diff --git a/examples/poly106_demo.py b/examples/poly106_demo.py
index bd4e3e7..dfa0da9 100644
--- a/examples/poly106_demo.py
+++ b/examples/poly106_demo.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """Play the Poly106: a pad progression, a bass line, brass stabs and a lead.
 
     python examples/poly106_demo.py            # writes wav_exports/poly106_demo.wav
diff --git a/examples/poly80_bench_mix.py b/examples/poly80_bench_mix.py
index 76f61d3..37352c4 100644
--- a/examples/poly80_bench_mix.py
+++ b/examples/poly80_bench_mix.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """The Poly80's cost with one channel heard: Init, eight notes, ``mix`` at channel I.
 
     python examples/poly80_bench_mix.py            # CPython
diff --git a/examples/poly_bench.py b/examples/poly_bench.py
index 1344782..18af6dc 100644
--- a/examples/poly_bench.py
+++ b/examples/poly_bench.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """How much of real time a polyphonic instrument takes.
 
     python examples/poly_bench.py [voice] [voices]       # CPython
diff --git a/examples/poly_render.py b/examples/poly_render.py
index 6901a72..af8784d 100644
--- a/examples/poly_render.py
+++ b/examples/poly_render.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """Render chords through the polyphonic adapter and check the steals for clicks.
 
     python examples/poly_render.py [voice]        # CPython
diff --git a/lib/audioinstruments/_rack/_lazy.py b/lib/audioinstruments/_rack/_lazy.py
index d964f8d..7690ace 100644
--- a/lib/audioinstruments/_rack/_lazy.py
+++ b/lib/audioinstruments/_rack/_lazy.py
@@ -3,7 +3,7 @@
 """Import a package's public names on first use.
 
 A package keeps a table of name -> submodule path and defines a module-level
-``__getattr__`` that calls :func:`load`. The first ``pkg.Name`` imports only the
+``__getattr__`` that calls `load`. The first ``pkg.Name`` imports only the
 submodule that holds ``Name`` and stores the result in the package's globals,
 so every later access is a plain attribute read.
 """
diff --git a/lib/audioinstruments/_rack/dsp/ensemble.py b/lib/audioinstruments/_rack/dsp/ensemble.py
index cb62cb7..3b05757 100644
--- a/lib/audioinstruments/_rack/dsp/ensemble.py
+++ b/lib/audioinstruments/_rack/dsp/ensemble.py
@@ -6,7 +6,7 @@
 each swept by the sum of a slow and a fast LFO, the three 120 degrees apart on
 both. Their sum in mono is steady (the three sweeps add to zero), while the
 left and right sides each take a different mix of the three and spread apart.
-It is a different animal from :class:`~audioinstruments._rack.dsp.chorus.BBDChorusCore`, which
+It is a different animal from `BBDChorusCore <audioinstruments._rack.dsp.chorus.BBDChorusCore>`, which
 is two lines swept in antiphase by one triangle with the dry signal always
 mixed in, so it sits beside it on the same node.
 
@@ -27,7 +27,7 @@
 
 Those are the defaults. The number of lines and everything about each one are
 constructor parameters, so a different ensemble is a different list of
-:class:`EnsembleLine` and a few keywords, not a different core: each line has
+`EnsembleLine` and a few keywords, not a different core: each line has
 its own slow rate, fast ratio, phase (a fraction of a period, on both sweeps;
 0.5 turns the sweep upside down), a level into the left and the right side,
 and a ``mix`` for its own node (2 is the wet alone, 1 the dry plus the wet).
@@ -186,7 +186,7 @@ def sweep_table(sweep, table_len=TABLE_LEN):
 class EnsembleCore:
     """Swept lines, wet only, and a bypass input that is as late as they are.
 
-    ``lines`` is a sequence of :class:`EnsembleLine` (the string ensemble's
+    ``lines`` is a sequence of `EnsembleLine` (the string ensemble's
     three by default). The other keywords are the delay every line sweeps
     around (``centre_ms``), the depths of the slow and fast sweeps in ms, the
     lines' low-pass (``tone_hz``), the length of a sweep table and the slow
diff --git a/lib/audioinstruments/_rack/dsp/organ.py b/lib/audioinstruments/_rack/dsp/organ.py
index 13183ad..87c3b6d 100644
--- a/lib/audioinstruments/_rack/dsp/organ.py
+++ b/lib/audioinstruments/_rack/dsp/organ.py
@@ -5,10 +5,10 @@
 An organ voice is a few pitches a key switches on together, each at a fixed
 interval from the key (a footage: 8' is the key's own pitch, 16' an octave
 under it, 4' an octave over it), each at a level a drawbar or a tab sets.
-:class:`DrawbarToneCore` sums those pitches into one single-cycle table at a
+`DrawbarToneCore` sums those pitches into one single-cycle table at a
 base pitch every footage is a whole harmonic of, and synthio plays the table.
 What each footage sounds like is the ``partial`` it is given: a tonewheel's is
-a near-sine (:func:`sine_partial`); a divide-down organ's is a square or a
+a near-sine (`sine_partial`); a divide-down organ's is a square or a
 pulse, which is one more function of the same form that sums a few odd sines.
 Which harmonic each footage plays on a given key (the foldback at the ends of
 the keyboard) belongs to the organ, so the voice hands the core both the
@@ -17,8 +17,8 @@
 A tonewheel organ has 91 wheels on one shaft, each in front of a pickup, so
 every pitch is a near-sine that is always turning; a key closes nine contacts
 that connect nine of them to the drawbars' busbars. The nine footages and the
-harmonic each takes of the 16' pitch are in :data:`TONEWHEEL_FOOTAGES`;
-:func:`tonewheel_harmonics` folds them at the keyboard's ends, as the organ
+harmonic each takes of the 16' pitch are in `TONEWHEEL_FOOTAGES`;
+`tonewheel_harmonics` folds them at the keyboard's ends, as the organ
 does because it has no wheels past its top and bottom.
 """
 
@@ -117,14 +117,14 @@ class DrawbarToneCore:
 
     ``harmonics`` lists every harmonic of the base pitch a footage may be asked
     to play (their partials are fetched once, here); ``partial(h)`` gives the
-    table of one footage at harmonic ``h`` (:func:`sine_partial` by default);
+    table of one footage at harmonic ``h`` (`sine_partial` by default);
     ``base_ratio`` is the table's pitch against the key's (0.5: the table is at
     16', so the 8' is its second harmonic); ``level`` scales the sum;
     ``table_len`` is the table's length, which every partial shares (synthio
     plays the table without interpolating, so a longer one sounds cleaner).
 
-    :meth:`set_levels` rebuilds the table when a level or a harmonic changes,
-    in buffers the core owns. :meth:`render` plays it through a
+    `set_levels` rebuilds the table when a level or a harmonic changes,
+    in buffers the core owns. `render` plays it through a
     ``ChunkedBridge``, so the tone trails the controls by 256 frames. The
     contacts (a 3 ms ramp each way) follow the gate 256 frames late too, which
     is when the tone at the key's pitch arrives, so a note never begins on the
diff --git a/lib/audioinstruments/_rack/modules/combo_organ/voice.py b/lib/audioinstruments/_rack/modules/combo_organ/voice.py
index 6e4c2a3..7dd308b 100644
--- a/lib/audioinstruments/_rack/modules/combo_organ/voice.py
+++ b/lib/audioinstruments/_rack/modules/combo_organ/voice.py
@@ -148,7 +148,7 @@ def _band_hz(note):
 
 
 def _rolloff(hz):
-    """The divider's level at ``hz`` (48 kHz), from :data:`ROLLOFF`."""
+    """The divider's level at ``hz`` (48 kHz), from `ROLLOFF`."""
     x = hz / 1000.0
     i = int(x)
     if i >= len(ROLLOFF) - 1:
@@ -202,7 +202,7 @@ def __init__(self):
         self._sine = None
 
     def partial(self, key):
-        """The partial ``DrawbarToneCore`` sums for a footage key (see :func:`footage_key`)."""
+        """The partial ``DrawbarToneCore`` sums for a footage key (see `footage_key`)."""
         table = self.squares.get(key)
         if table is None:
             table = self._square(key[0], key[1], key[2])
@@ -232,7 +232,7 @@ def _rank_table(self, count):
 
     def _square(self, h, count, f1):
         """An int16 table holding ``h`` cycles of a square of ``count`` odd harmonics,
-        :data:`PARTIAL_UNIT` counts high, each harmonic rolled off as a divider's
+        `PARTIAL_UNIT` counts high, each harmonic rolled off as a divider's
         square at ``f1`` Hz would be."""
         n = TONE_LEN
         sine = self._sine
diff --git a/lib/audioinstruments/_rack/sequencer.py b/lib/audioinstruments/_rack/sequencer.py
index edd2c1c..3690c25 100644
--- a/lib/audioinstruments/_rack/sequencer.py
+++ b/lib/audioinstruments/_rack/sequencer.py
@@ -165,7 +165,7 @@ def retrack(self, instrument, index=0, pattern=None):
         For a player changing kit mid-bar. The steps already on the queue
         belong to the instrument that is leaving, so they are taken back
         here rather than left to sound on a machine the dropdown says is
-        gone; :meth:`relay` lays them again on the new one.
+        gone; `relay` lays them again on the new one.
         """
         if not instrument.schedulable:
             raise ValueError("%r cannot be scheduled" % instrument)
diff --git a/tests/support/fake_pump.py b/tests/support/fake_pump.py
index b02f0f0..3de28a8 100644
--- a/tests/support/fake_pump.py
+++ b/tests/support/fake_pump.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """A stand-in for ``audiopump.Events`` and ``audiopump.now`` for the sequencer.
 
 A `Sequencer` writes to its queue through two calls, ``q.at(frame, op,
@@ -9,7 +13,7 @@
   wrapping comparison, hands back a token counting from 1, and answers **0**
   rather than raising when the queue is full;
 * ``cancel()`` is True only while the event is still pending;
-* :meth:`Events.apply` applies every event inside ``[now, now + block)`` in
+* `Events.apply` applies every event inside ``[now, now + block)`` in
   frame order, and counts late ones;
 * ``stats()`` returns the numbers in the same order.
 
diff --git a/tests/support/sf2_builder.py b/tests/support/sf2_builder.py
index a395bc4..b1ab174 100644
--- a/tests/support/sf2_builder.py
+++ b/tests/support/sf2_builder.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """Write a SoundFont 2 bank from scratch, for the `_sf2` reader's tests.
 
 Nothing here reads a `.sf2`; it only writes one, chunk by chunk, from the
diff --git a/tests/support/sf2_play.py b/tests/support/sf2_play.py
index 1c655a4..7e75df4 100644
--- a/tests/support/sf2_play.py
+++ b/tests/support/sf2_play.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """Render a SoundFont instrument and measure what came out, for the
 `soundfont` tests. No numpy: these run wherever audiodsp builds."""
 
diff --git a/tests/support/sf2_smoke.py b/tests/support/sf2_smoke.py
index 85bca0d..ca02c0e 100644
--- a/tests/support/sf2_smoke.py
+++ b/tests/support/sf2_smoke.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """Build the tiny bank, read it back, check it: the one check that runs on
 every interpreter.
 
diff --git a/tests/test_midi_dispatch.py b/tests/test_midi_dispatch.py
index 8c3accc..a8a2b65 100644
--- a/tests/test_midi_dispatch.py
+++ b/tests/test_midi_dispatch.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """`audioinstruments._rack.midi_dispatch`: MIDI in, instrument methods out.
 
 A recording instrument checks what each kind of input becomes: bytes with
diff --git a/tests/test_sequencer.py b/tests/test_sequencer.py
index d1d1f15..3663df2 100644
--- a/tests/test_sequencer.py
+++ b/tests/test_sequencer.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """`audioinstruments._rack.sequencer.Sequencer`, driven by a clock a test moves.
 
 The sequencer's whole claim is that the groove does not depend on when
diff --git a/tests/test_sf2.py b/tests/test_sf2.py
index ef862af..7744556 100644
--- a/tests/test_sf2.py
+++ b/tests/test_sf2.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """`_sf2`: a SoundFont 2 bank read back exactly.
 
 The bank under test is written by `tests/support/sf2_builder.py` from the
diff --git a/tests/test_soundfont.py b/tests/test_soundfont.py
index 86f5268..25055d6 100644
--- a/tests/test_soundfont.py
+++ b/tests/test_soundfont.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """`soundfont` and `soundfont_drums` on the bank `sf2_builder` writes.
 
 The plumbing, without a real bank: how the bank is found, the patch list
diff --git a/tests/test_soundfont_gm.py b/tests/test_soundfont_gm.py
index 744a2d9..70bc4b7 100644
--- a/tests/test_soundfont_gm.py
+++ b/tests/test_soundfont_gm.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """The General MIDI gate: a real bank plays every program and the kit.
 
 Against FluidR3_GM.sf2 (MIT), looked for at `$FLUIDR3_GM`, then
diff --git a/tests/test_soundfont_heap.py b/tests/test_soundfont_heap.py
index c5c8b26..2d89b18 100644
--- a/tests/test_soundfont_heap.py
+++ b/tests/test_soundfont_heap.py
@@ -1,3 +1,7 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+
 """A second SoundFont instrument after a big bank has loaded.
 
 Resolving every preset when a bank opened left a MicroPython heap in pieces