The Sauce on Source Codes: Obtaining DUI Dismissals from Software Non-Disclosures
Patrick T. Barone, Esq.
©2006 Michigan Beverage Journal, reproduced with permission, all rights reserved.
Michigan drivers violate the law when it can be proved that their bodily alcohol content was .08 or above at the time they operated a motor vehicle. In court this determination is most often based on the results of a breath test administered after the motorist is arrested. The machine used in Michigan for this purpose is called a "DataMaster" and is manufactured by National Patent Analytical Systems (NPAS), located in Mansfield Ohio.
The DataMaster measures the amount of alcohol in a breath sample by measuring the resistance of an infrared light beam as it passes through a sample chamber containing a small amount of the driver's breath. The basic theory employed for this purpose is that beverage alcohol (ethanol) absorbs infrared light and therefore any loss of energy in the sample chamber will be attributed to the presence of ethanol. This energy loss is then converted into a breath alcohol level by means of a simple computer that is on board the DataMaster, and like all computers source code or software drives this function.
In the typical drunk driving trial this DataMaster result is the only evidence of the motorist's bodily alcohol content, and this number is undermined if it can be established that the machine's software generates incorrect results or is not functioning properly.
What is "Source Code"?
Source code basically refers to the computer program that runs the DataMaster. What makes the DataMaster and other breath test devices unique among measuring devices is that after the amount of alcohol in the breath sample is measured, a variety of mathematical calculations must be performed to make the numbers relevant to the human system. In other words, to make the raw number relevant to the question at hand, does the motorist have an unlawful bodily alcohol level?
The questions posed by source code litigation are essentially four-fold. First, are the mathematical equations employed by the DataMaster properly formulated, calculated and applied, second, is the code acceptable to the relevant scientific community, third, has the code been subjected to an appropriate level of scientific scrutiny both inside and outside of the DataMaster itself, and forth, have any changes to the code been made after the device was approved in Michigan for use by law enforcement? The answers to these questions govern whether or not it can be persuasively argued that the test results should be excluded on the basis that the state cannot guarantee that they are reliable enough to sustain a criminal conviction. A complete understanding of this issue however will require a more vivid understanding of the science behind breath testing.
The Science of Breath Alcohol Testing:
The first thing that a breath testing device must be able to do properly is assure that an appropriate breath sample has been collected from the motorist. For example, the DataMaster measures the flow rate of the breath sample as it is delivered by the motorist, and in order for the sample to be accepted, certain pre-programmed parameters must be met. This is itself a critical calculation because a fundamental aspect of breath testing for forensic purposes is that only deep lung air be analyzed.
A related calculation is based on a breath/blood conversion ration, and in the forensic setting this conversion is necessary because breath alcohol is only relevant as it relates to a comparative blood alcohol level. Thus, the DataMaster must covert the breath alcohol measurement to a comparative blood alcohol measurement. This conversion is required because breath alcohol does not impair or intoxicate, and it is only after the alcohol has been distributed throughout the body via the blood that the alcohol can cause the brain to be affected. This conversion is based on "Henry's Law", which states that in a closed system, the amount of alcohol in a liquid will reach equilibrium with amount of alcohol in the air above the liquid. When applied to breath testing, the liquid is of course blood, and the air is the breath. The closed system is the lungs. Henry's law says that for drunk driving purposes a particular ratio (at a constant pressure and temperature) can be assigned to this system, and the ratio commonly accepted for law enforcement purposes is 2100/1. So, according to the theory, for every one molecule of alcohol in the breath, there are 2100 molecules of alcohol in the blood. Thus, in order for the breath test result to have relevance in drunk driving law enforcement, the source code used in the DataMaster must incorporate mathematical functions that will allow it to apply Henry's Law, along with the 2100/1 ratio.
Another calculation of great significance is loosely termed "slope detection." The purpose of slope detection is to allow the DataMaster to discern the presence of mouth alcohol because this would invalidate the test. The operator would know mouth alcohol has been detected because the DataMaster would produce an error message or status code that reads "invalid sample." Slope detectors are also used to help assure that alveolar (deep lung) air actually composes the breath that is being tested. This is essential because in order for Henry's Law to have any applicability, deep lung air must be measured because the deep lung tissue most closely approximates a closed system. Slope detection is an essential safeguard over the DataMaster breath test results, and like Henry's Law, it is totally dependant on properly formulated source code.
These issues are not some mere fanciful imagining of creative defense attorneys. For example, it has been established in court testimony in the state of Washington that changes have been made to the way the software calculates the 'invalid sample" error message. Sgt. Rod Gullberg supervises the Breath Test Section of the Washington State Patrol and in many Washington DUI prosecutions he provides the expert testimony required by the State to satisfy their burden of establishing the reliability of a breath test result. In the case of State v. Straka, 116 Wn.2d 859, 877, 810 P.2d 888 (1991), he testified that several changes were made to the way the DataMaster identifies an invalid sample. Also, that previous versions of the DataMaster "were too sensitive in their calculations as a result of programming errors."
Looking again at source code the question is whether or not the math used to make these conversions and calculations were properly formulated by NPAS when they developed the DataMaster, and also, whether or not this math was properly programmed into the source code. The only way to know this would be to have an independent computer expert look at the code, and this is often the very goal and purpose of source code litigation. If these calculations are in any way invalid, that is, they don't properly calculate the breath alcohol number in a way that is forensically acceptable, then the breath test results should not be admitted into evidence during drunk driving prosecutions. So, if inspections of the source code revealed and definitively determined that problems existed this would of course have a dramatic and deleterious impact on drunk driving prosecutions.
Compliance Testing Issues:
One might assume that NPAS did a great deal of internal testing during the production of the DataMaster, which presumably would have included compliance testing the source code in some way. Even if this is true, a problem arises because it appears that the source code has never been tested either outside the company or outside the DataMaster "box." It is within the realm of possibilities that the code is so riddled with errors that the validity or accuracy of breath test results could not be guaranteed, or that the code is so badly written that it could not carry out its intended functions. Independent testing is required to either prove or disprove these conjectures.
In scientific circles, this type of independent testing is called peer review. Peer review (also known as "refereeing") is a screening process used prior to the publication of manuscripts in scientific journals and in the awarding of funding for research. Peer review is used by publishers and funding agencies to select and to screen submissions for accuracy. In fact, peer review is a fundamental for publications that wish to assure that authors meet the standards of their discipline. Publications and awards that have not undergone peer review are often regarded with suspicion by scholars and professionals.
It appears that the DataMaster source code has never undergone any type of true peer review. In an ideal situation the peer review process would have been conducted before the source code for the DataMaster was "approved" for use. Defense attorneys believe that the source code itself has never really been provided in a way that would allow it to be independently analyzed.
A significant aspect to the source code litigation has been to obtain the source code so that it can be submitted to this type of testing by the relevant scientific community. Until such independent testing occurs in an unequivocal way, source code litigation will likely continue.
Software Changes After Approval:
Prior to being used by law enforcement, the DataMaster was subjected to an approval process by the Michigan Department of State Police. The DataMaster is also on the list of complying products maintained by the National Highway Traffic Safety Administration. This approval process is required by state law. Defense attorneys argue that if any changes to the equipment are made after the equipment is approved for use, then the equipment is no longer conforming because it is quite literally not the same equipment that was originally approved. Thus, each new change in the software should require a new approval process. Florida attorneys were able to show that changes were made after approval, and this was one of the main reasons (along with non-disclosure) that the Florida source code litigation has met with success.
Monitoring changes in the software is particularly important because the DataMaster has several internal "safeguards", and as a part of its operating function, it checks itself and the integrity of the breath sample to make sure everything is operating properly and that an appropriate breath sample is being obtained. This functionality is based upon algorithms and other information, including certain assumptions that are written into the software. If changes are made, then these changes may have been intended to correct "mistakes" (software bugs) made by earlier versions. It is obvious why this information would be useful to the criminally accused, both past and present.
The History of Source Code Litigation:
The code used in the DataMaster is antiquated by today's standards, and excluding patches and updating, the technology currently in use dates back several decades. Most of the DataMasters in use in Michigan were manufactured in 1993 and 1995. Some source code litigation relative to the DataMaster relates back to that period of time.
It should also be noted that not all states use the DataMaster to test a motorist's breath for alcohol. Florida for example uses a device called an Intoxilyzer, and recently it is the litigation in Florida that has received the most publicity. But even in Florida there is incomplete agreement among the judiciary as to whether or not source code must be produced or even if it is relevant. So far, Judges in the Florida counties of Manatee, Sarasota, Seminole and Volusia have ruled that the defense was entitled to the Intoxilyzer's source code. In an order from one such case the court held:
When the law expressly provides that defendants are entitled to full information about the instrument that is used to establish their guilt, such full information logically includes making the instrument available for open inspection. Full information should include the software that runs the instrument. To construe the statute otherwise, is tantamount to granting the state authority to use confidential information (i.e., source code) to establish the guilt of a criminal defendant without disclosing the information to the defendant for inspection and possible impeachment. Unless the defense can see how the breathalyzer works and verify it is an approved machine, it remains nothing more than a mystical machine used to establish an accused's guilt.
However, Judges in other Florida counties, such as is Palm Beach County have denied access to DUI defendants who sought production of source code. Many cases have been dismissed in Florida because the company that makes the intoxilyzer had refused to comply with a court order to release the source to the defense.
It is important to note that the successful defendants in some of the Florida Counties were able to establish that several Intoxilyzers in use at the time each contained different hardware from one another and in some instances different software. What was even more troubling to the judges was that the state agency charged with assuring the scientific reliability of these devices had no explanation for why these machines were not exactly alike. Had these changes not already been established then the judges certainly may have ruled otherwise. To a large extent this is why the judges determined that the defense was entitled to examine the source code.
Source code litigation has been pursued in the past, most notably in the state of Washington where, like Michigan, the DataMaster is used. In that litigation some code was released by NPAS and was analyzed by a defense expert. However, the defense was dissatisfied with what was produced, and argued that NPAS had not produced the appropriate version of the software and also that the software engineers were not able to do the necessary hands-on testing. Nevertheless, they did inspect the version of the software produced, and the Final Software Reliability Analysis dated March 17, 1995, expressed several concerns including a belief that DataMaster software had not undergone "even minimal [compliance] testing necessary for a product of this size and impact." Also, that the DataMaster software was written in the style of a video game and clearly showed a history of many modifications. It was their opinion that such modifications were "notoriously unreliable" and "would be nearly impossible to detect even under extensive testing." Another issue identified in the report was what they believed was a probable error in the sequence during which the DataMaster purges the test chamber by drawing room air through. Their conclusion was that as a result of this coding error, the sample chamber "is almost never clean after the purge cycle." This litigation was eventually resolved. Remaining unsatisfied however, the Washington defense bar continues to raise source code challenges from time-to-time.
At the time of this writing, similar litigation is being pursued in other states, including Michigan. It is likely that this litigation will meet with varying levels of success and will in part be dependant on the viewpoint of the various judges presiding over the particular case, and perhaps more importantly, by the skill and knowledge level of the defense attorneys involved. The defense attorney's role is particularly crucial because it is he or she who will ultimately be charged with the task of appropriately educating the judiciary.
DataMaster's Response to Source Code Issues
DataMaster has released source code under a protective order/non disclosure agreement since 1994, and it's fair to say that the owner of NPAS, Mr. John Fusco, believes that the entire issue is much ado about nothing. He has indicated that "Defense attorneys, by and large, are like a bunch of little old ladies at a card party. The software gossip is popping up all over and rumors are running rampant." Regarding the Washington litigation he has stated that "in that case, frankly, both the defense attorney and the experts were so inept that if there were any significant issues, they wouldn't have recognized them anyway. Once all the wrong conclusions reached by the experts were corrected (presumably including those listed above), they were left with no significant software issues."
Mr. Fusco has indicated that NPAS has released the software in Washington a dozen or so times over the years. It is Mr. Fusco's opinion that: "Releasing does avoid the problems that Florida is having, but by and large, does not help the media. The primary problem in Florida is not that the software does or does not work, but whether or not the current instrument is still the same one that was originally approved." Mr. Fusco correctly points out that the source code is only one way of making this determination.
Mr. Fusco also does not believe it is either possible or necessary to examine the code separately (from the DataMaster) in order to determine if it is working. It is his opinion that while it "may be necessary to look at the software to make some determinations, but to look at it separately from the system (the instrument) is patently flawed and next to impossible. The experts in Washington proved this beyond all doubt."
Again, according to Fusco, "it is not possible to understand or evaluate the software unless the people doing this understand what the instrument is doing and test it in its' operating environment. The initial evaluation report submitted in Washington 10 years ago contained about 22 items they said were so bad that the instrument could not possibly measure alcohol. Sgt. Gulberg (the head of Washington's breath testing department) took his time to instruct them. The final report contained 2 items, neither of which was even remotely significant."
While Mr. Fusco does have a fairly pragmatic approach to the entire issue, it does seem apparent that he is frustrated by the time and money required for at all levels to both prosecute and defend issues that he believes are irrelevant. Interestingly, he also believes that drunk driving defense attorneys may be ethically required to pursue source code litigation.
Source code litigation is alive and well, and is likely to continue for many more years before it is finally resolved. It is likely that the defense bar will not be satisfied until the source code used in breath testing has been thoroughly examined and has been unequivocally shown to be forensically reliable. On the other hand, the prosecution, emboldened by Fusco's comments, will likely argue to either block disclosure, or to convince the judiciary that source code is simply not relevant.
If the source code issue makes it to the Michigan Appeals Court then it seems likely that any opinion issued by them will ignore the science, and will instead follow the previously established trend for administrative rule violations, which is to find that source code issues go to "weight" rather than to "admissibility." This means that the trial judge will be required to allow the defense attorney only to argue the source code issue to the jury. This is unfortunate because the integrity of the source code goes to the very core of the scientific reliability of the test result. As such it is quite different from a comparative technical violation of an administrative rule. It is also unfortunate because such a ruling would once again require the trial courts to disregard their gate-keeping function.
If this were to happen to the source code issue at the appellate level then regardless of any source code errors that may exist, the breath test results will come it to evidence, and it will be up to the non-scientific jury to decipher and determine if these issues are important. Because juries tend to give great weight to the breath test "number", this may mean that drivers that are otherwise lawfully operating their cars after drinking will find themselves convicted of drunk driving, and that their convictions might well be based on forensically unacceptable evidence.