- Manual references
- Initial photos and conditions
- Day 1: Power checks and initial diagnostics
- Day 2: diagnostics, A5 board test and installation
- Day 3: diagnostics, A2 board
- Day 4: A3 reference measurements
- Day 5: A1 test
- Day 6-9: New A3 installation
- Day 10-22: Modification for lower LTZ1000ACH oven temperature
- Day 23: DC 10VDC test with reference input
- Day 30: Calibration adjustment prior to final test
- Day 33: Final calibration tests and report
- Day 40-50: Comparisons to other xDevs.com lab gear
- Day 52: Minor fixes and touch-up
- Thermal infrared images
After a year of use in original rusty repaired 3458A it’s advantage in lab and operation performance was obvious and clear. Biggest benefit of this precision DMM is due it’s ultra-high ADC performance and benefits which come from it: Calibration from single 10 V and 10 KΩ standards and ACAL function to remove drift and thermal coefficient errors.
My good friend once mentioned – it’s nice to have at least two pieces of everything. In case of primary reference instrument failure, you still have backup B unit to chase the precision rabbits. While this sounds like a great idea, buying even used good for 3000-5000$ USD DMM just to keep spares is a bold move, especially for hobby home lab. This plan was put on hold, till very recently, when I got a chance to buy another unit. After some seller-related shenanigans, second unit was finally shipped and this repair work-log of second HP 3458A is now possible!
And to follow our best tradition, unit is completely broken and non-functional. Not even photos of power on were possible. There is no fun in buying working equipment. No room for learning how to fix stuff! :)
There are service notes/engineering changes were published during years of 3458A’s lifecycle. If your unit is old, worth to check if any of them required to do.
Initial photos and conditions
Like a wizard with all-see crystal sphere, here are list (likely not complete) of issues present with the box.
- Missing A5 PCBA (Outguard controller) Replacement bought and ready
- Bad VFD? VFD is good and bright
- Missing top inguard cover. TBC
- Damaged front panel bezel Replacement ready
- Scratched display window Replacement ready
- Damaged front terminal block Will swap with rear. Use triax DIY ports for rear instead.
- A3 has balanced rundown convergence issue, stops working instantly A3 replacement installed, stable and good!
- A2 has offsets on all ranges Fixed by recalibration
- A1 can pass ST/ACAL with good A2/A3 OK
Day 1: Power checks and initial diagnostics
Take a look on that sick box. Understanding is beyound me, how these precision very specialized instruments can end up in shape like this.
Rear side is okay, nothing bad there, except obvious sign of missing A5 Outguard board with GPIB connector.
Plastic front panel is literally torn in pieces, just like some angry hulk was hungry. Front panel terminals did not survive the abuse, so that’s a $229 USD right there, if one decide to replace the damaged terminals (Keysight P/N 03458-62111).
Rear terminals are OK, and they are exactly same part as one used in front. So I will move terminal block from rear to front, and use another solution on the rear side. More details to come later in the article about this.
Key rubber pads are not held in place correctly, perhaps due to front panel plastic damage?
FP PCBA is there at least, we can see a VFD glass thru the scratched front panel window.
Panel plastic is definately going to be replaced. During disassembly it was literally falling apart.
And now it’s clear why rubber keypads were not in place. Every single FP PCBA locking notch is broken. That is the result of front face impact. Perhaps someone dropped unit on it’s front face? I’m afraid fragile VFD glass is broken.
But VFD glass looks okay. Will see if it’s fully functional upon power up with replacement A5 digital board.
|Red-Red/Yellow||1.8 Ω, 23 VAC||1.73 Ω, 22.8 VAC|
|Red-Red/Yellow||1.8 Ω, 23 VAC||1.71 Ω, 22.8 VAC|
|Yellow-Yellow||0.4 Ω, 10 VAC||0.35 Ω, 10.2 VAC|
Table 1: Transformer test results
Looking good, transformer appear to be healthy. It was quite a pricey part, as we learned during first HP 3458A repair. Schematics of transformer is shown on Image X, in case our readers need a reminder.
Image 16: HP 3458A’s transformer configuration
Now let’s test main power supply rails, both inguard (on A4 PCBA) and outguard (A6 PCBA). If these rails are bad, this could be a disaster and path to the infinite rabbit hole.
|Outguard J400||+5 VDC||+5.09 VDC, <3mVpk-pk|
|Outguard C8||+15 VDC||+18.118 VDC, <3mVpk-pk|
|Outguard U200||+8 VDC||+8.12 VDC, <3mVpk-pk|
|Inguard +5||+5 VDC||+5.05 VDC, <3mVpk-pk|
|Inguard +18||+18 VDC||+18.48 VDC, <3mVpk-pk|
|Inguard -18||-18 VDC||-18.68 VDC, <3mVpk-pk|
Table 2: Power supply test results
No problems here too, everything is well within spec. Still have a chance for this repair project to be easy and quick? Yeah, I wish. Time will tell.
Time to dig deeper now.
Day 2 diagnostics, A5 board test and installation
Outguard controller has arrived and ready for initial check and service. This A5 PCBA is newer than Rev.A from original HP 3458A as it’s manufactured in 1996, and features single firmware ROM chip, which is ST M27C4002 in DIP40 package. ROM is UV-eraseable EPROM, so firmware update is possible in field, using UVA-UVC lamp and EEPROM-programmer.
Board P/N is 03458-66505 Rev.B and Engineering Revision tag is 3313. Two-layer PCB made in USA.
Board came with RAM socket and old 1995-dates Dallas chips, so those are to be removed and replaced with good quality collet sockets, to install fresh new Dallas NVRAMs. RAM on this board formed by pair non-volatile Dallas (today it’s MAXIM) DS1230, high and low bytes separately. If these NVRAM become bad, meter during power-on selftest will throw error RAM TEST 1 LOW or RAM TEST 1 HIGH depends on which IC died and lockup further operation.
No bodges or issues on bottom side of the board, overall pretty clean.
New DS1230AB and DS1220 installed in sockets, ready for test. I programmed DS1220 with calibration dump from my good 3458A, so I could test operation of this A5 without loosing calibration. Digital boards in 3458A can have different revisions and firmware versions without loosing calibration, as far as calibration ROM is properly transferred between boards.
Firmware is Rev 7. I will update it later to latest Rev.9. Here is collection of firmware ROM dumps for reference:
Firmware ROM dumps for HP/Agilent/Keysight 3458A, read by general purpose ROM programmer. I used TL866CS from eBay to work with firmware ROMs. From information we have, firmware dumps are compatible with all hardware versions.
|ROM||Revision 2||Revision 4.6||Revision 6||Revision 7 (1992)||Revision 8 (1998)||Revision 9 (latest)|
|03458-88877||03458-88887 Single ROM|
Table 3: Firmware ROM dumps
Board installed in good 3458A revealed no problems, everything work as expected, including GPIB interface. And to confirm that calibration is not lost or changed, here’s quick GPIB datalog chart, showing LTZ1000 module readout.
Reading is within ±0.2ppm from original value, prior to A5 board swap.
Since I already installed this fresh A5 in my good meter, I decided to keep it there, so no other hidden issues would arise, while this second sick 3458A will get it’s old A5. There is no functional difference, as both boards running now same Rev.9 firmware.
Day 3 diagnostics, A2 board
As high voltage AC path has incorrect reading, let’s check input HV AC attenuator operation. Schematics fragment from CLIP will be a great help on this.
First suspects are U201, which is DG211 analog switch, low leakage protection JFETs and zeners Q202,Q203,CR203,CR204 and attenuation network resistors R204,R205,R206,R207,R208,R209.
Day 4, A3 reference measurements
Measurement of ADC references, which are derived by Linear LT1001 opamps and resistor network inside custom HP ASIC U180, which is famous for main cause of A/D failures and excessive drift.
Here’s result taken by Keithley 2002 with scanner card and F732B as 10V cross-check reference.
Image 30: Front panel
No need further comments, U180 is dead. These +5, -12, +12 should be fixed ratio of main 7V output of LTZ1000A module A9, but due to bad resistors inside U180 hybrid, we can see large drifts over just few hours of time.
Also few extra photos were taken
Now we need to repeat all that “find-good-A3” quest once again. Anyone still believe it’s possible to get good working 3458A on eBay under 3000-4000$ USD? Cost of A/D PCBA from Keysight service center is around $1300 USD, if reminder is required.
Day 5, A1 Test with good A2 and A3 boards
This board have later upgraded 40.000 KΩ reference resistor, R207 with Vishay VHP101 mark on it, so it’s expected to have less than ±0.3ppm/°C TCR.
To verify correct A1 DC board function faulty original AC Board (A2) and A/D Converter board (A3) were temporarily replaced by ones from good 3458A. This will help us to isolate any issues of A1, and by using verified A2/A3 guesswork is out.
And here we go with first test:
Surprisingly it passed self-test just fine. Artifact calibration ACAL ALL also went OK, so there is high chance of good condition at least in A1 PCBA. Funny enough, I would rather prefer this board to be faulty than A3, as it’s relatively easy to fix, while with bad U180 hybrid not much we can do.
Next would be test with DCV signal, to make sure it’s all good and stable, before making conclusions on A1.
Day 6-9: New A3 installation and initial tests
After several weeks replacement A3 finally was acquired. There are multiple revisions of A3 board assembly in 3458A units over its 28 years of lifetime.
Table 4: A3 PCB versions
It’s very recent, according to date codes on ICs made in week 16 year 2016, with most of opamps and logic ICs made in 2015. As Keysight running into more and more difficulties to replace old obsolete parts, they were forced to design patch PCB with P/N 03458-26550 with ALTERA CPLD (RTL code 03458-85550) to replace gate array ASIC with digital logic of A/D converter.
Soldering points around patch boards is still covered in flux and has solder bits, so it’s bit how-ya-doing job there. There is also visible flux around integrating capacitor, perhaps it was selected and replaced after board was originally assembled?
Since used MAX V CPLD is not a direct replacement, without support of 5V logic level, array of Texas Instruments SN74LVC8T245 8-Bit Dual-Supply Bus Transceivers used to interface main A3 PCB signals. There are also two LDOs in center, U11 and U12 with bunch of 0402 and 1210 caps around for decoupling. Three LEDs LD1/LD2/LD3 used for status reporting and unpopulated 6-pin 2.54mm spaced header is likely CPLD JTAG port for ISP.
Test pads on the corner edge are reserved for JTAG connector to handle ISP.
Image 53: CPLD detected with JTAG ISP tool
ALTERA USB Blaster was able to read CPLD configuration without any problems, with dump stored in backup file. Checksum 0056E14F matches one on sticker, so it’s indeed correct dump.
Image 54: Test pad connections using Tek grabbers
Another patch PCB 03458-66530 is installed to replace obsolete Elantec EL2018CN fast ±15V comparator.
Let’s see if this ADC board is good, as it came untested without any warranty. To do this I’ll run test covered in Service Note SN18 which compares ADC A3 stability versus A9 LTZ1000 reference, assuming later one is good and stable. A week or so of data would be sufficient to tell the initial result.
SN18 ADC stability test
|Date/time||Meter temperature||CAL? 72||CAL? 1,1||CAL? 2,1||Deviation, ppm|
|18.DEC.2016 20:48||35.5||983.313029E-03||39.9991607E+03||7.07743961||Reference, 0.00|
|18.DEC.2016 22:35||35.3||983.313077E-03||+0.0488 (0.0488)|
|19.DEC.2016 08:32||35.5||983.313098E-03||+0.0214 (0.0702)|
|19.DEC.2016 18:29||35.7||983.313055E-03||-0.0437 (0.0264)|
|20.DEC.2016 07:48||35.3||983.313080E-03||+0.0254 (0.0519)|
|20.DEC.2016 18:36||35.7||983.313100E-03||+0.0203 (0.0722)|
|20.DEC.2016 23:18||35.3||983.313055E-03||-0.0458 (0.0264)|
|21.DEC.2016 00:56||35.3||983.313155E-03||+0.1070 (0.1281)|
Table 5: Initial SN18-stability check
So far initial test revealed no problem, let’s proceed with quick calibration to Fluke 732B 10V DC standard by executing CAL 10.000000 command and repeat same test, together with datalog of 10V. This will let us to see stability with DC signal applied on 10VDC range.
|Date/time||Meter temperature||CAL? 72||CAL? 1,1||CAL? 2,1||Deviation, ppm|
|21.DEC.2016 09:06||35.5||983.331695E-03 A||39.9991607E+03||7.07037491||Reference, 0.00|
|21.DEC.2016 20:48||36.2||983.331710E-03||+0.0153 (+0.0153)|
|22.DEC.2016 08:26||36.2||983.331676E-03||-0.0346 (-0.0193)|
|23.DEC.2016 00:52||35.6||983.331701E-03||+0.0254 (+0.0061)|
|23.DEC.2016 08:30||36.0||983.331641E-03||-0.0610 (-0.0549)|
|24.DEC.2016 17:42||36.2||983.331664E-03||+0.0234 (-0.0315)|
|24.DEC.2016 18:16||36.1||983.331626E-03||-0.0386 (-0.0702)|
|24.DEC.2016 23:21||35.5||983.331697E-03||+0.0722 (+0.0020)|
|25.DEC.2016 16:45||36.6||983.331594E-03||-0.1047 (-0.1027)|
|26.DEC.2016 01:23||35.2||983.331753E-03||+0.1617 (+0.0590)|
|26.DEC.2016 08:35||35.4||983.331695E-03||-0.0590 (+0.0000)|
|27.DEC.2016 03:41||35.2||983.331765E-03||+0.0712 (+0.0712)|
|28.DEC.2016 11:20||35.9||983.331619E-03 B||-0.1485 (-0.0773)|
Table 6: Post-calibration SN18-stability check
Overall A/D drift rate result per Service Note 18 is [*C = [(A – B) * 1000000] / [A * 7 days] formula*] = -0.0110 ppm, which is well within 0.43 ppm requirement. If this rate is greater than 0.43 ppm/day, then the A3 assembly needs to be replaced.
There is no much repair possible on drifty A3 boards. This is also confirmed by many 3458A owner reports, since usually root cause of excessive drift/integration errors is faulty U180 custom hybrid chip, which is not repairable. Last year I’ve spend more than 200 hours, trying different things to get A3 working in my first 3458A with no avail. Final solution was replacement of A3 PCBA altogether.
Day 10-22: Modification for lower LTZ1000ACH oven temperature
As we will be doing own calibration for this meter, and it’s main use targeted for aircon environment, so very high nominal ~90°C oven temperature of main LTZ1000A-based reference (PCBA A9) is undesired. Higher temperature cause very large long-term drift, which is specified by HP/Agilent/Keysight at 8 ppm/year for standard reference modules and 4 ppm/year for special pre-selected option 002 “high-stability” references. This design decision was made by HP engineers to satisfy +50°C operation temperature range for tight rack-mount usage, such as military applications.
So for metrology and lab purposes this is not good, but solution is rather simple. We can reduce on-die oven temperature of LTZ1000A reference chip by just adding one resistor in parallel with original temperature setpoint one. There is even empty footprint on A9 PCBA, perfectly suitable for this task.
Image 56: Schematics of lower oven temperature tweak
Added resistor need to be high enough value to reduce original X411 15 KΩ resistor down to ~13KΩ which will reduce temperature to ~55°C, improving long term-stability to <2 ppm/°C. Such modification was already successfully performed on my first 3458A and meter indeed is good and stable after the recalibration.
If you change the oven temperature, you will lose the calibration, as the LTZ1000A will change its reference voltage, invalidating previous calibration constants.
Additional resistor has to be high-stability and low tempco type, to ensure that temperature set-point does not drift over time. Vishay Precision Group VAR Z-foil 100 KΩ with specified maximum TCR ±2 ppm/°C resistor satisfy requirements well. Resistor temperature coefficient must be no more than 3ppm/°C. Still most of error comes from original 15 KΩ metal foil resistor, so there is no need for ultra-stable 0.2 ppm/°C resistors unless original one is replaced too. Retail price for VAR BMF resistor range in $USD 15-30 mark, depends on tolerance, value and order quantity required.
Here are few photos of modified LTZ1000 reference, A9 PCBA with 95 KΩ in our first 3458A:
And this second unit A9 module, modified same way but with VAR 100KΩ resistor instead:
Now after modification unit was assembled and warmed up few hours to perform stability test once again. No problems revealed at this stage.
Day 23: DC 10VDC test with reference input
Now as we have unit functioning, it’s vital to determine if there is hidden drift over long-term period. And there is no workaround but to actually connect very stable source, such as +10 VDC ovenized DC reference, in this case mighty Fluke 732B and log data over extended period of time, such as week or two. By monitoring deviation and temperature, one can reveal drift component of measured voltage. It’s also important to keep stable environment conditions, such as air temperature/humidity to reduce error from these condition variations.
To draw a conclusion regarding new A3/meter stability next set of targets should be achieved:
- Voltage need to be stable within ±0.1 ppm window over 10 minutes of constant TEMP? PASS
- Stable within ±0.275 ppm window over 24 hour of constant TEMP? PASS, 0.2 ppm
- Stable within ±0.7 ppm window over 7 days of constant TEMP? PASS, 0.43 ppm
- Change less than ±0.5 ppm with TEMP? change 5°C PASS
- Stable within ±1.0 ppm window over 14 days of constant TEMP? PASS
Day 30: Calibration adjustment prior to final test
After stability check next step in repair is calibration adjustment and accuracy testing against know external references. Due to very tight specification for HP 3458A finding “good enough” is often a mighty challenge. However this time opportunity to access stable and well aged Fluke 5700A and calibrated Fluke 732B & ESI SR104 standards was exploited. “Exploit” is correct word here, because meter was located in remote lab for few weeks, with wiring connected between 5700A and our DUT 3458A for days without interferences. Both instruments controlled by Raspberry Pi 3B via GPIB with “linux-gpib and NI GPIB-USB-HS pod”. Environment conditions were logged from “Bosch BME280 THP sensor”.
All equipment was left to warm up and stabilize few days before actual tests ran. Calibration was performed following guidelines in “HP 3458A Calibration manual”. Recently calibrated and well cared DC Standard Fluke 732B was used to execute DC gain calibration, ESI SR104 used for 10 KΩ adjustment, and setup with HP 3325A, Ballatine 1384 3V and 1V thermal converters with nanovoltmeter HP/Agilent/Keysight 34420A.
Day 33: Final calibration tests and report
Day 40-50: Comparisons to other xDevs.com lab gear
|Standard under test||HP 3458A #1||This HP 3458A #2||K2002 #1||K2002 #2|
|HI/LO Bias current, 10 DCV, AZERO OFF, ARM HALT||pA||pA||pA||pA|
|+7.1846680 VDC, LTZ1000A, HP A9 STD||+7.1846284 [-5.51 ppm]||+7.1846288 [-5.45 ppm]||+7.1846272 [-5.67 ppm]||+7.1846243 [-6.08 ppm]|
|+7.1298358 VDC, LTZ1000A, KX module||+7.1298212 [-2.04 ppm]||+7.1298214 [-2.02 ppm]||+7.1298195 [-2.29 ppm]||+7.1298161 [-2.76 ppm]|
|+7.1366740 VDC, LTZ1000A, KX module||+7.1366613 [-1.78 ppm]||+7.1366595 [-2.03 ppm]||+7.1366599 [-1.97 ppm]||+7.1366570 [-2.38 ppm]|
|+7.1366450 VDC, LTZ1000A, KX module||+7.1366414 [-0.50 ppm]||+7.1366398 [-0.73 ppm]||+7.1366397 [-0.74 ppm]||+7.1366368 [-1.15 ppm]|
|10V DCV, LTZ1000A, HP 3245A||+9.9999596 [-4.04 ppm]||+9.9999567 [-4.33 ppm]||+9.9999643 [-3.57 ppm]||+9.9999512 [-4.88 ppm]|
|10.00001 KΩ, VPG VHD200 #1||10000.024 Ω [1.4 ppm]|
|9.999894 KΩ in G9330 body, VPG VHP202||NT||9999.892 Ω [-0.3 ppm]|
|0.997091 GΩ, DIY STD with Ohmite MOX1125-23|
Table 7: Comparisons with our previous 8½-digit DMMs
Also separate set of tests done using Wavetek 4920M as AC voltage standard measurement system.
|Standard under test||Wavetek 4920M||HP 3458A #1||This HP 3458A #2||K2002 #1|
|10 VAC 1kHz, HP 3245A|
Table 8: Comparison with Wavetek 4920M
Day 52: Minor fixes and touch-up
As all functions and operation is confirmed good, I also performed some last minor fixes, such as replacement of all electrolytic capacitors on A6 and A4 power supply boards, replacement of Schaffner mains filter, fan replacement and new front panel installation.
These still intrusive tasks were done after all calibration transfers and main standards measured, to ensure minimal errors on calibration accuracy. Usually any disassembly or repairs, even minor as fan replacement are big no-no after final calibration is done. And exception to this very good rule can be made only if you really know what are the results and ready to loose calibration accuracy. These are not an issue for 1% accurate tool, but remember, no detail is minor once we talking about 1ppm (0.0001%) accuracy!
Thermal infrared images
Thermal analysis and understanding of hot spots and thermal gradients present in metrology-level instrument like HP/Agilent/Keysight 3458A is vital to achieve stable and repeatable ppm-level performance.
HP engineers spend significant amount of resources for overall system design with thermal zoning in play. It is not easy to see this at first glance, but thanks to thermal imagery there are ways to grasp overall key areas of focus. Procedure of capturing images below was next:
- Meter warmed up for over 24 hours in +25°C ±2°C environment.
- Outer covers were inplace to ensure thermal distribution as by design.
- Top cover was removed and thermal photos of A1,A9,A5,A4 boards taken within first 3 minutes by Fluke Ti32 camera.
- Cover installed back, meter rotated upside and left for 20 minutes to stabilize.
- Bottom cover removed, photos of A6,A3,A2,A4 taken within 3 minutes.
Goal of getting second 3458A in order was achieved, even though actual final effort and cost was way over initial estimated numbers. This is a good lesson to possible volt-nuts, as buying such instrument for low $500 looked like a bargain at first glance, but quickly turned into significant repair expense. It was not unexpected on my part, but there were hopes to avoid expensive A3 board swap.
And this is not considering calibration cost or getting 3458A on Keysight warranty program (which is pricey, but well worth for such expensive instrument). I was lucky to have temporary access to standards and stable calibrator, such as Fluke 5700A to perform own adjustments and calibrations. This also helped to obtain yet another international transfer for DC voltage, resistance and DC current as well as AC calibration for my home lab.
Performing such transfers is key importance to any calibration facility, be it official accredited laboratory or little hobbyist dungeon like mine. Metrology bodies like NIST, NIM, PTB doing same equipment swaps for their “travel” standards and even intrinsic Josephson Voltage standards to build up and maintain confidence in their equipment accuracy and stability. And using 3458A is second-best option (transfer of primary standards would be more accurate, but yet much more risky to ship over half-world) to do simpler transfers thanks to full calibration from just 10 VDC and 10 KΩ reference levels + AC/TVC correction for &mt;1MHz AC.
|Dead HP 3458A with chewed FP and no A5 PCBA||520$||190$||eBay|
|HP 3458A Outguard controller PCBA, Rev.B||400$||N/A||eBay|
|Dallas/MAXIM NVRAMs 1 x DS1220, 2 x DS1230||80$||N/A||Maxim|
|03458-40212 Front Panel Sub Assembly||$97.50||1|
|03458-47911 Rear plastic bezel||$76.23||1|
|03458-49331 Window, Front Panel||$26.00||1|
|03458-66513 TBR Analog to Digital In-guard PCA||$1306||1|
Table 9: Replacement parts orders
Total restoration cost for this meter: $2695.73 USD. Not cheap by any measure, yet still in range of secondary-market value.
|10/28/2016||Received unit, initial teardown and checks||3 hour|
|10/29/2016||Disassembly and diagnostic checks||2 hour|
|11/6/2016||Troubleshooting A2/A3||8 hour|
|11/11/2016||Test A1 with good A2/A3||3 hour|
|18/12/2016||Testing SN18 start with replaced A3||8 hour|
|18/12/2016||Testing SN18 start with replaced A3||8 hour|
|20/12/2016||Modified A9 to lower reference oven temperature with 15KΩ VAR||3 hour|
|31/12/2016||Meter confirmed good and stable, ready for calibrations||30 hours|
|04/01/2017||Calibration complete with F732B,SR104 and HP3325A+TVC+F8920A||20 hours|
|05/01/2017||Calibration report generated using 24h calibrated Fluke 5700A||8 hours|
|2016-2017||Documentation, photos, article write-up||45 hours|
Table 10: Timelog summary
I would like to express my gratitude and appreciation to supporters of this project: Todd Micaleff, Dr.Frank, Kleinstein, plesa and of course all our readers. Stay tuned, many more interesting ideas will be in testing, using fresh duo of Keysight 3458A and Keithley 2002’s real soon.
Modified: Oct. 17, 2017, 11:26 p.m.