
+the *mixer* is correct — `check-mix` reads the saved session. Ctrl+S in Ardour
+before trusting a green fader line.
+
+> Note on `silent-eval`: run it **`--seeded`**. Unseeded it reports 12 "SILENT"
+> orbits across these three tracks, all of which are just `midiOn` gates sitting
+> at their default-off — a property of the harness, not of your music.