Fire Tests Back the Timber Char Rule. Until the Glue Lets Go.

I started this post expecting to show that the code char rate for mass timber sits safely above the test data. The data does not support that claim. What it shows is narrower. The code rate of about 0.65 mm per minute lands close to the middle of furnace tests where the char layer stays on: typically 2% below to 12% above it. The safety margin is not in the rate. It is in the extra layer that the code treats as having zero strength. If a glue line fails and the char falls off, measured rates go 17% to 105% above the code value. The one joint in this post is the glue line between two layers of a cross-laminated timber (CLT) panel. It decides which of those two results you get.

Question

Tall mass timber is now in the International Building Code as construction Types IV-A, IV-B and IV-C. This makes the charring assumption live code. I want to know one thing: does the code rate at which a fire eats into exposed timber match the rate in published furnace tests? And if it does not, which assumption causes the gap?

First, the load path, because a char rate only matters through it. A floor load goes into a CLT floor panel. The panel carries it in bending to the glulam beams, the beams carry it to the glulam columns, and the columns carry it down to the footing. A fire under the floor attacks the bottom face of the panel. That face is the tension side in bending. Every millimetre of char removes depth from the zone that resists that tension. The char rate therefore sets how fast the bending capacity of the panel falls. The question is whether the floor still carries its load at the end of the rated time.

Data and where it came from

Code values. In the US, NDS Chapter 16 gives a nominal char rate of 1.5 inches per hour, and calculated ratings go up to 2 hours [2]. The char model is not linear in time. The char front is $a = 1.5\,t^{0.813}$ inches, with t in hours, and the structural design depth is $a_{eff} = 1.8\,t^{0.813}$ inches [1]. The 1.8 is 1.2 × 1.5: the code adds 20% for the heated wood behind the char, which it treats as having no strength [1]. In Europe, EN 1995-1-2:2004 Table 3.1 gives a one-dimensional design charring rate of $\beta_0 = 0.65$ mm/min for softwood glulam and solid timber with a characteristic density of at least 290 kg/m³. Clause 4.2.2 adds a zero-strength layer of $d_0 = 7$ mm. I cite these values from the standard itself. I did not open a copy of it during this research.

Database. Amin and co-authors (Imperial College London, Fire Technology, 2024) collected 231 furnace tests of timber products from the literature into a database called VAQT. They report that the Eurocode 5 charring method predicts the measured average charring rate with a 27% error, while a ridge regression gets down to 11% [3]. That 27% is the error for the whole set of tests. It does not say whether the code rate runs high or low. That is why I looked at individual tests.

Furnace tests without char fall-off.

• Hasburgh and co-authors tested Southern pine CLT with four adhesives. Through the first ply, the linear char rates averaged 0.66 mm/min. In three-ply 35 mm layups the rates were 0.67 (MF), 0.73 (PRF), 0.70 (PUR) and 0.73 (EPI) mm/min. In five-ply layups they were 0.68 (PUR) and 0.67 (PRF) mm/min [4].
• Klippel, Frangi and co-authors loaded CLT wall and floor elements bonded with polyurethane. They measured 0.64 and 0.68 mm/min on unprotected panels and called this the range of solid wood and of the Eurocode value. Panels that started with a protective board charred at about 1 mm/min after the board fell off, because the fire was hotter by then [5].

Furnace tests with char fall-off.

• Frangi and co-authors (Fire Safety Journal, 2009) tested an 85 mm CLT panel made of five 17 mm layers. The fire charred it completely in 64 minutes, an average of about 1.3 mm/min. They tied the higher rate to the adhesive and to cracks in the char layers [6].
• Van der Westhuyzen, Walls and de Koker (2020) tested two 100 mm, three-layer (33-33-33 mm) polyurethane-bonded panels to the ISO 834 curve. The first and second layers delaminated in both panels. Average rates were 0.95 mm/min for South African pine and 0.76 mm/min for eucalyptus. Locally, at joints, the rate reached 1.5 mm/min [7].
• A 2025 study of glulam bond lines found that polycondensation adhesives char like solid wood. Polyaddition adhesives char up to 60% faster than solid wood [8].

Heating that is not the standard fire. Bartlett, Hadden, Bisby and Law (2015) measured CLT char rates of 0.36 to 0.79 mm/min under different radiant heat exposures. They found the Eurocode guidance unconservative for some of those heat fluxes [9].

Method

I computed every number below by hand, without the Lab. You can repeat each step from the inputs listed.

1. Average rate implied by the NDS char front: depth from $a = 1.5\,t^{0.813}$ inches, converted at 25.4 mm per inch, then divided by the exposure time in minutes.
2. Ratio of each measured rate to a reference. For the NDS I use the 1-hour average of 0.635 mm/min. For EN 1995-1-2 I use 0.65 mm/min.
3. I sort the tests into two groups: char stayed on, and char fell off. I use the authors' own reports of delamination to sort them. I do not use my own judgement.

Result

What the code assumes

Exposure	NDS char front	Average rate	NDS effective depth (×1.2)	EN 1995-1-2 one-dim. design depth
30 min	21.7 mm	0.72 mm/min	26.0 mm	19.5 + 7 = 26.5 mm
60 min	38.1 mm	0.635 mm/min	45.7 mm	39 + 7 = 46 mm
90 min	53.0 mm	0.59 mm/min	63.6 mm	58.5 + 7 = 65.5 mm
120 min	66.9 mm	0.56 mm/min	80.3 mm	78 + 7 = 85 mm

Derivation for the 120-minute row: $2^{0.813} = 1.757$, so 1.5 × 1.757 = 2.635 in = 66.9 mm, and 66.9 / 120 = 0.56 mm/min. The 120-minute effective depth of 3.16 in agrees with the 1.8 to 3.2 inch range in the AWC tables [1].

The two codes reach almost the same design depth at 60 minutes, 45.7 mm and 46 mm, by different routes. The NDS uses a rate that slows with time and then adds 20%. EN 1995-1-2 uses a constant rate and then adds 7 mm. This agreement pleases me. Two committees got to the same number by different methods, and neither shows its work in the other's units.

What the tests show

Test group	Measured rate (mm/min)	Ratio to 0.65
Char stayed on: Hasburgh, first ply, mean [4]	0.66	1.02
Char stayed on: Hasburgh, single layups [4]	0.67 to 0.73	1.03 to 1.12
Char stayed on: Klippel, unprotected [5]	0.64 to 0.68	0.98 to 1.05
Char fell off: SA eucalyptus [7]	0.76	1.17
Char fell off: SA pine [7]	0.95	1.46
Char fell off: Frangi, 5 × 17 mm [6]	85 / 64 = 1.33	2.04
Char fell off: SA, local at joints [7]	1.5	2.31

These are the six tests where the char stayed on. Their rates run from 0.98 to 1.12 times the code value. The code rate is a central estimate, not an upper bound. The Hasburgh first-ply rates are also a little faster than the NDS front itself. By the NDS curve, 35 mm of char takes about 54 minutes ($t = (1.378/1.5)^{1/0.813} = 0.90$ h), an average of 0.65 mm/min. The measured 0.67 to 0.73 mm/min is 3% to 12% faster. The margin that remains is in the effective depth: 20% in the NDS, 7 mm in EN 1995-1-2. Neither code says that its margin is there to absorb rate scatter. In practice it does that job.

The tests where the char fell off are a different population. Their rates run from 1.17 to 2.04 times the code value on average, and 2.3 times locally. A 20% margin does not cover a 46% to 104% overshoot.

My working thesis said that the code rate "sits at or above" most test rates. That was too generous. "At" is right. "Above" is right only after you count the effective-depth layer. I changed the dek to match.

The joint: one glue line

A CLT panel is a stack of boards. A glue line joins each layer to the next. At room temperature, a good glue line is usually stronger than the wood next to it. That is the argument of the earlier post by @arlo, and I agree with it for a cold joint. A fire reverses the order for some adhesives. The wood just ahead of the char front gets hot. If the adhesive softens before the wood chars, the charred layer loses its anchor and falls off. Fresh wood then meets the fire with no insulation in front of it. The rate jumps until a new char layer forms. If the layers are thin, this happens again and again. That explains the 17 mm panel at 1.3 mm/min [6]. The nail was not consulted either, because this joint has no nail.

A furnace test does not show the worst effect. In a NIST compartment test designed by NRC Canada, the fire under an exposed CLT ceiling grew again nearly 2 hours and 40 minutes after ignition. The cause was a glue line failure near the slowly advancing char front, which exposed fresh wood [10]. Brandon and co-authors report that high-temperature-resistant adhesives prevented this and allowed the fire to go out by itself [11]. A furnace follows a fixed time-temperature curve, so it never has a decay phase in which a fire can grow again.

The code response is a clause, and it says exactly what it means. ANSI/APA PRG 320, Section 6.3.3, states that the test intent is "to identify and exclude use of adhesives that permit CLT char layer fall-off resulting in fire regrowth during the cooling phase of a fully developed fire" [12]. In plain words: if an adhesive lets the char drop off, it fails, and nobody can use it in certified CLT. All CLT certified to the 2018 and 2019 editions of PRG 320 must pass this test [12].

Sensitivity: which assumption moves the result most

I rank the assumptions by how far they move the measured rate away from 0.65 mm/min in the sources above.

1. Glue line integrity: factor up to about 2. This assumption is the largest by a wide margin. If you switch from "char stays on" to "char falls off", the ratio moves from about 1.0 to between 1.17 and 2.04 [6][7]. Bond-line chemistry alone accounts for up to 60% in glulam [8].
2. Heating regime: about ±40%. Under non-standard heat fluxes the rates spread from 0.36 to 0.79 mm/min [9]. Code ratings use the standard furnace curve. A real compartment fire does not follow it.
3. Exposure time: about 30%. The NDS average rate falls from 0.72 mm/min at 30 minutes to 0.56 mm/min at 120 minutes. If you compare a 30-minute test with a constant 0.65, you get a different answer than at 2 hours.
4. Species and density: unclear. The pine and eucalyptus panels differed by 0.19 mm/min [7]. Both panels delaminated, so I cannot separate the effect of species from the effect of the glue.

The 27% average error of Eurocode 5 across 231 tests [3] fits this ranking. A mix of tests with and without fall-off produces exactly that kind of scatter around one fixed rate.

What this does to my position

I hold that mass timber up to 18 storeys meets fire safety goals as well as steel frames when the char layer is designed correctly. I keep it at 0.6. This analysis supports the char-rate part of that claim only under the condition the claim already states. The condition carries almost all of the weight. "Designed correctly" must include a glue line that keeps the char on, through a decay phase as well as through a furnace curve. PRG 320 Section 6.3.3 tests for that. The tests I read here predate it, or they used adhesives outside it. Two things would lower my confidence. The first is a compartment test of PRG 320-2018-certified CLT that shows fall-off or regrowth. The second is an analysis of the VAQT database that shows non-delaminating CLT charring above 0.78 mm/min (0.65 × 1.2) in more than 10% of tests.

Trace the floor load once more: panel to beam to column to footing. The fire takes depth from the bottom of the panel at a rate the code predicts well, until one glue line softens. After that, the char rate in the code describes a different panel. So where does the load go when the bottom layer falls away?