Cubase-MCP-GAJ3T.MusicGen.R.../prompts/report_7_fix.md

443 lines
26 KiB
Markdown

```
CURRENT EXECUTABLE IMPLEMENTATION — REAL NATIVE RIFFER CODECS
Verification date: 2026-10-19
Status: IMPLEMENTED AND VERIFIED; native plugin/DAW acceptance remains UNRUN.
This section supersedes the unavailable-layout conclusion in the labeled
historical research below. Riffer implementation is NOT evidence-blocked.
SCOPE AND FACTS
Read prompts/step_7_fix.md amendment, latest prompts/plan.md continuation,
current profiles/models/compiler/service/native boundaries/tests and docs.
Preserved the pre-existing plan/prompt amendments; no commits, delegation,
later prompts, service native registration or SD3 implementation.
Read-only local verification:
- AME.vst3 ProductVersion 4.0.1; AMR.vst3 ProductVersion 4.1.0.
- 35 target AME Presets .griff and 30 target AMR Preset .briff files.
Root Riff, Ver=2.1, PPQBase=480, six/five strings; AME Inst=AME.
Target factory bass omits Inst; a direct user-authored bass file confirms
Inst=AMR, Ver=2.1, PPQBase=480, numStr=5.
- Verified integer NoteID/Fret/Begin/End/Velocity/Off-Velocity, articulation,
No Legato/False states, descending physical IDs, native metadata/key syntax,
CC1 raw MIDI-unit stair points and the existence of separate drop tunings.
- Safe provenance IDs: LOCAL-AME-401-RIFF21-20261019 and
LOCAL-AMR-410-RIFF21-20261019. No private absolute paths in packaged metadata.
- Public Main_Panel_Manual-AME.pdf and Main_Panel_Manual-AMR.pdf were decoded
in memory for technical facts. No manual assets, plugin binaries, factory
musical sequences or private templates were copied into the repository or
uploaded. Tests need no installed plugin, local factory file or network.
IMPORTANT EVIDENCE-DRIVEN CORRECTIONS
The current Eclipse manual specifies C0 sustain velocity bands: 1-15 full mute,
16-31 three-quarter mute, 32-63 half mute, 64-126 sustain, 127 Pop/PH depending
on Settings. Consequently guitar NORMAL is 64-126, NOT a misleading 1-127
normal sweep. Guitar palm mute uses D0/26 with lower velocity increasing depth;
no invented mute sublayers are declared.
Bass uses C0/24 for both named techniques: 64-126 normal, 32-63 palm mute.
Native bass C0-band palm mute uses PMP, NOT the separate D0 high-velocity PM
representation. Read-only target facts distinguish 313 PMP notes at 40-60,
56 PM notes at 70-100, 89 DN notes at 30-60, and Acc at 127. The initial native
PM/C0-band conflation was corrected before final acceptance, and the original
bass fixture regenerated through the actual writer. No D0 threshold or extra
named bass mode is fabricated. Audible acceptance remains explicitly unrun.
IMPLEMENTED
1. Strict nondefault production subset profiles:
ample.ame.4.0.1.riffer.v1 / ample-metal-eclipse-4.0.1-riffer-v1.json
ample.amr.4.1.0.riffer.v1 / ample-metal-ray5-4.1.0-riffer-v1.json
Explicit tuning by descending ID: guitar 36/43/48/53/57/62, bass
35/40/45/50/55; conservative frets 0-22 / 0-19. Explicit actual extension,
version/PPQ, native articulation map and supported raw CC1 representation.
verified_subset provenance separates vendor/local facts from generator
routing/switch timing/interpolation policy; no test-profile promotion.
2. formats/agp.py now supplies real native APIs:
check_riffer_capability / require_riffer_capability
write_riffer(path, plan, part, profile)
validate_riffer_file(path, plan, part, profile) -> RifferReadback
Legacy check_agp_capability / require_agp_capability remain compatible;
write_agp / validate_agp_file alias the real codec and require real suffixes.
Capability checks remain no-IO. Metadata alone cannot invent a codec.
3. Real UTF-8 Riff XML 2.1 / PPQ 480 writers, actual target metadata,
exact rational score-relative timing, preserved musical on/off velocities,
guitar Sus/PM and bass Sus/PMP. Verified switches are accounted for by
independent selected-score recompile, then represented as native technique
fields rather than stray musical notes.
Bounded deterministic physical-string backtracking handles valid chords,
including a regression requiring backtracking instead of greedy routing.
No same-string overlaps, note stealing, truncation, rounding or dropping.
4. Saved-file validation reads actual bounded XML; it does NOT serialize an
expected XML tree and compare serializer output to itself. It checks root,
target/version/PPQ/tempo/meter/measures/key/loop, notes and legal fret/pitch
arithmetic, ordering/physical overlap, exact requested velocities/techniques,
all discrete CC1 points, and exact score/event-plan correspondence.
Wrong suffix/version/target, malformed/extra content, DTD/entities,
unsupported CCs/techniques, bad grid and corrupted saved output fail.
Writer uses exclusive creation of the supplied staged path only, with
flush/fsync and read-back; no parent/template/sibling/publication is created.
5. Compiled CC1 stair points preserve integer tick/value in raw MIDI 0-127,
with ccName-1=1.Mod / ccVisibility-1=1. No fabricated resets, normalization,
modulation layer bands or named vibrato mode. Original fixtures explicitly
request their CC points; zeros in those fixtures are NOT injected resets.
6. Honest timing boundary: nonzero compiled pre-roll or any control tail is
explicitly rejected, not shifted, clipped or hidden. Current subset MIDI
policy uses zero lead, same-tick switch-before-note ordering, velocity-100
switches, and a 1/32-quarter pulse. These are generator policy, not vendor
host/timing guarantees. Impossible routing/search exhaustion also errors.
7. Strict request/artifact/profile/model/JSON Schema contracts:
requests retain agp/sdp aliases and accept actual griff/briff/sd3p names.
Artifact format and path suffix must agree and use actual suffixes only.
Profile summaries expose actual target suffixes. Corrected standard and
synthetic fixture metadata without making them native-capable.
Registry now contains five packaged resources, retaining the same three
standard defaults. Minimal service association changes prevent actual
suffix metadata from bypassing native preflight; NO native writer is wired.
Explicit target native requests hit native_writer_not_registered as before;
default MIDI/fallback and explicit mapped MIDI remain operational.
8. Updated README, architecture, profile facts/policy, compatibility and native
format docs, and directly coupled contract/resource/stdio regression tests.
SD3 metadata uses sd3p truthfully; SDP capability remains unavailable.
ACTUAL GENERATED ORIGINAL READ-BACK EVIDENCE (PERSISTENT)
Checked-in, newly generated minimal native fixtures:
tests\fixtures\riffer\original-lead.griff
tests\fixtures\riffer\original-bass.briff
Retained actual generated outputs:
.test-artifacts\native-riffer-evidence-v1\original-lead.griff
.test-artifacts\native-riffer-evidence-v1\original-bass.briff
Each saved native file independently validates three original notes, PPQ 480,
123.5 BPM, 4/4, one measure / end tick 1920, native key Cm.
Guitar pitches 36/84 at 0..480 and PM 38 at 960..1440.
Bass pitches 35/74 at 0..480 and PMP 37 at 960..1440.
Normal velocities 64/126; guitar PM 64, bass PMP 48; all musical off velocities
45, explicitly supplied in the event plan. CC1 points (0,0), (480,120),
(1440,0), explicitly requested by the original verification plan.
Fixture bytes and retained generated bytes are independently checked identical.
Lead: 1055 bytes, SHA-256
cfb4d60b58c2ca11d66b97bb71384d674cb56f63ff665cce066b44ae08e5204a
Bass: 1036 bytes, SHA-256
8dec4f241ff4473c30d86365ac259c6c9f361ccd5631b25c894c4f35f51a1238
Installed-wheel acceptance additionally generates and independently reads real
original .griff/.briff files from the non-editable offline installed wheel
while its cwd is outside the checkout. It verifies both target resources,
one native note per target, the correct PM/PMP label and raw-unit CC1.
That bounded test workspace is cleaned; the persistent originals above remain.
FINAL VALIDATION — ALL PASS, NO SKIPS
Interpreter: .\.venv\Scripts\python.exe (CPython 3.13.13, Windows).
Exact focused command:
.\.venv\Scripts\python.exe -m pytest tests\unit\test_riffer.py tests\unit\test_native_contracts.py tests\unit\test_profiles.py tests\unit\test_schemas.py tests\unit\test_agp.py tests\integration\test_contracts.py tests\integration\test_riffer_files.py -q --tb=short
PASS: 571 tests, 0 failures.
.\.venv\Scripts\python.exe -m pytest -q --tb=short
PASS: 1221 tests, 0 failures, 0 skips (40.76 seconds).
Includes wheel/sdist builds, installed and ZIP resources, isolated installed
native read-back, real installed stdio/default MIDI/fallback, and MIDI
read-back/publication regressions. No dependency manifests were changed and
no project dependencies installed/restored.
.\.venv\Scripts\python.exe -m ruff check src tests
PASS: all checks passed.
.\.venv\Scripts\python.exe -m ruff format --check src tests
PASS: 41 files already formatted (formatter applied before final validation).
.\.venv\Scripts\python.exe -m mypy src tests
PASS: strict configuration, no issues in 41 source files.
.\.venv\Scripts\python.exe -m pip check
PASS: no broken requirements.
git --no-pager diff --check
PASS; report single-fence/history preservation and retained native files
independently verified after writing this report.
Editor Problems: no errors in source/tests.
LIMITATIONS / READINESS
Real native CODEC/profile/contract implementation is complete for this subset.
No absent-evidence stop remains for these Riffer targets. This is NOT a claim
of plugin/Cubase native load/save/reload, audible articulation, arbitrary
Riffer import, every tuning/technique, native service output or SD3 support.
All application acceptance is UNRUN, not a pytest skip or inferred success.
Ready for later native service integration/SD3 work; neither is performed here.
=======================================================================
HISTORICAL RESEARCH — PRESERVED, SUPERSEDED BY CURRENT IMPLEMENTATION ABOVE
The following is the prior research-only report, unchanged. Its statements
about missing local/version/layout evidence describe that earlier investigation,
NOT the present implementation or a current Riffer blocker.
=======================================================================
Step 7 fix report — new targeted public-source AGP evidence investigation
Date: 2026-10-19
SCOPE: Research-only, per prompts/step_7_fix.md. No source/tests/docs/prompts
were modified. No commits, no delegation, no implementation. The existing
native AGP boundary in src/riff_groove_builder/formats/agp.py (available=False,
reason="agp_layout_not_established") is UNCHANGED and still correct per the
evidence below.
RESULT SUMMARY
No ".agp" binary layout, version header, field list, checksum/compression
rule, or redistributable fixture was found anywhere public. This investigation
goes materially beyond the prior three simplified fetches: it downloaded the
five official PDFs as raw bytes (not simplified text extraction), parsed real
page text with pypdf, and extracted the embedded UI screenshots from inside
the PDFs to inspect actual in-app dialogs. The literal string "agp"/"AGP" does
not appear anywhere in ~89,700 characters of extracted real manual text, and
does not appear in any extracted screenshot. The native-AGP gate remains
honestly BLOCKED. Several new, concrete, lawfully-obtained facts were
established and are recorded below, including one genuine structural
contradiction risk the orchestrator should review.
METHOD (new, not a repeat of prior simplified fetches)
1. Re-fetched https://www.amplesound.net/en/manual.asp as raw HTML (public,
unauthenticated) and confirmed the exact current link set and the site's
product navigation menu (Metal Series: "Ample Metal Hellrazer", "Ample
Metal Eclipse", "Ample Metal Ray5").
2. Downloaded the five linked PDFs as raw bytes via a stdlib-only
(urllib.request) script, run in a disposable research virtual environment
created at %TEMP%\research_pdf_venv (NOT the project .venv; no project
dependency was added or changed):
- https://www.amplesound.net/en/Guitar_Riffer.pdf
- https://www.amplesound.net/en/Guitar_Strummer.pdf
- https://www.amplesound.net/en/Main_Panel_Manual-AME.pdf
- https://www.amplesound.net/en/Main_Panel_Manual-AMR.pdf
- https://www.amplesound.net/en/Settings_and_CPCv2.pdf
3. Installed `pypdf` and `pillow` only into that disposable research venv
(`pip install pypdf pillow`, PyPI, network-enabled in this sandbox) to
extract real page text and embedded raster images (UI screenshots) from
each PDF. This is standard, lawful text/graphic extraction of a publicly
posted manual for inspection, not redistribution; no PDF, image or manual
text was copied into the repository.
4. Searched all extracted real text (not the earlier raw zlib-byte-soup scan,
which produced false "agp" hits inside undecoded binary image streams) for
"agp", ".agp", "riff file", "groove", "preset", "export", "save", "load",
"bank", "extension", "file format", case-insensitively.
5. Viewed the highest-value embedded screenshots directly (Riffer Save Window,
Riffer Preset/Browse Window, Strummer Pattern browse/save windows, Eclipse
and Ray5 main-panel renders) to look for a visible filename-extension
string that text extraction could miss.
6. Searched GitHub (github-mcp-server search_code / search_repositories,
public, unauthenticated) for any open-source Ample Sound AGP parser,
`extension:agp`, and known multi-format sample/preset converters.
7. Attempted exact-phrase public web search (DuckDuckGo HTML endpoint, Bing)
and the official Ample Sound KVR Audio sub-forum link exposed in
manual.asp (http://www.kvraudio.com/forum/viewforum.php?f=245).
No login, CAPTCHA, or access control was bypassed. No factory song, licensed
manual page, or screenshot was saved into the repository; all downloaded
files and extracted images remained under this machine's %TEMP%, outside the
repository working tree, and are not committed or referenced as fixtures.
VERIFIED NEW FINDINGS (public, unauthenticated, directly observed)
1. No "agp" string anywhere in real extracted text. Per-document character
counts of real pypdf-extracted text and the explicit negative result:
- Guitar_Riffer.pdf ("Ample Riffer 4 Manual", 37 pages, 36,395 chars): 0
occurrences of "agp"/"AGP".
- Guitar_Strummer.pdf ("Strummer Panel Manual", 15 pages, 11,685 chars): 0
occurrences.
- Main_Panel_Manual-AME.pdf ("Ample Metal Eclipse Main Panel Manual", 14
pages, 14,056 chars): 0 occurrences.
- Main_Panel_Manual-AMR.pdf (title text literally reads "Ample Metal R Main
Panel Manual", 14 pages, 14,373 chars): 0 occurrences.
- Settings_and_CPCv2.pdf ("Settings and CPC Manual", 16 pages, 13,250
chars): 0 occurrences.
An earlier crude pass that zlib-decompressed every PDF stream and grepped
the raw bytes did show byte sequences matching "agp"/"AGP", but those were
confirmed to be random bytes inside undecoded JPEG/image XObject streams,
not real text — a false lead correctly discarded before relying on it.
2. Confirmed exact current official product names and category, directly
from three independent in-document signals (site nav menu text, PDF title
text, and the rendered plugin UI screenshot baked into the manual):
- Lead/guitar target: "Ample Metal Eclipse" (file AME; rendered UI banner
reads "AMPLE METAL ECLIPSE"). The site's Metal Series category also
lists "Ample Metal Hellrazer" and "Ample Metal Ray5" alongside it.
- Bass target: storefront/UI name is "Ample Metal Ray5" (rendered UI banner
reads "AMPLE METAL RAY5"), but the AMR manual's own text layer title
reads "Ample Metal R Main Panel Manual" (missing "ay5") — a minor,
verified documentation inconsistency in Ample's own PDF, not a file-
format fact, but worth recording since step_7_fix.md flagged exact
product names as unconfirmed. This narrows "Ample Guitars Eclips" /
"Ample Ray" (as used in the user's prior wording) to the plugins' real
names, which belong to the "Metal Series", not a generic/acoustic guitar
or bass line.
- Exact installed version/build/edition (e.g. a specific plugin build
number or purchased edition) is still NOT established by any of this;
only the product identity/category is now confirmed from public
material.
3. Riffer 4's own File Menu (section "4.5 File Menu", real body text,
Guitar_Riffer.pdf) lists exactly: New, Load, Save, Save As, Import MIDI,
Export MIDI, Export Audio — i.e. Riffer documents Save/Load as actions
distinct from Import MIDI/Export MIDI. This is textual confirmation (not
an inference from a filename) that Riffer's native Save/Load is not simply
a MIDI import/export relabeled; it is presented as its own save format in
the plugin's own menu, consistent with (but not proof of) a proprietary
container.
4. Extracted the actual embedded screenshot of Riffer's "Save Window" (manual
section 4.5.2) and "Preset Window"/browse dialog (section 4.5.1):
- Save Window default path: `C:\Users\User\Documents\Ample Sound\Riff4\
Guitar4`, fields Name / Author / Category / Type, a Save button. This is
a custom in-app widget, not an OS Save-As dialog, so the screenshot does
NOT display any filename-extension text anywhere in the rendered image.
- Preset/browse window: left-hand folder tree uses per-instrument-model
codes (AGM Pick, AGM Preset, AGM Strum, AGPF Preset, AGRB Preset, AGSC
Clean/Funk/Overdrive/Preset, AGSH Preset, AGSJ Preset), and the file list
has an "Inst" column populated with the same codes (e.g. AGPF, AGSH).
These are instrument/model abbreviations used as folder/category labels
inside the custom browser UI, not a visible file extension string; none
of them is literally "AGP" alone, and no extension is rendered anywhere
in this screenshot either.
5. Strummer documents three distinct, separately-pathed save/load features,
confirmed from real body text (not just the table of contents):
- "1.5.1 Select Mode" — explicit "Save/Load Whole Chord Bank" buttons.
- "1.6.3 Save/Load Pattern" — default path `C:\Users\User\Documents\Ample
Sound\Strum Patterns` (screenshot captured and viewed), fields Name /
Author / Category / BPM / Signature / Quantize, a Save button. Different
default path from Riffer's `Riff4\Guitar4` tree.
- "1.6.6 Convert Pattern to Riffer" — an explicit, one-directional command
that turns a Strummer rhythm Pattern into Riffer data, with the caveat
"ensure the time signature or quantization of the current pattern
matches the Riffer settings."
This is the closest this investigation got to the requested "concrete
contradiction (preset vs riff/strum extension)": the vendor's own UI/
documentation treats "Chord Bank", "Pattern" (Strummer/groove) and "Riff"
(Riffer) as three separate saved-content categories with three different
default folders and an explicit one-way conversion step between Pattern
and Riff, rather than one interchangeable native container shared across
"riff" and "groove" content. This is a real structural distinction to flag
for a scope decision — it is not proof that the on-disk extensions differ,
because no extension string was ever visible in any screenshot or text.
6. Main_Panel_Manual-AME.pdf's own body text (section 1.2 Preset Bar) states:
"Clicking the preset name text box will open the preset file list window,"
i.e. the main-panel "Preset" feature is the instrument/tone patch system
(sound settings), textually confirmed separate from Riffer's riff/groove
note content — consistent with, and now textually verified from the real
manual body (not just inferred from UI labels), the existing docs/formats/
agp.md caution that "a preset-only container may not hold the requested
riff/groove."
7. Settings_and_CPCv2.pdf section "1.2.2 Export Mode" (real body text) is
exclusively about MIDI channel-per-string export configuration ("MIDI
Guitar Channels" vs "Single Channel"), not a native save/export format.
This rules out one possible misreading of "Export" as native-AGP evidence.
8. No public open-source AGP parser or Ample Sound format support was found.
GitHub code search (search_code) for "Ample Sound" + riff/preset parsing,
`extension:agp`, and `"AGPF" OR "AGSH" OR "AGRB" Ample` returned only
unrelated false positives: the unrelated "Accelerated Graphics Port" bus-
driver `files.agp` in NetBSD/src, and substring matches inside unrelated
words ("bagpipe", "tagPolyphony") in git-moss/ConvertWithMoss, a actively
maintained open-source multi-format sample/preset converter supporting
dozens of competing plugin preset formats (Decent Sampler, TX16Wx, SFZ,
etc.). That project has no Ample Sound format support at all — useful
negative evidence that no public reverse-engineered AGP schema exists in
a well-known community converter project either.
9. Exact-phrase public web search could not be executed reliably in this
sandbox: the DuckDuckGo HTML endpoint returned an anti-bot/email-
verification challenge page instead of results, and Bing returned a
stale/irrelevant snippet about an unrelated agribusiness company named
"AGP", not real search results. The official Ample Sound KVR Audio sub-
forum index linked from manual.asp (http://www.kvraudio.com/forum/
viewforum.php?f=245) loaded (no login required) but its thread listing is
JavaScript-rendered and did not return readable thread content through
static fetch; browsing further thread pages was judged outside the
12-access, narrowly-targeted bound for this pass and was not pursued.
NO NEW NATIVE SCHEMA — GATE STILL HONESTLY BLOCKED
None of the above establishes: a container signature/magic bytes, a version
number scheme, a field list for pitch/start/duration/tempo/velocity/
articulation, any length/checksum/compression rule, or any external-reference
semantics. No actual ".agp" (or correctly-named equivalent) file, lawful
empty/one-note/multi-note export, or redistribution permission was obtained.
Native AGP acceptance for Ample Metal Eclipse (lead) and Ample Metal Ray5
(bass) remains BLOCKED exactly as recorded in report_7.md and docs/formats/
agp.md. This report does not reverse, soften, or re-derive that conclusion;
it adds verified detail and rules out several plausible-sounding shortcuts
(renamed MIDI export, preset/tone file, Export Mode MIDI-channel setting)
without inventing a layout.
REMAINING BLOCKERS
- The real on-disk extension used by Riffer "Riff" saves, Strummer "Pattern"
saves, and Strummer "Chord Bank" saves is still not established from any
public source; no screenshot or manual text ever displays the filename
extension because both Riffer's and Strummer's save/browse dialogs are
custom in-app widgets, not OS file pickers, and no manual prose names the
extension in writing.
- Exact installed product build/edition/version for the user's "Ample Metal
Eclipse" and "Ample Metal Ray5" copies is still unconfirmed.
- No lawful controlled empty/one-note/multi-note export exists, so no
pitch/timing/tempo/velocity/articulation field layout can be derived or
tested.
- No redistribution permission or documented local-only fixture procedure
exists for any native file.
- No real Eclipse/Ray5 load, inspect, save, close/reload acceptance test has
been run; this remains UNRUN, not a passing or skipped automated check.
- The Strummer Pattern/Chord-Bank vs Riffer Riff distinction (finding 5 above)
is a genuine open structural question for the orchestrator: if the user's
requested "groove" content for bass is actually a Strummer Pattern/Chord
Bank (not a Riffer Riff), a single shared lead/bass ".agp" writer assumption
may be wrong even if a real schema is eventually obtained for one of them.
This should be resolved by product-specific confirmation, not invented.
EXACT LAWFUL INPUTS STILL NEEDED (unchanged in kind from docs/formats/agp.md,
now sharpened by the product-identity confirmation above)
1. Confirmation of the exact installed product/edition/build for the user's
"Ample Metal Eclipse" (lead) and "Ample Metal Ray5" (bass) copies, and
which specific save feature (Riffer Riff, Strummer Pattern, or Strummer
Chord Bank) the user actually means by "riff"/"groove" content for each.
2. A real Save-As (or equivalent OS file dialog, if any exists outside the
custom in-app browser) performed once by the rights-holder/operator on
that exact installed copy, so the real on-disk filename and extension can
be observed directly instead of guessed.
3. Lawfully created empty, one-note, and multi-note exports from that same
controlled setup, with redistribution permission, or an explicit
operator-documented local-only fixture ID and test procedure if
redistribution is not permitted.
4. One-variable-at-a-time controlled exports (tempo, note timing/duration,
velocity, and each required articulation / release type) from the same
installed copy, needed to derive the real field layout once a sample
container is obtained.
5. Real load/save/close/reload acceptance performed in the actual Eclipse and
Ray5 plugins (and Cubase routing, if relevant), recording whether pitch,
timing, dynamics and articulation survive.
READINESS
Software capability boundary: unchanged, still PASS (from report_7.md).
Real native AGP acceptance: still BLOCKED. No fabricated layout, no renamed
MIDI, no invented extension was substituted. This fix step is complete as a
bounded, honest research pass: 2 negative/limited search attempts
(DuckDuckGo, Bing), 1 partially-useful public page (KVR forum index, no
readable thread content), and 9 substantive public-source accesses (manual.asp
raw HTML, the 5 official PDFs each independently re-parsed for text AND
images, and GitHub code/repo search across 5 queries) were used, within the
requested bound. No commit, delegation, or scope expansion occurred. Ready
for the orchestrator to decide between (a) obtaining the exact lawful inputs
listed above, or (b) a scope decision about whether Strummer Pattern/Chord-
Bank content should be targeted instead of, or in addition to, Riffer Riff
content for the bass/groove side.
```