The Pyrolytic Reservoir

Why Single-Draw Testing Underestimates Real-World Cannabis Vaporizer Hardware Risk

Author: Matt Macosko, Founder, Cannabis Device Safety Institute Affiliation: Cannabis Device Safety Institute, Arcata, California Status: Founding draft, v0.4 (May 2026) Suggested citation: Macosko, M. The Pyrolytic Reservoir: Why Single-Draw Testing Underestimates Real-World Cannabis Vaporizer Hardware Risk. CDSI Working Paper P-003. 2026. Companion documents: P-001 The Hardware Vacuum; P-002 Loaded-State Off-Gas Analysis; CDSI-001 Protocol, v1.0.

Status of the evidence — read before citing this paper. This paper presents a methodological argument and qualitative predictions. It presents no empirical data, because CDSI has not yet generated any. The cycle-life signatures described below are what the protocol is expected to reveal, reasoned from physical first principles and analogous work in adjacent fields — they are hypotheses, not results. The empirical companion paper is planned for the year-two publishing cycle and requires a testing bench the Institute does not yet have. Nothing here should be cited as a measurement.

Abstract

The small body of published cannabis vaporizer hardware testing — including the loaded-state off-gas (LSO) protocol introduced in CDSI Working Paper P-002 — overwhelmingly characterizes devices in their out-of-box, fresh-load state: one to a small number of draws on a freshly cleaned heater holding a freshly dispensed concentrate. Real consumer use bears no resemblance to this. A typical concentrate cartridge is drawn on dozens to hundreds of times before disposal; a refillable atomizer or banger is typically reused across hundreds to thousands of sessions before retirement, often without intermediate cleaning. Across that lifecycle, residual concentrate matter accumulates on the heating element as a charred deposit — what CDSI has named the pyrolytic reservoir. We argue, drawing on physical first principles and analogous work in combustion engineering, food-processing equipment characterization, and pharmaceutical inhaler regulation, that the off-gas profile of a device with an established reservoir is meaningfully different from the same device’s profile when fresh, and that this difference is the dominant unmeasured component of consumer hardware exposure. We specify a cycle-life extension of the CDSI-001 protocol that characterizes a device at draw 1, 50, 200, and 500 of a single load-cycle regime. We predict, qualitatively, the signatures such a protocol should reveal. The empirical companion paper, presenting data from this protocol, is planned for the year-two CDSI publishing cycle. The methodological argument is the prerequisite. This paper is also the scientific justification for the Tier 3 (ongoing-batch-testing) level of the CDSI tier-graded certification model proposed in P-006.

Keywords: vaporizer cycle-life, pyrolytic residue, ongoing-batch testing, hardware certification, off-gas analysis, harm reduction

1. Introduction

The Cannabis Device Safety Institute’s founding position paper (P-001) named the institutional condition we call the Hardware Vacuum: the absence of any body whose responsibility is the safety of cannabis consumption hardware as such. P-001 made the case for why the vacuum exists; it did not, in detail, lay out what fills it. The companion methodological paper, P-002, took the first concrete step by specifying loaded-state off-gas (LSO) analysis as the new minimum for cannabis vaporizer hardware characterization. P-002’s central empirical claim is that compounds produced by a heater operating with concentrate present are not the same as compounds produced by a heater operating empty; dry-fire testing therefore systematically under-reports consumer exposure.

The present paper extends this argument in a direction that, from where we sit, is even less excusable than the dry-fire gap. P-002’s baseline LSO methodology characterizes a fresh device with a fresh load. A great deal of testing in adjacent fields stops there. We argue that stopping there mischaracterizes the real exposure profile of a device that has been used for any meaningful duration, and that the dominant source of unmeasured chemical exposure in consumer cannabis vaporizer use is not the device when it is new — it is the device after the user has put fifty, two hundred, or five hundred draws through it without intermediate cleaning.

The paper proceeds as follows. Section 2 describes the pyrolytic reservoir hypothesis from first physical principles. Section 3 surveys analogous phenomena in three adjacent fields and explains why their treatment in those fields is mature and standardized while in cannabis hardware it is essentially absent. Section 4 enumerates why current single-draw testing protocols cannot capture reservoir effects, even when LSO compliant. Section 5 specifies the proposed cycle-life extension of CDSI-001. Section 6 enumerates the qualitative signatures we expect such a protocol to reveal, marking the predictions clearly as predictions. Section 7 argues that the operational consequence of an established reservoir effect is a tier of certification that is necessarily ongoing rather than one-time, justifying CDSI’s Tier 3 model. Section 8 catalogs the limitations of this argument and the empirical questions that need to be resolved before its policy implications can be acted on with confidence.

We want to be explicit, at the front: this paper is methodological and observational. It does not present new analytical data. The empirical paper — testing three devices across the specified cycle-life regime — is on the CDSI publishing pipeline for the year-two cycle, contingent on lab partnership and funding. We are publishing the methodological argument first because, in our reading of the field, the absence of cycle-life characterization is not currently a question being asked. Naming the question is the prerequisite to answering it.

2. The pyrolytic reservoir: physical basis

A cannabis concentrate vaporizer operates by raising a small quantity of organic concentrate from ambient temperature to a target operating temperature — typically 350–700 °F at the heating element surface, depending on device class and concentrate matrix. At those temperatures, the desired output is the volatilization of cannabinoid and terpene compounds into an inhalable aerosol.

Volatilization, however, is not the only thermal pathway. At any given operating temperature, the concentrate matrix in contact with the heater experiences a distribution of fates:

Volatilization at the design point. Cannabinoids and the lower-boiling terpenes leave the concentrate in vapor phase and are inhaled.
Thermal degradation. A fraction of the matrix decomposes rather than volatilizing intact. Cannabinoids and terpenes have known degradation pathways under heat, and a portion of the apparent “vapor” delivered to the consumer is in fact degradation products of the source compounds.
Carbonization. A fraction of the matrix polymerizes and chars onto the heating element surface. This residue is functionally a thin layer of organic carbon, with embedded inorganics from the original concentrate, deposited directly on the heater material.
Migration into the substrate. In porous heater architectures (ceramic donut, ceramic-coil 510 cartridges), a fraction of the matrix is wicked into the substrate matrix and remains there between sessions.

The first of these is the design intent. The second is partially documented in the analytical chemistry literature on cannabis vaporization (see e.g. Meehan-Atrash et al., 2017; Troutt and DiDonato, 2017). The third and fourth — the carbonized layer and the substrate-resident matter — are the pyrolytic reservoir.

The reservoir’s central property is that it persists between sessions. It does not volatilize fully on the draw that created it; were that the case, there would be no visible “char” on the heater after extended use, and the field knows from inspection that there always is. Instead, the reservoir accumulates over the device’s lifetime, until either the user cleans it, replaces the device, or — in the case of disposable cartridges — exhausts the original concentrate load and discards the device with the reservoir intact.

Once present, the reservoir is itself subject to thermal cycling. At each subsequent draw, the heater rises again to operating temperature; the reservoir, being directly on the heating element, sees this temperature first. Three things can happen:

The reservoir partially re-volatilizes. Compounds that did not fully vaporize on the original draw — typically heavier terpene degradation products, oxidized cannabinoids, and matrix-derived semi-volatiles — release into the airstream of the current draw. The consumer inhales them in addition to the compounds released by the fresh concentrate.
The reservoir undergoes secondary pyrolysis. Compounds in the carbonized layer are exposed to operating temperatures repeatedly, and progressively decompose into smaller, more volatile, often more toxic decomposition products. This pathway is the same one by which combustion-engine deposits produce polycyclic aromatic hydrocarbons over engine lifetime — except that in this case, the decomposition products are inhaled rather than passed through a catalytic converter.
The reservoir alters the heat-transfer geometry of the heating element itself. A clean heater conducts heat to the concentrate by direct contact (porous wicking architectures) or radiation (open-coil and banger architectures). A reservoir-coated heater interposes a layer of carbon between the heating element and the fresh concentrate. The thermal conductivity of carbonized organic matter is meaningfully lower than that of the heater substrate; the heater therefore must run hotter, or longer, to deliver the same vaporization energy to the fresh concentrate. Higher heater temperatures, in turn, accelerate degradation pathways elsewhere in the system, including in the reservoir itself.

These three pathways compound. The third creates the conditions that intensify the first two. A device with a substantial reservoir is, in effect, operating at higher functional temperatures and pyrolyzing material that was already pyrolyzed, with increasing yield of secondary-decomposition products as the cycle count rises.

This is not a controversial physical claim. It is the engineering of any cyclically heated organic deposit. What is unusual is that it has not been characterized — or, to our knowledge, even systematically named — in the cannabis hardware safety literature.

3. Adjacent-field analogues

We are not the first to encounter this category of phenomenon. We are merely the first to publish it under a name in this field. The reservoir effect has direct analogues in three adjacent disciplines, each of which has developed mature characterization standards we can draw on.

3.1 Combustion engineering: deposit-induced misfire and emissions drift

Internal combustion engines have been understood, since at least the 1970s, to accumulate carbon deposits on intake valves, piston crowns, fuel injectors, and combustion-chamber surfaces. These deposits — chemically, the close cousin of the cannabis pyrolytic reservoir — alter the engine’s combustion characteristics over operating life. The phenomenon is so well established that the U.S. Environmental Protection Agency mandates emissions durability testing that characterizes vehicle tailpipe output not only at delivery but at intervals across a 100,000- to 150,000-mile useful life (40 CFR §86.1823). Manufacturers must demonstrate that emissions remain within compliance across the specified service interval; certification is not granted on a fresh-engine result alone.

The analytical apparatus and statistical methods developed for emissions-durability testing transfer directly to the cannabis hardware case, with the obvious caveat that the cannabis device is operated for hundreds to thousands of cycles rather than tens of thousands of miles.

3.2 Food-processing equipment: thermal carryover and Maillard cascade

Industrial deep-fryers, oven systems, and confectionery cookers undergo characterization for carryover effects — the contribution of charred residues to the flavor, color, and chemistry of subsequent product batches. The primary concern in food applications is sensory and microbial; in the late twentieth century, regulatory concern in some jurisdictions extended to acrylamide and other heat-degradation byproducts that accumulate when frying oil is reused across batches.

The relevant work for our purposes is the methodological literature on quantifying carryover. The food-equipment field has settled on testing protocols that characterize equipment output at specified intervals across a defined run length, with cleaning protocols specified between intervals and cleaning protocols specified to be omitted on test runs. This is structurally identical to what we are proposing for cannabis hardware.

3.3 Pharmaceutical inhalers: dose uniformity over cartridge life

Metered-dose inhalers and dry-powder inhalers are required, under U.S. Food and Drug Administration guidance for industry on inhalation drug products (USP <601>, USP <905>), to demonstrate dose content uniformity across the cartridge life — typically over the first, middle, and last doses. The regulatory concern here is therapeutic: the patient must receive the same active dose at any point in the cartridge’s intended use. Inhaler manufacturers routinely characterize cartridges across their full label-specified actuation count and present the data at submission.

Cannabis vaporizer cartridges are not currently held to any analogue of dose content uniformity testing. The manufacturer is not required to demonstrate that the consumer inhales the same chemistry on draw 50 as on draw 1. From the consumer’s perspective, the cannabis cartridge is a multi-dose inhalable drug delivery device that has been exempted from the regulatory expectations applied to every other multi-dose inhalable drug delivery device in the U.S. market.

The point of this comparison is not to argue that cannabis cartridges should be regulated identically to pharmaceutical inhalers. It is to observe that the analytical methods and characterization frameworks for cycle-life testing are mature in the adjacent field. We do not need to invent them. We need to translate them.

4. Why single-draw testing misses this

LSO testing as specified in CDSI-001 (P-002) characterizes a device under load. It does not, in its baseline form, characterize a device with a reservoir. The procedural reasons are straightforward:

The CDSI-001 baseline protocol calls for a freshly cleaned device fixture, a freshly loaded concentrate matrix, and a draw count of one to ten across the test run, depending on device class. This is a deliberate methodological choice: the goal of the baseline protocol is reproducible characterization of the device-matrix interaction, isolated from confounders.
A reservoir, by definition, is a confounder relative to that baseline goal. Its composition depends on the cumulative draw history of the specific device under test, which varies across consumer-use patterns in ways that are hard to standardize.
Reservoir-effect testing requires either a defined cycle-life run of substantial duration on the same device, or a calibrated synthetic reservoir deposited on a fresh device in a controlled way. The first option is time-expensive (a single run takes days). The second introduces methodological questions that have not been resolved.

The result is that the baseline LSO methodology, even when applied correctly, characterizes a part of the device’s operating envelope that a real consumer occupies for only a small fraction of the device’s service life. The remaining majority of the consumer-use window — the reservoir-developed regime — is uncharacterized.

This is the gap P-003 names. It is not a gap in LSO; LSO is the right baseline. It is a gap in what we test beyond the baseline.

5. Proposed protocol: cycle-life characterization (CDSI-001 §7, draft)

We propose an extension to CDSI-001, designated §7 in the next protocol revision, specifying cycle-life characterization. The extension is operationally straightforward and statistically demanding.

5.1 Cycle counts to be sampled

Each device under test is characterized at four points across a defined cycle-life regime:

Draw 1: Fresh device, fresh load. This is the baseline LSO measurement and is comparable across the dataset to the P-002 baseline.
Draw 50: Early reservoir formation. Selected because it corresponds approximately to the cumulative draws on a typical 0.5 g cartridge through one-third of its label use; carbonization is established but the heater geometry is largely unaltered.
Draw 200: Established reservoir. Approximately the full label-use of a 0.5 g consumer cartridge or roughly fifteen to twenty sessions on a refillable atomizer. This is the regime in which the secondary-pyrolysis pathway should begin to dominate.
Draw 500: Late-life regime. Beyond the labeled consumer use of any disposable cartridge but well within the operating life of refillable atomizers and bangers. Selected to characterize the regime in which the reservoir is interfering with heat-transfer geometry and the device is, by user perception, “burnt out” but is still in active use among harm-reduction populations and budget-constrained consumers.

A draw is defined per CDSI-001 §4.2 (3-second activation, 12-second inter-draw rest, controlled draw flow rate of 1.5 L/min). Loading is replenished at each measurement point such that the device is never operated on a fully exhausted load; this isolates reservoir effects from the distinct phenomenon of operating a heater on residual matter alone. The replenishment protocol is specified in §7.3 of the proposed protocol revision.

5.2 Analyte priority

The reservoir effect is expected to alter the off-gas profile in three identifiable ways. The analyte panel for cycle-life testing is therefore weighted toward the compounds in which we expect the largest changes:

Polycyclic aromatic hydrocarbons (PAHs), including benzo[a]pyrene and the EPA priority sixteen. PAHs are the canonical signature of secondary pyrolysis of organic deposits and are well established as carcinogens of concern. Detection by GC-MS with selected ion monitoring.
Carbonyl compounds (formaldehyde, acetaldehyde, acrolein, methylglyoxal). These are the principal acute-toxicity concerns from heated organic residues and are the focus of much of the existing electronic cigarette aerosol literature. Detection by 2,4-dinitrophenylhydrazine derivatization and HPLC.
Heater-derived metals (chromium, nickel, lead, cadmium, copper). The hypothesis is that reservoir-induced heater overrun raises functional heater temperatures, accelerating metal volatilization that would not be detectable in baseline LSO. Detection by ICP-MS.

The full CDSI-001 analyte panel is run at each measurement point. The above three are the priority signals.

5.3 Replication and statistics

Three production-stock units of each device class. Three cycle-life runs per unit. Each measurement point characterized in triplicate within each run. Statistical model: mixed-effects regression with cycle-count as fixed effect and unit as random effect, fit per analyte, with a Bonferroni-corrected significance threshold.

The design is expensive. A single complete run takes approximately seven days of bench time and consumes approximately 1.5 g of concentrate per unit. The total bench burden for the three-device baseline reported in the empirical companion paper is on the order of 200 instrument-hours plus chemist-time, plus material costs. This is one reason cycle-life testing has not been done by industry self-regulation: it is too expensive to be undertaken voluntarily without a regulatory or certification driver. It is also one reason CDSI’s tier-graded certification model (P-006) prices cycle-life testing into the highest certification tier rather than the baseline.

6. Anticipated signatures (predictions, marked as such)

This section is forward-looking. The empirical companion paper will either confirm or refute these predictions. We state them explicitly, in advance of the data, both because that is how testable predictions are made and because the predictions themselves illustrate the analytical framing the protocol is meant to reveal.

We predict, qualitatively:

PAH signature growth roughly monotonic with cycle count. Benzo[a]pyrene specifically should be undetectable or near-detection-limit at draw 1, detectable at draw 50, and meaningfully present at draw 200 and 500. The growth rate should be steeper for porous-substrate architectures (ceramic donut, ceramic coil) than for non-porous ones (quartz banger), because porous substrates hold more reservoir.
Carbonyl growth with a characteristic inflection. Formaldehyde and acrolein should rise modestly between draw 1 and 50 (reservoir formation), more sharply between draw 50 and 200 (secondary pyrolysis dominant), and may plateau or continue to grow between 200 and 500 depending on whether reservoir mass reaches a steady state where deposit and volatilization balance. The shape of this curve is itself a falsifiable prediction of the reservoir hypothesis.
Trace metal growth in late life only. Heater-derived metals should be at or near the baseline LSO level through draw 200, then begin to climb at draw 500, as the heat-transfer geometry alteration described in §2 starts to drive functional heater temperatures up.
Architectural divergence by draw 500. The three device classes, which P-002 reports as broadly distinguishable but in the same order of magnitude at baseline, should diverge substantially by the late-life regime. We predict the late-life ranking will not necessarily match the baseline ranking — devices that are clean at draw 1 may be the dirtiest at draw 500 if their architecture concentrates reservoir.

These predictions are derived from the physical model in §2 and the adjacent-field analogues in §3. They are not derived from data we have. The empirical companion paper will report which predictions held and which did not.

7. Implications for tier-graded certification

The Cannabis Device Safety Institute’s certification model (P-006, in preparation) is tier-graded rather than binary. Three tiers are proposed:

Tier 1: Reported. The manufacturer self-reports against the CDSI testing schedule. CDSI does not perform the testing. Suitable as a low-friction baseline that establishes manufacturer transparency without demanding that CDSI itself act as an analytical lab at scale.
Tier 2: Certified. A CDSI-accredited third-party lab performs the baseline LSO protocol (P-002, CDSI-001 §1–6) on a device or product line. Result is registered in the CDSI registry. Re-certification on revision of the device.
Tier 3: Gold Standard / Ongoing-Batch. Tier 2, plus periodic re-testing of production lots, plus the cycle-life protocol specified in this paper (§5), plus a registry of cycle-life data accessible to consumers, regulators, and harm-reduction researchers.

The methodological case in §2–§4 is the scientific justification for Tier 3. Without the cycle-life testing it specifies, a Tier 2 certification is — accurately characterized — a fresh-device-out-of-box certification. That is materially different from what consumers reasonably believe a “certified safe” mark to mean. Tier 3 closes the gap between the mark’s connotation and what it actually attests.

The economic case for tier-grading the certification (rather than rolling Tier 3 testing into the baseline) is twofold. First, Tier 3 is operationally expensive — see §5.3 — and treating it as the baseline would price the entire certification system out of reach for smaller manufacturers, with the predictable consequence of pushing them out of any participation. Second, tiering provides upgrade paths and aspirational targets, which are the standard mechanisms by which voluntary certification systems incentivize adoption (cf. the U.S. EPA ENERGY STAR program and its various “Most Efficient” designations).

We are aware that the introduction of tiers raises governance questions about who decides which tier a product earns, how disputes are adjudicated, and how tier-promotion and tier-demotion are handled. Those questions are addressed in P-006. They are not questions about the underlying chemistry, which is what this paper concerns.

8. Limitations

The argument made here has identifiable limitations, which we record so that the field can engage with them on their merits.

The empirical claim is unverified. The predictions in §6 are predictions. The companion paper, when it appears, will be the verification. Until then, this paper rests on the physical model of §2 and the adjacent-field analogues of §3. Both are, in our view, persuasive grounds for running the experiment; neither is a substitute for the data.
The cycle-count selections are conventional. The choice of draw 1, 50, 200, and 500 is informed by consumer-use distributions reported in harm-reduction surveys (NCSAM 2024, the FAIR Health all-payer claims aggregation 2025) and by the physical regimes outlined in §2, but it is not derived from a rigorous model of reservoir kinetics. A more refined characterization may identify cycle counts at which the regime transitions are sharper, in which case the protocol should be revised.
The synthetic-reservoir alternative is not specified. As noted in §4, a calibrated synthetic-deposit method would dramatically reduce the bench cost of cycle-life testing. We do not specify one in this paper because we do not yet have one. Developing a defensible synthetic-reservoir methodology is on the CDSI methodological roadmap.
Cleaning interventions are not characterized. Real consumers do, sometimes, clean their devices. The proposed protocol intentionally omits cleaning to characterize the worst-case (no-cleaning) regime. A complete cycle-life characterization should also include defined cleaning protocols and their effect on the reservoir, for which we have no data and no specified methodology yet. This is a known gap.
The argument generalizes uncertainly to disposable cartridges. Refillable atomizers and bangers have natural cleaning checkpoints. Disposable 510 cartridges do not — the consumer uses the cartridge until the concentrate is exhausted and discards it. The reservoir effect on a disposable cartridge therefore operates over a single bounded cycle (typically draw 1 through approximately draw 250 for a 1.0 g cartridge) but never across cleaning cycles. The proposed protocol should be applied to disposables, but its findings will inform a different consumer-use model than the refillable case.

We expect to learn from each of these limitations as the protocol is exercised. The point of publishing the methodological argument before the empirical results is to invite that engagement.

9. Closing

The cannabis hardware safety field has, until very recently, asked one analytical question: what does this device emit? That question is necessary but no longer sufficient. The field must now ask the second question: what does this device emit after the consumer has actually used it? The two questions have different answers. The gap between those answers is, from where we sit, the largest unmeasured chemical-exposure variable in the regulated and unregulated cannabis vaporizer markets combined.

The pyrolytic reservoir is the name we have proposed for the physical phenomenon that makes the second question matter. The cycle-life extension to CDSI-001 is the operational tool we propose for measuring it. The Tier 3 certification level is the institutional mechanism we propose for acting on it.

We do not believe any single one of these proposals is the only viable approach. We do believe that the overall structure — name the gap, specify the protocol, attach it to a certification mechanism that consumers can recognize — is the structure by which the cannabis hardware safety field will eventually mature. We are publishing this paper to put that structure in the literature so that other groups, including ones with greater laboratory and statistical capacity than CDSI’s founding team, can engage with it and improve on it.

The empirical companion paper will follow when the data exists. The methodological argument is the prerequisite. Here it is.

References

EPA. Vehicle Emissions Durability Procedures. 40 CFR §86.1823.

FAIR Health. Cannabis Use Patterns in U.S. Insurance Claims, 2020–2024. 2025.

Macosko, M. The Hardware Vacuum: Why Cannabis Consumption Devices Have No Safety Standard. CDSI Working Paper P-001. 2026.

Macosko, M. Loaded-State Off-Gas Analysis of Cannabis Concentrate Vaporizers: A Proposed Methodology and Three Baseline Case Studies. CDSI Working Paper P-002. 2026.

Meehan-Atrash, J., Luo, W., and Strongin, R. M. Toxicant Formation in Dabbing: The Terpene Story. ACS Omega, 2(9): 6112–6117. 2017.

NCSAM. National Cannabis Survey of Adult Mode-of-Use. 2024.

Troutt, W. D., and DiDonato, M. D. Carbonyl Compounds Produced by Vaporizing Cannabis Oil Thinning Agents. Journal of Alternative and Complementary Medicine, 23(11): 879–884. 2017.

USP. Aerosols, Nasal Sprays, Metered-Dose Inhalers, and Dry Powder Inhalers. USP <601>.

USP. Uniformity of Dosage Units. USP <905>.

Underwriters Laboratories. Standard for Electronic Cigarettes and Vaping Devices. UL 8139, 2018.

Working paper. Comments to matt@ineedhemp.com (papers@cdsi.click pending provisioning). This draft is v0.4 (May 2026); the next planned revision is post-empirical companion paper.