# Errata in ISO 12354 and EN 12354

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/reference/errata/iso-12354/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/reference/errata/iso-12354/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-10-28
Updated: 2026-11-16T21:36:14Z
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: Errata in ISO 12354 and EN 12354: 17 defects in the printed text, with the evidence and what phonometry does about each.

---

These are the 17 entries of the [errata registry](https://docs.jmrp.io/phonometry/reference/errata/) on EN 12354-1:2000, ISO 12354-1:2017, EN 12354-2:2000, EN 12354-3:2000, EN 12354-5:2009 and ISO 12354-2:2017, in the order the registry lists them. The registry's introduction says how each claim is checked against the printed page and what the status of an entry means.

## EN 12354-1:2000 Formula (E.5) / ISO 12354-1:2017 E.3.4 (K24 clamp misprint)

- **Location:** EN 12354-1:2000, Annex E, the wall-junction-with-flexible-
  interlayers block printed under Figure E.5 and numbered Formula (E.5)
  (printed p. 46), and ISO 12354-1:2017, E.3.4 NOTE 4. Annex E of the 2000
  edition has only two numbered clauses, E.1 "Determination methods" and E.2
  "Empirical data", so "E.5" is a formula number, not a clause; an earlier
  revision of this entry cited it as a clause.
- **The print:** $K_{24} = 3{,}7 + 14{,}1 M + 5{,}7 M^{2}\ \text{dB}$;
  $0 \le K_{24} \le -4\ \text{dB}$ ; $0\ \text{dB / octave}$, i.e. the bound
  on the $K_{24}$ junction term is an empty interval; the 2017 edition repeats
  the 2000 misprint verbatim.
- **The problem:** the interval is impossible as printed; the accompanying
  figure and the physics (the term is a reduction bounded below) indicate
  $-4\ \text{dB} \le K_{24} \le 0\ \text{dB}$.
- **Evidence:** the Figure E.5 curve family on the same page runs the $K_{24}$
  branch from 0 dB down to about −4 dB over the plotted mass ratios, which is
  the interval read in the other order. Verified on PDF page 48 (printed p.
  46) of EN 12354-1:2000 and PDF page 52 (printed p. 46) of ISO 12354-1:2017.
- **Library behaviour:** implements the clamp as $-4 \le K_{24} \le 0$ with a
  misprint note in the docstring.
- **Status:** unreported.

## EN 12354-1:2000, Figure E.9 (E.7) (K24 stated in the figure-axis mass ratio)

- **Location:** Annex E, Figure E.9 / Formula (E.7) (junction of lightweight
  double leaf wall and homogeneous elements), the $K_{24}$ line.
- **The print:** $K_{24} = 3{,}0 - 14{,}1 M + 5{,}7 M^{2}\ \text{dB}$ (for
  $m_2/m_1 > 3$), under a figure whose x-axis is $m_2/m_1$.
- **The problem:** Annex E defines $M$ per transmission path as
  $M = \lg(m'_{\perp,i}/m'_i)$ (perpendicular element over the element
  carrying the path). The $K_{24}$ path 2→4 is carried by the homogeneous
  element ($m_2 = m_4$) with the leaf ($m_1$) perpendicular, so the per-path
  $M$ is $\log_{10}(m_1/m_2)$, but the printed $K_{24}$ line only matches its
  own figure's curve when $M$ is read as the x-axis variable
  $\log_{10}(m_2/m_1)$ (e.g. −2.4 dB at $m_2/m_1 = 3$, −5.4 dB at 10). Read
  with the annex's declared $M$, the line contradicts the figure by
  $28.2 \cdot |\log_{10}(m_2/m_1)|\ \text{dB}$. The same edition's other
  $K_{24}$ line (Figure E.5, Formula (E.5)) *does* follow the declared
  per-path $M$, so the two $K_{24}$ prints of the 2000 edition silently use
  different conventions. ISO 12354-1:2017 E.3.5 prints the relation
  consistently in the per-path convention of its Formula (E.3),
  $K_{24} = 3.0 + 14.1 M + 5.7 M^{2}$; the two editions agree
  numerically (an earlier revision of this entry read the 2017 print as a sign
  misprint; re-derivation against both editions' figures shows it is a
  convention recast, not a defect of the 2017 text).
- **Evidence:** numerical evaluation of both forms against the Figure E.9
  curve. Verified on PDF page 44 (printed p. 42), PDF page 48 (printed p. 46)
  and PDF page 50 (printed p. 48) of EN 12354-1:2000, and of PDF page 53
  (printed p. 47) of ISO 12354-1:2017, whose E.3.5 prints its K24 line beside
  a Figure E.7 that carries no mass-ratio axis at all.
- **Library behaviour:** implements the per-path convention uniformly
  (`junction_vibration_reduction`, mass_ratio = $m'_{\perp,i}/m'_i$ for every
  branch), so the E.7 double-leaf branch takes leaf-over-homogeneous ratios
  below 1/3 and evaluates $3.0 + 14.1 M + 5.7 M^2$.
- **Status:** unreported.

## EN 12354-2:2000, Formula (3) vs Annex E.3 (standardized impact level)

- **Location:** Formula (3) and worked example E.3.
- **The print:** Formula (3) defines
  $L'_{nT} = L'_n - 10 \lg(0{,}16 \cdot V/(A_0 \cdot T_0))$, which reduces
  exactly to $L'_n - 10 \lg(0{,}032 \cdot V)$, i.e. a reference volume of
  $31{,}25\ \text{m}^3$. Annex E.3 states "from equation (3):
  $L'_{nT,w} = L'_{n,w} - 10 \lg(V/30)$".
- **The problem:** the annex's $V/30$ is a rounding of the formula's own
  constant; the two differ by a constant 0.177 dB.
- **Evidence:** direct algebra; both variants recomputed for the E.3 case
  (42.959 vs 42.782 dB, both rounding to 43 in that example). Verified on PDF
  page 7 (printed p. 5) and PDF page 34 (printed p. 32) of EN 12354-2:2000.
- **Library behaviour:** implements the exact $0.032 \cdot V$ form and
  documents the annex's rounding.
- **Status:** unreported.

## EN 12354-3:2000, Formula (5) (reduced form of the normalized level difference)

- **Location:** clause 3.1.5 "Relations between quantities", Formula (5)
  (printed p. 6).
- **The print:**
  $D_{2m,n} = D_{2m,nT} - 10 \lg[0{,}16\,V/(T_0 A_0)] = D_{2m,nT} - 10 \lg 0{,}32\,V\ \text{dB}$.
- **The problem:** the reduced form is off by a factor of ten. Six lines above
  it, the where-list of clause 3.1.4 defines $A_0$ as "the reference
  equivalent sound absorption area, in square metres, for dwellings given as
  $10\ \text{m}^2$", and the where-list of clause 3.1.3 on the preceding page defines
  $T_0$ as "the reference reverberation time, in seconds, for dwellings given
  as 0,5 s". So $0.16/(T_0 A_0) = 0.16/5 = 0.032$, not 0.32. Applied as
  printed, the reduced form shifts every normalized façade level difference by
  exactly $10\log_{10} 10 = 10\ \text{dB}$. The exact analogue in the
  companion part, EN 12354-2:2000 Formula (3), prints the same algebra
  correctly:
  $L'_{nT} = L'_n - 10 \lg[0.16\,V/(A_0 T_0)] = L'_n - 10 \lg 0.032\,V\ \text{dB}$.
  ISO 12354-3:2017 dropped the reduced form altogether: its Formula (5) prints
  only $D_{2m,n} = D_{2m,nT} - 10 \lg[C_\text{sab} V/(A_0 T_0)]$ with
  $C_\text{sab} = 0.16\ \text{s/m}$.
- **Evidence:** direct algebra with the standard's own $A_0$ and $T_0$, and
  the side-by-side comparison with the correctly reduced Formula (3) of Part
  2. Verified on PDF page 8 (printed p. 6) and PDF page 7 (printed p. 5) of EN
  12354-3:2000, on PDF page 7 (printed p. 5) of EN 12354-2:2000 for its
  Formula (3), and on PDF page 12 (printed p. 6) of ISO 12354-3:2017 for the
  2017 Formulae (4) and (5).
- **Library behaviour:** unaffected. No code path implements the reduced form:
  the façade model computes $D_{2m,nT}$ from Formula (13)
  ([`facade.py`](https://github.com/jmrplens/phonometry/blob/main/src/phonometry/building/prediction/facade.py)), and the
  survey method converts with the unreduced
  $D_{2m,n} = D_{2m} + k + 10\log_{10}[A_0 T_0/(0.16 V)]$ of ISO 10052
  Clause 3.15
  ([`survey_insulation.py`](https://github.com/jmrplens/phonometry/blob/main/src/phonometry/building/measurement/survey_insulation.py)).
  The two standardization constants that *are* pre-folded elsewhere in the
  library are both correct: $0.032$ for the Part 2 impact form and $0.32$
  for the Part 1 airborne form
  $D_{nT} = R' + 10\log_{10}(0.16 V/(T_0 S_s))$, where the denominator is an
  area rather than $A_0$.
- **Status:** unreported.

## EN 12354-3:2000, Formula (13) vs its own Annex F example (the "6" constant)

- **Location:** clause 4.1, Formula (13) (printed p. 9), against the worked
  example of Annex F (printed pp. 27-28).
- **The print:** Formula (13) gives
  $D_{2m,nT} = R' + \Delta L_\text{fs} + 10 \lg[V/(6 T_0 S)]\ \text{dB}$,
  while the Annex F.1.3 result table prints a $D_{2m,nT}$ row that is exactly
  $R' + 1{,}5\ \text{dB}$ in all five octave bands and in the single-number
  column (25,9/23,0/26,4/36,9/39,0 against 24,4/21,5/24,9/35,4/37,5, and 29,3
  against 27,8).
- **The problem:** on this constant the *example* is self-consistent and the
  *formula* is the outlier. (Two cells of the same annex table do not follow
  from its element rows, which is the subject of the next entry; the printed
  $+1.5$ dB row holds in every band regardless, so the two defects are
  independent.) With the example's own inputs ($V = 50\ \text{m}^3$,
  $S = 11.3\ \text{m}^2$, $T_0 = 0.5\ \text{s}$,
  $\Delta L_\text{fs} = 0$), the Sabine form gives
  $10\log_{10}[0.16 \cdot 50/(0.5 \cdot 11.3)] = 1.5104\ \text{dB}$,
  which is the printed +1.5 dB row; Formula (13) as printed gives
  $10\log_{10}[50/(6 \cdot 0.5 \cdot 11.3)] = 1.6877\ \text{dB}$. The
  gap is the constant: Formula (13)'s "6" is a rounded $1/0.16 = 6.25$,
  and $10\log_{10}(6.25/6) = 0.177\ \text{dB}$ is exactly the discrepancy.
  ISO 12354-3:2017 replaced it with an explicit Sabine constant, printing
  Formula (4) as
  $D_{2m,nT} = R' + \Delta L_\text{fs} + [10 \lg(C_\text{sab} V/(T_0 S))]$
  with $C_\text{sab} = 0.16\ \text{s/m}$, which is the constant the 2000
  example already used. A previous revision of this entry attributed the 1.5
  dB row to the example; the attribution is the other way round.
- **Evidence:** evaluation of both constants against the printed Annex F rows,
  which agree with 0.16 to the 0.05 dB the table carries and disagree with the
  rounded 6 by a uniform 0.18 dB; and the 2017 recast, which adopts the
  example's constant. The example's single-number result
  $D_{2m,nT,w} = 33\ \text{dB}$ is insensitive to the difference and
  reproduces either way. Verified on PDF pages 11 (printed p. 9), 29 (printed
  p. 27) and 30 (printed p. 28) of EN 12354-3:2000, and of PDF page 12
  (printed p. 6) of ISO 12354-3:2017.
- **Library behaviour:** implements Formula (13) as printed, with the rounded
  6; the test data records that the Annex F rows follow the exact 0.16
  constant and sit 0.18 dB below the model.
- **Status:** unreported.

## EN 12354-3:2000, Annex F.1.3 (the 1 kHz and 2 kHz R' cells)

- **Location:** Annex F, table F.1.3 "Results for façade" (printed p. 28), the
  `R' (equation 10)` row.
- **The print:** $R'$ = 24,4 / 21,5 / 24,9 / 35,4 / 37,5 dB at 125 / 250 / 500
  / 1000 / 2000 Hz.
- **The problem:** the last two cells do not follow from the table's own
  element rows. Formula (10), $R' = -10\log_{10} \sum \tau_{e,i}$, applied to
  the four $-10\log_{10} \tau_e$ columns printed immediately above gives 24.41
  / 21.50 / 24.86 / **35.78** / **37.99** dB. The first three cells reproduce
  to the 0.05 dB the table carries; the 1 kHz and 2 kHz cells are printed 0.4
  dB and 0.5 dB low.
- **Evidence:** energy summation of the printed element rows band by band (1
  kHz: 60.7 / 40.0 / 46.6 / 38.5 dB; 2 kHz: 66.7 / 41.0 / 43.6 / 44.5 dB). The
  $D_{2m,nT}$ row below is a uniform $R' + 1.5\ \text{dB}$ in every band
  including those two, so it inherits the same offset, and the single-number
  result $D_{2m,nT,w} = 33\ \text{dB}$ is insensitive to it and still
  reproduces. Verified on PDF page 30 (printed p. 28) of EN 12354-3:2000.
- **Library behaviour:** the test data notes the inconsistency next to the
  affected anchor.
- **Status:** unreported.

## EN 12354-5:2009, Table F.1 and clause F.4.2 (reference force printed as 1 pN)

- **Location:** Annex F, clause F.4.2: the symbol list of Formula (F.9), the
  sentence introducing the closed form, and the caption of Table F.1 (printed
  p. 59).
- **The print:** "$L_\mathrm{F}$ is the force level in the source room, in dB re 1 pN";
  "$L_\mathrm{F} = 10\lg 2{,}5f/10^{-12}$ dB re 1 pN or $L_\mathrm{F} = 10\lg 0{,}8f/10^{-12}$ dB
  re 1 pN for one-third octave bands"; and "Table F.1 – Force level $L_\mathrm{F}$ re
  1 pN for the ISO tapping machine in octave bands", whose eight cells read
  139, 142, 145, 148, 151, 154, 156 and 156 dB.
- **The problem:** the reference force of those levels is $10^{-6}$ N, not
  1 pN. Three independent readings agree, and none of them is compatible with
  the printed reference. **(a) The annex's own algebra.** A power level re
  1 pW built from a force level and a mobility is
  $L_\mathrm{W} = L_\mathrm{F} + 10\lg(F_0^2 Y / W_0)$. Formula (D.5a) prints
  $L_\mathrm{Ws,c} = L_\mathrm{F,eq} + 10\lg Y_\mathrm{s}$ and Formula (D.9a) prints
  $L_\mathrm{Ws,c} = L_\mathrm{F} - 5 - 10\lg f$, which is the same expression evaluated at
  the mass-like source mobility $Y_\mathrm{s} = (2\pi f M)^{-1}$ of a 0.5 kg tapping
  hammer. Neither carries a term for $F_0^2/W_0$, so both balance only when
  $F_0^2 / W_0 = 1\ \text{s}^{-1}$, that is $F_0 = 10^{-6}$ N; read re 1 pN
  each would fall 120 dB short of the level it defines. The velocity
  counterpart, Formula (D.10a), does print its reference term
  $10\lg(v_\text{ref}^2/W_\text{ref})$ and states the result cancels the
  $10\lg Z_s$ exactly, which it does at the $10^{-9}$ m/s the standard itself
  gives as the velocity-level reference in clause F.4.2. The annex is
  therefore explicit and correct about the velocity reference and silent about
  the force one. **(b) The machine that produces the table.** The ISO tapping
  machine drops 0.5 kg hammers from 40 mm at ten impacts per second, so each
  impact transfers a momentum of 0.443 N·s and the force is a 10 Hz impulse
  train every harmonic of which carries 6.26 N r.m.s. Summing the harmonics
  that fall inside each octave band gives 139.4 / 142.4 / 145.4 / 148.4 /
  151.4 / 154.4 dB re $10^{-6}$ N from 31.5 Hz to 1 kHz, reproducing the first
  six cells of Table F.1 to within 0.5 dB; the 2 kHz and 4 kHz cells sit below
  that line, which is the roll-off the standard itself flags with "up till
  about 1000 Hz". Re 1 pN the same cells would describe forces of tens of
  micronewtons, which no impact machine produces. **(c) The companion
  standard.** UNE-EN 15657:2018 Formula (15), which is where the structure-borne
  source data of Annex D comes from in the first place, writes the same
  force-to-power conversion "in dB re $F_0 = 10^{-6}$ N", and $10^{-6}$ N is
  the preferred reference force of ISO 1683.
- **Evidence:** verified on PDF pages 61 and 62 (printed pp. 59 and 60) of
  BS EN 12354-5:2009, carrying clause F.4.2 with the symbol list of
  Formula (F.9), the closed form, the whole of Table F.1 and the symbol list
  of Formula (F.11) with its $10^{-9}$ m/s velocity reference; and on PDF
  pages 45, 48 and 50 (printed pp. 43, 46 and 48) of the same edition,
  carrying Formulae (D.5a), (D.9a) and (D.10a).
- **Library behaviour:** ships the printed cells unchanged and documents them
  re $10^{-6}$ N. `tapping_machine_force_level` returns the eight values of
  Table F.1, `tapping_machine_force_level_estimate` the closed form and
  `tapping_machine_characteristic_power_level` Formula (D.9a) as printed;
  `test_table_f1_is_referred_to_1e_6_newton_not_1_piconewton` pins the reading
  against the mechanics of the machine.
- **Status:** unreported.

## EN 12354-5:2009, Figure D.3 Key (three curves under one symbol)

- **Location:** Annex D, the Key of Figure D.3 (printed p. 47).
- **The print:** three key rows, each labelled with the same symbol:
  $L_\mathrm{Ws,c,A} = 124\ \text{dB}$, $L_\mathrm{Ws,c,A} = 119\ \text{dB}$ and
  $L_\mathrm{Ws,c,A} = 102\ \text{dB}$.
- **The problem:** the figure's own caption reads "Structure-borne sound power
  for the ISO-tapping machine: characteristic source power, installed power on
  a wooden floor and installed power on a concrete floor; the A-weighted power
  level is also indicated". Only the first curve is a characteristic power; the
  other two are installed powers and their A-weighted totals are
  $L_\mathrm{Ws,inst,A}$. The plotted curves settle the assignment: the first
  is flat at about 114.5 dB re 1 pW, which is the frequency-independent
  Formula (D.9a) result for the tapping machine, while the other two rise with
  frequency and lie below it, the concrete floor lowest, as
  $L_\mathrm{Ws,c} - D_{\mathrm{C},i}$ requires.
- **Evidence:** verified on PDF page 49 (printed p. 47) of BS EN 12354-5:2009,
  the page carrying Figure D.3 with its Key and its caption.
- **Library behaviour:** none required; no value is read from Figure D.3.
  `test_formula_d9a_is_flat_at_about_115_db_per_third_octave` pins the flat
  characteristic curve that the first key row belongs to.
- **Status:** unreported.

## ISO 12354-1:2017 Table L.3 / ISO 12354-2:2017 Table G.3 (perimeter sums)

- **Location:** the input-data block below Table L.3 (printed p. 81) and the
  identical block below Table G.3 (printed p. 38), which lists the perimeter
  absorption sum $\sum l_k \alpha_k$ of Formula (C.1) for the worked example.
- **The print:** one value per element *type*: separating floor 2,364 m
  ($S = 20\ \text{m}^2$), external wall 2,375 m ($S = 11\ \text{m}^2$),
  internal wall 1,840 m ($S = 13{,}75\ \text{m}^2$).
- **The problem:** Formula (C.1) needs one sum per *element*, and the example
  has five elements with three different areas. Only two of the three printed
  values reproduce the columns they are supposed to drive: 2.375 m with
  $S = 11\ \text{m}^2$ gives external wall 1 exactly, and 1.840 m with
  $S = 13.75\ \text{m}^2$ gives internal wall **2** exactly. The separating
  floor's printed 2.364 m does not reproduce its own column at any band
  (0.074 9 against the printed 0.083 1 at 50 Hz, 0.026 4 against 0.029 0 at
  500 Hz); 2.659 m does, at every band. The two elements with no printed value
  need 2.548 m (external wall 2, $S = 13.75\ \text{m}^2$) and 1.636 m
  (internal wall 1, $S = 11\ \text{m}^2$).
- **Evidence:** all five sums re-derived from Formula (C.4),
  $\alpha_k = \sum_j \sqrt{f_{c,j}/f_\text{ref}}\ 10^{-K_{ij}/10}$, over the
  example's own junction geometry with the unrounded Annex E indices: 2.659 /
  2.375 / 2.548 / 1.636 / 1.839 m. The derivation returns the two printed
  values that are self-consistent with their own columns (2.375 m, and 1.839 m
  against the printed 1.840 m) and supplies the three that are missing or
  wrong, and every $\eta_\text{tot,situ}$ column of Table L.3 / G.3 then
  reproduces to $5 \cdot 10^{-5}$. The printed values applied to the wrong
  element of the same type miss by far more than that rounding: 2.375 m on
  external wall 2 gives 0.108 5 against the printed 0.114 9 at 50 Hz, and
  1.840 m on internal wall 1 gives 0.085 0 against 0.077 0.
- **Library behaviour:** `in_situ_total_loss_factor` takes $\sum l_k \alpha_k$
  as an input and `perimeter_absorption_coefficient` implements Formula (C.4);
  the Annex L fixture derives all five sums that way rather than using the
  printed block, and says so
  ([`tests/building/prediction/test_detailed_model.py`](https://github.com/jmrplens/phonometry/blob/main/tests/building/prediction/test_detailed_model.py)).
- **Status:** unreported.

## ISO 12354-1:2017 Table L.3 / ISO 12354-2:2017 Table G.3 (external wall ηint)

- **Location:** the same input-data block, external-wall line.
- **The print:** $\eta_\text{int} = 0{,}013$ for the 365 mm autoclaved aerated
  concrete external walls.
- **The problem:** the example's own element specification, and Annex B Table
  B.3 for autoclaved aerated concrete, give 0.012 5. Only 0.012 5 reproduces
  the tabulated $\eta_\text{tot,situ}$: at 500 Hz Formula (C.1) gives
  $0.012\,5 + 0.001\,41 + 0.034\,57 = 0.048\,5$, the printed value,
  where 0.013 would give 0.049 0.
- **Evidence:** term-by-term recomputation of Formula (C.1) for both external
  walls at every band with each candidate $\eta_\text{int}$.
- **Library behaviour:** the Annex L fixture uses 0.012 5.
- **Status:** unreported.

## ISO 12354-1:2017, Table L.4 (second path block labelled 2d)

- **Location:** Annex L, Table L.4 (printed p. 82), the right-hand block
  headed "Transmission path 2d".
- **The print:** the block gives $\alpha_{i,\text{situ}}$ = 6,3 to 14,1,
  $D_{\mathrm{v},ij,\text{situ}}$ = 11,0 to 13,6 and $R_{ij}$ = 43,9 to 84,6 dB.
- **The problem:** those are the numbers of path **4d** (internal wall 2 to
  the separating floor), not of path 2d (external wall 2). Table L.1 of the
  same annex prints the whole $R_{4d}$ column, 43.9 to 84.6 dB, and the
  block's $R_{ij}$ column is that column cell for cell. What settles it band
  by band is the other two columns, which cannot be confused: external wall 2
  has $\alpha_{i,\text{situ}} = 10.3\ \text{m}$ at 50 Hz
  ($S = 13.75\ \text{m}^2$, $\eta_\text{tot} = 0.114\,9$) while internal
  wall 2 has 6.3 m ($\eta_\text{tot} = 0.070\,3$), the printed value; and
  $D_{\mathrm{v},ij,\text{situ}}$ follows the floor-to-internal-wall $K_{ij}$ of 8.8
  dB, which gives 11.0 to 13.6 dB, not the floor-to-external-wall 6.4 dB,
  which gives 9.6 to 11.9 dB.
- **Evidence:** independent recomputation of Formulae (10), (11) and (15) for
  both candidate paths at every band. Path 4d reproduces all three columns of
  the block, $\alpha_{i,\text{situ}}$ to 0.05 m and $D_{\mathrm{v},ij,\text{situ}}$ and
  $R_{ij}$ to 0.05 dB, which is the printed resolution. Path 2d departs from
  the block's $R_{ij}$ column by 0.1 dB to 7.0 dB depending on the band, and
  comes closest between 100 Hz and 160 Hz (0.5 / 0.5 / 0.1 dB), so $R_{ij}$
  alone does not identify the path over those bands; $\alpha_{i,\text{situ}}$
  (10.3 against 6.3 m at 50 Hz) and $D_{\mathrm{v},ij,\text{situ}}$ (1.4 dB to 1.7 dB
  apart in every band) do.
- **Library behaviour:** the test that asserts the block builds it as path 4d
  and names the mislabelling.
- **Status:** unreported.

## ISO 12354-1:2017, Table L.1 (non-integer weighted ratings)

- **Location:** Annex L, Table L.1 (printed p. 79), the $R_w$ row and the
  sentence below it, and the corresponding $L_{n,w}$ row of ISO 12354-2:2017
  Table G.1.
- **The print:** the $R_w$ row gives one decimal for every path (75,1 / 84,5 /
  70,6 / … and 57,8 in the total column) while the sentence immediately below
  states $R'_w\,(C\,;\,C_\text{tr}) = 57{,}9\ (-2\,;\,-8)\ \text{dB}$.
- **The problem:** ISO 717-1 rates by shifting the reference curve **in 1 dB
  steps**, so a weighted rating is an integer; the printed one-decimal values
  are the reference curve shifted *continuously* until the sum of unfavourable
  deviations equals exactly 32.0 dB. The airborne $R_w$ row of Table L.1
  *truncates* that continuous value to one decimal while the sentence below it
  rounds, which is why the same quantity appears twice as 57.8 and 57.9; the
  impact $L_{n,w}$ row of Table G.1 rounds instead (29.58 prints as 29.6 and
  40.98 as 41.0), so the truncation is a property of the airborne row only.
  The spectrum adaptation terms inherit the offset: with the ISO 717-1 rating
  of 57 dB they are $C = -1$ and $C_\text{tr} = -7$, and the printed (−2 ; −8)
  is exactly the pair shifted by the same 0.86 dB.
- **Evidence:** a continuous-shift solve of the ISO 717-1 reference curve
  against the printed per-band spectra reproduces every printed value in both
  rows ($R_{Dd}$ 75.12 against 75.1; $R_{D1}$ 84.54 against 84.5; $R_{11}$
  70.66 against 70.6; the total 57.86 against 57.8 / 57.9; on the impact side
  $L_{n,Df1}$ 29.58 against 29.6 and the total 40.98 against 41.0), whereas
  the ISO 717-1 1 dB-step ratings of the same spectra are 75, 84, 70 and 57
  dB. Verified on PDF page 85 (printed p. 79) of ISO 12354-1:2017.
- **Library behaviour:** `weighted_rating` / `weighted_impact_rating`
  implement ISO 717-1/-2 as written, so the detailed model returns
  $R'_w = 57\ \text{dB}$ and $L'_{n,w} = 41\ \text{dB}$ ($C_I = 2$) for the
  example; the test pins those and documents the printed values.
- **Status:** unreported.

## ISO 12354-2:2017, Table G.1 (50 Hz to 80 Hz flanking columns)

- **Location:** Annex G, Table G.1 (printed p. 36), the four $L_{n,Df}$
  columns, 50 Hz, 63 Hz and 80 Hz rows.
- **The print:** $L_{n,Df1}$ = 47,3 / 44,9 / 46,2 dB.
- **The problem:** Table G.4 of the same annex prints the same path Df for
  external wall 1, from the same inputs, as 47.8 / 45.9 / 47.0 dB. The two
  tables cannot both be right, and from 100 Hz upwards they agree exactly.
- **Evidence:** Formula (12) evaluated from the annex's own Table G.3 columns
  ($L_{n,\text{situ}}$, $R_\text{situ}$) and the Table G.4
  $D_{\mathrm{v},ij,\text{situ}}$ and $\Delta L_\text{situ}$ columns gives 47.80 /
  45.85 / 46.95 dB, reproducing the printed 47.8 / 45.9 / 47.0 of Table G.4 to
  0.05 dB and Table G.1 only from 100 Hz upwards. Carrying the same
  recomputation through the whole chain puts external wall 2 low by 0.5 dB to
  1.0 dB over the same three bands and the two internal walls low by up to 0.5
  dB at 50 Hz and 63 Hz (their 80 Hz cells agree). From 100 Hz upwards no
  flanking column deviates by more than 0.15 dB. Correcting the affected cells
  raises the printed total $L'_n$ only slightly: 58.6 to 58.7 dB at 50 Hz,
  57.0 to 57.2 dB at 63 Hz, 55.9 to 56.1 dB at 80 Hz.
- **Library behaviour:** the test asserts Table G.4 in full, the Table G.1
  direct column over the whole range, and the Table G.1 flanking columns from
  100 Hz upwards, naming the disagreement.
- **Status:** unreported.

## ISO 12354-2:2017, Table G.8 (junction Kij and m'i)

- **Location:** Annex G, Table G.8 (printed p. 40), the internal wall to
  external wall rigid T junction.
- **The print:** row "Int. wall 1/2 - Ext. wall 1/2" gives
  $K_{ij} = 6{,}6\ \text{dB}$; the row below it, "Ext. wall 1/2 - Ext. wall
  1/2", gives $m'_i = 2{,}19\ \text{kg/m}^2$.
- **The problem:** two independent misprints. The rigid-T corner branch
  $K_{12} = 5.7 + 5.7 M^2$ with $M = \log_{10}(360/219) = 0.215\,6$
  gives 5.97, i.e. **6.0**, and ISO 12354-1:2017 Table L.8 prints 6.0 for the
  identical junction of the identical example. And the external wall's mass
  per unit area is $219.0\ \text{kg/m}^2$ throughout the example, not 2.19
  (a factor 100).
- **Evidence:** Annex E evaluation of the corner branch; the same table's own
  other rows and the whole of ISO 12354-1 Annex L use
  $219.0\ \text{kg/m}^2$. Verified on PDF page 46 (printed p. 40) of ISO
  12354-2:2017, whose Table G.8 mass columns are headed `m'i` and
  `m'orthogonal`, and PDF page 89 (printed p. 83) of ISO 12354-1:2017.
- **Library behaviour:** uses 6.0 dB and $219.0\ \text{kg/m}^2$.
- **Status:** unreported.

## ISO 12354-2:2017, Table G.6 (mislabelled row)

- **Location:** Annex G, Table G.6 (printed p. 40), internal wall to
  separating floor rigid cross junction.
- **The print:** a row labelled "Ext. wall 1/2 – Int. wall 1/2" with `m'i` =
  360,0, `m'orthogonal` = 484,0 and $K_{ij} = 11{,}0\ \text{dB}$.
- **The problem:** Table G.6 describes the *internal wall to separating floor*
  cross junction; no external wall meets it. The masses and the value are
  those of the in-line internal-wall path, and ISO 12354-1:2017 Table L.6
  prints the same row correctly as "Int. wall 1/2 - Int. wall 1/2".
- **Evidence:** the rigid-cross through branch $8.7 + 17.1 M + 5.7 M^2$
  with $M = \log_{10}(484/360)$ gives 10.99, the printed 11.0, for the
  internal wall. Verified on PDF page 46 (printed p. 40) of ISO 12354-2:2017
  and PDF page 89 (printed p. 83) of ISO 12354-1:2017.
- **Library behaviour:** treats the row as the internal-wall in-line path.
- **Status:** unreported.

## ISO 12354-1:2017 Table L.10 / ISO 12354-2:2017 Table G.10 (element label)

- **Location:** the simplified-model input table of both parts, fourth row:
  Table L.10 (printed p. 84) and Table G.10 (printed p. 41).
- **The print:** ISO 12354-1 prints "Internal wall 4 (F = f = 4)"; ISO 12354-2
  prints "Internal wall 4 (f4)": the two parts label the row differently, and
  an earlier revision of this entry quoted the Part 1 form for both.
- **The problem:** the example has two internal walls; the element indexed
  $F = f = 4$ is internal wall **2**
  ($5.00\ \text{m} \times 2.75\ \text{m}$, $S = 13.75\ \text{m}^2$), as
  the detailed-model tables of the same annexes label it.
- **Evidence:** the row's own $S = 13.75\ \text{m}^2$ and
  $l_{ij} = 5.0\ \text{m}$ match internal wall 2 of Table L.1 / G.1.
  Verified on PDF page 90 (printed p. 84) of ISO 12354-1:2017 and of PDF page
  47 (printed p. 41) of ISO 12354-2:2017, with the detailed-model column
  labels read on PDF page 85 (printed p. 79) of ISO 12354-1:2017 and of PDF
  page 42 (printed p. 36) of ISO 12354-2:2017.
- **Library behaviour:** none needed; the numbers are unaffected.
- **Status:** unreported.

## ISO 12354-1:2017, Table D.1 (1 600 Hz covered by two rows)

- **Location:** Annex D, Table D.1 (printed p. 39), which reads the weighted
  sound reduction index improvement of an interior lining off its resonance
  frequency.
- **The print:** the last two rows are "630 to 1 600 -> -10" and "1 600 <= f0
  <= 5 000 -> -5".
- **The problem:** 1 600 Hz belongs to both rows, with different values, and
  Clause D.2.2 requires $f_0$ to be "rounded to the centre frequency of the
  one-third-octave band in which fo falls", so 1 600 Hz is a value the table
  is actually read at rather than an unreachable edge. Because the rounding is
  mandatory, the ambiguity is not a single point: every raw resonance
  frequency in the 1 600 Hz band, that is from 1 412.5 Hz to 1 778.3 Hz (ISO
  266 band edges), lands on it. Every other boundary in the table is a
  distinct band centre (200, 250, 315, 400, 500 Hz), and no other pair of rows
  overlaps.
- **Evidence:** the printed table itself, on PDF page 45 (printed p. 39) of
  ISO 12354-1:2017: the two rows are separately ruled and share the endpoint
  verbatim, "630 to 1 600" and "1 600 <= f0 <= 5 000". Neither row can be
  discarded, because 630 Hz to 1 250 Hz has no other entry and 2 000 Hz to
  5 000 Hz has none either. The predecessor edition gives the earlier,
  unambiguous reading: EN 12354-1:2000 Table D.3, verified on PDF page 43
  (printed p. 41) of that edition, prints the same pair of rows as "630 -
  1 600 -> -10" and "> 1 600 -> -5", strictly greater, so in 2000 exactly
  1 600 Hz took -10 dB with nothing to decide. The 2017 rewrite replaced ">
  1 600" with "1 600 <= f0 <= 5 000" while leaving "630 to 1 600" untouched,
  which is what creates the overlap; what the rewrite intended at the shared
  endpoint the text does not say.
- **Library behaviour:** `weighted_lining_improvement` returns the more
  conservative -10 dB at exactly 1 600 Hz and -5 dB above it, the 2000
  reading, with the ambiguity named in the docstring and pinned in
  [`tests/building/prediction/test_resilient_layers.py`](https://github.com/jmrplens/phonometry/blob/main/tests/building/prediction/test_resilient_layers.py).
- **Status:** unreported.

- **Related, not an erratum:** NOTE 1 of the same table sets a floor of 0 dB
  on the 30 Hz to 160 Hz branch $74.4 - 20\log_{10}(f_0) - R_w/2$. Inside
  the validity box Clause D.2.2 states for the table
  ($30\ \text{Hz} \le f_0 \le 160\ \text{Hz}$,
  $20\ \text{dB} \le R_w \le 60\ \text{dB}$) the branch never reaches it: its
  minimum is $74.4 - 20\log_{10}(160) - 60/2 = 0.32\ \text{dB}$. The floor
  is therefore inactive for every input the table is stated for, but it was
  not always: the 2000 edition tabulated the low branch as four discrete rows
  ending in "160 -> 28 - Rw/2", whose minimum is $28 - 60/2 = -2\ \text{dB}$,
  so NOTE 1 was operative there. The 2017 continuous fit sits 2.3 dB above it
  at that corner and left the note vestigial. The library keeps the floor
  because the note is still printed.
