An amp's signal finally reaches the speaker and becomes vibration of air. For a long time, that was the "end" of the sound.
Load boxes and IRs changed that ending — you could say, for now. But summarize it as "the speaker variable was simply swapped for digital" and you drop the most interesting part. In fact, the signal chain is cut at one point, and the front stays as live analog, only part of the severed side is replaced by computation, and part is lost. This piece digs into "what was kept and what was lost."
The speaker was the last — and nonlinear — analog stage
First, pin down exactly what the speaker was doing.
A speaker is a transducer that turns an electrical signal into vibration of air. But not an accurate one. It emphasizes certain bands, resonates with the box, and — this is key — behaves nonlinearly at high volume. The cone collapses at the limit (cone breakup), the voice coil compresses from heat, and the response itself changes with the input level. That a distorted signal settles into the ear as "the amp's sound" is after passing this nonlinear final stage. The speaker wasn't just coloring the sound — it was a live analog stage that dynamically deformed the signal according to its level.
First came attenuators and loads
Follow the history and it didn't go straight to digital.
The first motive was simply "volume." Drive the power stage and it sounds good, but that needs high volume. So a device was born that receives the power stage's output and drops only the volume before it reaches the speaker. Tom Scholz's Power Soak in the late 1970s and the THD Hot Plate in the '90s are early representatives. This developed into load boxes like the Fryette Power Station, Suhr Reactive Load, Two Notes Torpedo, and Universal Audio OX Amp Top Box — gear that keeps the power stage fully rung while taking out the signal without driving a speaker loud. That you can record the sound of a cranked 100W power stage at midnight is thanks to this lineage.
A load box is not "swapping a variable for digital"
Here's the crucial point the summary hides.
A load box isn't a box that swaps the speaker for digital. It's a box that cuts the signal chain right after the power stage. So the tube power stage's distortion, the output transformer's saturation, the supply's sag — behaviors that make up a large part of an amp's "character" — stay alive as analog and are extracted as is.
And there's a quality to how they remain. A speaker's impedance varies greatly with frequency (rising at the low resonance and in the highs). The power stage makes the sound and playing feel by interacting with this varying load. So with a resistive load that merely receives with a resistor, the load the power stage "feels" differs from the real thing, and the sound and response change. Against this, a reactive load modeling the speaker's impedance curve keeps even that interaction. That recent load boxes are all reactive is because, even severed, they try to keep the relationship of power stage and speaker. What was replaced is only the speaker's "acoustic conversion"; the power stage's behavior wasn't replaced.
An IR is only a linear snapshot
So where did the severed speaker acoustics go? The IR (impulse response).
An IR records how a certain speaker, cab, and mic respond to a signal, as one linear filter. It faithfully reproduces, by convolution, the frequency response and phase — that is, the "color" the speaker-plus-mic makes. As Celestion distributes official IRs, countless cab × mic colors are now data. Up to here, it's really well made.
But what an IR captures is only linear color. The speaker's nonlinear behavior described above — cone collapse at high volume, voice-coil compression, response changing with input level — doesn't fit into a static linear filter. An IR is a photograph of the "color" at one point, and the liveness of a speaker dynamically flexing with level doesn't remain as is. That a plain IR can sound somehow flat compared to an actually-mic'd cab is because of this loss. Recently, methods capturing multiple levels to approach dynamic behavior have appeared, but the principle — that a single static IR is a linear approximation — doesn't change.
So the question isn't "physical or digital"
Decomposed this way, the story stops being "the speaker was swapped from physical to digital."
The chain splits into three layers. (1) The power stage's behavior (the load box keeps it as analog), (2) the speaker-plus-mic's linear color (the IR reproduces it well), (3) the speaker's nonlinear dynamics (a linear IR loses it). What load boxes and IRs did wasn't turning the single speaker variable wholly digital — it was separating these three layers and handling each differently. What stays alive, what's approximated, what's lost. More accurately: the ending didn't change — the ending was decomposed into multiple layers.
So how is CSL's record different from an IR?
Here's something to honestly admit. CSL ultimately turns the rung sound into a Kemper profile. And a Kemper profile is, pushed to the end, a relative of the IR.
A Kemper profile captures the amp's nonlinear part (how it makes distortion) as a model. But the cab-and-speaker part is treated as a linear response, almost the same as an IR. That's exactly why you can swap to another cab or layer an external IR on the Kemper. So even when CSL mics a real Marshall 1912, the moment it's turned into a profile, (3) the speaker's raw nonlinearity — cone collapse and compression at high volume — is frozen just like an IR, convolved into a near-linear form. It's not that "because we record with a real cab, we keep the raw physics an IR loses." I want to make this clear. What Kemper adds over a plain IR is (1) the amp's nonlinearity, and that's obtainable via a load box too. As for the speaker layer, what CSL does is effectively no different from an IR.
So where's the meaning in CSL insisting on a real cab and mic? It isn't "keeping the raw nonlinearity." It's freezing one state that actually existed, whole and integrated. Rather than multiplying a generic DI by a generic IR someone else measured separately, "this real amp actually rang this real cab (1912), at this setting, with an SM57 and a GT-67" — capturing that integrated result, nature and all. It's the sound of the moment the power stage and the real speaker actually interacted. The difference isn't whether you can keep physics live, but the provenance of the record. A combination of generic parts, or a copy of a specific state that existed. That CSL sells not "the best sound" but "the record of a sound that actually existed" is consistent here too.
In short
Load boxes and IRs severed the speaker, the final stage. But that isn't "a variable simply swapped for digital." A reactive load keeps the power stage's behavior as analog; a linear IR reproduces the speaker's color well but loses the nonlinear dynamics. The ending wasn't replaced — it was decomposed into three layers.
And — this is key — a Kemper profile is a relative of the IR too, and the speaker layer is likewise frozen linearly. So CSL recording with a real cab and real mic isn't to keep raw physics live. What can be kept is a copy of a specific state that existed, nature and all. Where and how a sound "ends" is now an object of design and choice, and every one of them — IRs, modelers, CSL's profiles — is, to differing degrees, a "freeze." The difference is in what was made the source of the freeze, and in the honesty about it.