What 149,098 shipped drivers say about our production consistency
Three years of real end-of-line measurements from two high-volume OEM
programs — every shipped unit counted once, published as distributions rather than claims.
Every driver REDCATT ships passes an automated acoustic test before it leaves the line.
This page shows what those tests recorded for two OEM products in series production
from 2023 to 2026: 149,098 shipped units in total, each counted once by its
final test record. The products are REDCATT's own
transducer designs, manufactured for OEM customers — anonymized here under the
confidentiality those programs require — and the data is untouched: it comes directly from
the test-station databases, unit by unit, with serial-number and timestamp traceability.
How the measurement works — read this first
Each production line ends in an in-line acoustic test station: a measurement
box of roughly 1.2 × 1.2 m with a microphone about 40 cm from
the driver under test. Every unit is placed in the box and measured automatically —
DC resistance, impedance curve, resonance frequency, output level, frequency response,
distortion, Rub & Buzz, and polarity.
This is not an anechoic chamber — by design. A small box with
a close microphone has reflections and boundary effects, so the absolute SPL values and
curve shapes here differ significantly from free-field datasheet measurements. That is
irrelevant to the question this station answers, because the test is comparative:
before production starts, a golden sample — a reference unit validated in the lab —
is measured in the same box, and every production unit is compared against it under
identical conditions. Acceptance limits are derived from the golden sample's measurement.
What you should read from every chart on this page is spread, never absolute values.
Who is in the data: all units that passed the end-of-line test — that is,
the units that actually shipped — each counted once, by its final test record. Units
that never met the limits are not shipped and are excluded, so these distributions
describe what a customer receives; the size of that share is published for
each program below as the final acceptance yield, year by year. Test temperature and
humidity are recorded with every measurement.
How to read the charts. Histograms show how many units (vertical) measured a given
value (horizontal); a tall narrow shape means high consistency. σ ("sigma") is the standard
deviation: about two-thirds of all units fall within ±1σ of the mean. Red dashed lines are
acceptance limits; the red curve or line is always the golden reference.
Program A — high-frequency compression driver
A REDCATT-designed transducer in series production for an OEM
program since 2023-07. Every figure below
is computed from the final end-of-line test record of every unit that shipped —
each unit counted once.
52,621
shipped units in this dataset
2023-07 – 2026-08
production period
15 / 39
production batches / test-station sessions
±0.18 dB
sensitivity spread (1σ) vs golden sample
98.8%
final acceptance yield, 2026 to date
99.5% of 52,621 units measured within
±0.5 dB of the golden-referenced mean — and >99.99% within ±1 dB.
The worst single month in 30 months of production averaged 0.36 dB
from the golden sample.
Final acceptance yield, year by year
This table shows, of the units reaching the end-of-line test station, the
share that ultimately met every acceptance limit and shipped. A unit that fails is not necessarily
scrapped: where the design allows a mechanical adjustment it is re-worked and re-tested,
and every unit is counted once, by its final test result. Units that never meet the
limits are not shipped. Units rejected at earlier assembly stations never reach this
test, so this is the acoustic-test yield, not a total build yield.
Year
Units reaching final test
Shipped (met every limit)
Final yield
2023
2,376
2,127
89.5%
2024
23,211
21,307
91.8%
2025
22,244
21,698
97.5%
2026 (through Aug)
7,579
7,489
98.8%
Whole period
55,410
52,621
95.0%
For this compression driver the typical adjustment is re-shimming the magnetic gap — a normal tuning step, not a defect repair. The 2023–2024 figures reflect the program's ramp-up phase; as the assembly and gap-shimming process matured, the yield rose to 97.5% in 2025 and 98.8% in 2026.
Sensitivity, unit by unit
Each driver's output level is measured and compared with the golden sample
on the same station. This chart shows how far each shipped unit landed from its golden
reference — 0 dB means "identical to the golden sample". Mean deviation -0.02 dB,
spread ±0.18 dB (1σ); every unit fits inside the plotted range. Red dashed lines are
the most common acceptance limits.
Resonance frequency (fs)
The resonance frequency of every shipped unit, against the most common
acceptance window (800–1,300 Hz). Mean 1,077.8 Hz, spread ±59.7 Hz (1σ), i.e. ±5.5% relative.
A distribution well inside its window means consistent moving mass and suspension
stiffness — the parameters that drift first when a production process is not
controlled. Every shipped unit lies inside this window.
Voice-coil DC resistance (Re)
Re reflects the winding process itself: wire, layers, tension.
Mean 12.840 Ω, spread ±0.368 Ω (1σ). 9 of 52,621 units sit outside the plotted range — accepted under period windows re-derived from the golden sample, wider than the most common one drawn here; no unit shipped outside its own window.
Month by month, 30 months
The blue line is the monthly average output level of production units; the
band around it is ±1σ for that month; the red line is the golden reference itself as
re-measured over time. Note the magnified scale: the total range of the monthly means over
the whole period is 0.63 dB. The
two lines moving together is the point — shared movement comes from the station and
its environment, and production tracks the reference through 30 months, seasons,
and material lots. Months with fewer than 100 units are left off this chart.
The same view for resonance frequency, with the fixed
acceptance limits in red — note how much headroom the production spread keeps inside the window.
150 drivers against one golden sample
The blue band shows where the frequency responses of 150 production
units from one batch fall: the darker band holds the middle half of all drivers, the lighter
band 90% of them. The red line is the golden reference they were tested against — it runs
through the middle of the band because production and reference are, in practice, the same.
Through the working band the unit-to-unit spread is about ±0.26 dB (1σ),
widening toward the very top of the range where cone/diaphragm break-up makes every
driver's response most individual — normal physics, not process variation. Remember from
"How the measurement works": the curve shape belongs to the 1.2 m test box and
the 40 cm microphone, not the driver's free-field response — read the band's tightness,
not the shape.
The engineer's view of the same data — every one of
the 150 units drawn individually, golden reference in red:
Program B — low-frequency woofer
A REDCATT-designed transducer in series production for an OEM
program since 2023-08. Every figure below
is computed from the final end-of-line test record of every unit that shipped —
each unit counted once.
96,477
shipped units in this dataset
2023-08 – 2026-08
production period
14 / 60
production batches / test-station sessions
±0.21 dB
sensitivity spread (1σ) vs golden sample
99.5%
final acceptance yield, 2026 to date
98.4% of 96,235 units measured within
±0.5 dB of the golden-referenced mean — and >99.99% within ±1 dB.
The worst single month in 34 months of production averaged 0.34 dB
from the golden sample.
Final acceptance yield, year by year
This table shows, of the units reaching the end-of-line test station, the
share that ultimately met every acceptance limit and shipped. A unit that fails is not necessarily
scrapped: where the design allows a mechanical adjustment it is re-worked and re-tested,
and every unit is counted once, by its final test result. Units that never meet the
limits are not shipped. Units rejected at earlier assembly stations never reach this
test, so this is the acoustic-test yield, not a total build yield.
Year
Units reaching final test
Shipped (met every limit)
Final yield
2023
13,374
13,284
99.3%
2024
28,779
28,365
98.6%
2025
24,125
23,874
99.0%
2026 (through Aug)
31,111
30,954
99.5%
Whole period
97,389
96,477
99.1%
Sensitivity, unit by unit
Each driver's output level is measured and compared with the golden sample
on the same station. This chart shows how far each shipped unit landed from its golden
reference — 0 dB means "identical to the golden sample". Mean deviation +0.09 dB,
spread ±0.21 dB (1σ); every unit fits inside the plotted range. 242 early units lack a stored reference value and are not part of this comparison. Red dashed lines are
the most common acceptance limits.
Resonance frequency (fs)
The resonance frequency of every shipped unit, against the most common
acceptance window (56–76 Hz). Mean 64.1 Hz, spread ±2.45 Hz (1σ), i.e. ±3.8% relative.
A distribution well inside its window means consistent moving mass and suspension
stiffness — the parameters that drift first when a production process is not
controlled. Acceptance windows are re-derived from the golden sample's measurement per production period; 4 units from an early run were accepted under a slightly different window and sit outside the lines drawn here — no unit shipped outside its own window.
Voice-coil DC resistance (Re)
Re reflects the winding process itself: wire, layers, tension.
Mean 2.021 Ω, spread ±0.087 Ω (1σ). The two visible groups are separated in time, not mixed production: units in the slightly higher group were built during 2024–2025, and 2026 production is back at the original value. Both groups sit well inside the acceptance limits. 621 of 96,477 units sit outside the plotted range — accepted under period windows re-derived from the golden sample, wider than the most common one drawn here; no unit shipped outside its own window.
Month by month, 34 months
The blue line is the monthly average output level of production units; the
band around it is ±1σ for that month; the red line is the golden reference itself as
re-measured over time. Note the magnified scale: the total range of the monthly means over
the whole period is 0.91 dB. The
two lines moving together is the point — shared movement comes from the station and
its environment, and production tracks the reference through 34 months, seasons,
and material lots. Months with fewer than 100 units are left off this chart.
The same view for resonance frequency, with the fixed
acceptance limits in red — note how much headroom the production spread keeps inside the window.
150 drivers against one golden sample
The blue band shows where the frequency responses of 150 production
units from one batch fall: the darker band holds the middle half of all drivers, the lighter
band 90% of them. The red line is the golden reference they were tested against — it runs
through the middle of the band because production and reference are, in practice, the same.
Through the working band the unit-to-unit spread is about ±0.25 dB (1σ),
widening toward the very top of the range where cone/diaphragm break-up makes every
driver's response most individual — normal physics, not process variation. Remember from
"How the measurement works": the curve shape belongs to the 1.2 m test box and
the 40 cm microphone, not the driver's free-field response — read the band's tightness,
not the shape.
The engineer's view of the same data — every one of
the 150 units drawn individually, golden reference in red:
Questions this data usually raises
Why do the response curves look different from a datasheet?
Datasheet curves are measured free-field in an anechoic environment. These are
measured in a 1.2 m in-line test box with the microphone at 40 cm, where
reflections shape the curve. The station measures every unit the same way and compares
against the golden sample measured in the same box — so the spread between curves is
meaningful, while the curve shape itself is a property of the box.
What exactly is a golden sample?
A reference unit of the product, validated in the laboratory, kept at the line, and
measured on the same station. Production acceptance limits are derived from its
measurement, and it is re-measured over time. That is why the red reference line in
the monthly charts moves slightly: the golden sample itself reads a little differently
as the station's environment changes — and production moving together with it is
exactly what those charts are there to show.
Why show only units that passed? And how many is that?
The distributions describe the population a customer receives, so they cover
shipped units only. The count is published in each program's
"Final acceptance yield" table above: over the whole period, 95–99%
of units reaching the end-of-line test ultimately met every limit, and the
current-year figures are 98.8% and 99.5%. Units may be tested more than once — a failing compression driver
is typically re-tuned in the magnetic gap (shim washers) and re-tested, a normal
mechanical tuning step — so every statistic on this page counts each unit once, by
its final test record; mixing retest records into a shipped-product statistic would
overstate variation.
Why are there more test-station sessions than batches?
A production batch is often tested across several station sessions: a session is
opened whenever the line starts a test run, so line changeovers, multi-day batches and
re-test passes each add a session. Batches count what was ordered and built; sessions
count how the station recorded the work.
Are these all the units, or a sample?
All of them. The scalar statistics (sensitivity, fs, Re) cover
every shipped unit in the period, with two small exceptions each chart states in
place: a fraction of a percent of records lack a stored value for a given quantity
(for example, the sensitivity comparison skips units with no stored golden-reference
value), and the monthly charts leave off months with fewer than 100 units. Only the
curve-overlay charts use a sample of the stored curves, and say so in their caption.
Can this be verified?
Yes. Every value traces to a per-unit test record with serial number and timestamp
in the production test-station database. Qualified OEM customers can review
project-specific records, control plans, and acceptance criteria under NDA — and
factory visits are welcome.