HP 3458A — non-converging rundown, single-slope only: is this faulty U180?

[vf33184]

Hi all,

I've been working through a dead HP3458A I recently acquired and I'd really appreciate a sanity check from people who've seen similar failure, before I spend money on a replacement A3 board or NU180 alternative solution.
I've done a fairly thorough top-down trace of the A/D and everything points at the U180 hybrid — but I want to be sure.

The instrument
- HP3458A, series 2823A, build 1997
- A3 (A/D) board: 03458-66503 Rev D (U210 is the original Fujitsu MB651314, U180 is the HP 1SJ8-0108 hybrid)
- A5 outguard controller: 03458-66505 Rev C (single EPROM version), software revision 9.2
- Calibration Dallas DS1220Y memory is intact (battery still OK regardless almost 30 years old)

Symptoms / errors
The meter never produces a reading — it halts on the errors.

At power-up:
- ERR 202 — HARDWARE FAILURE: SLAVE TEST: CONVERGENCE
- ERR 110 — CALIBRATION REQUIRED: ACAL

On ACAL DC:
- ERR 209 — HARDWARE FAILURE: TIMEOUT: UNABLE TO READ A/D 61
- ERR 210 — HW FAILURE: CALIBRATION REQUIRED
- ERR 101 — CALIBRATION ERROR
- ERR 114 — SYSTEM ERROR: balanced rundown convergence

ACAL AC and ACAL OHM also fail.

The key observation I made:
The rundown is single-slope. The integrator (TP112) charges up, runs down as one straight ramp, the main comparator U142 EL2018CN (TP142) trips once at the zero crossing, and then nothing — the sequence never ping-pongs through the multislope steps, so it times out as non-convergence. This is unchanged when zoomed to 2 µs/div and unchanged at NPLC 100.


3-ch capture: CH1=TP112 (integrator), CH3=TP140 (slope amp), CH2=TP142 (comparator). Single integrate, single rundown slope, one comparator edge at zero crossing.


wider view: clean single slope, no slope reversals / no staircase, timeout at the end


schematic diagram with highlighted measuring points
 

my A3 board


What I've verified GOOD (top-side, in circuit)
- Inguard & outguard power supplies — all rails in spec (+18.6 / −18.6 / +5.0 V etc.)
- Integrator (U110/U111/U112, TP112) — ramps cleanly
- Slope amp (U140, TP140) — follows integrator
- Main comparator U142 (EL2018, TP142) — toggles cleanly, correct edge at zero crossing
- Minor-slope drivers U150 (74HCT14) — XS64 verified toggling pin 13 → pin 12 during rundown
- Clocks into U210 — CK20 = 20 MHz, CK10 = 10 MHz, both clean (U230 + U131 OK)
- U210 is alive — it communicates with A5; unplugging either optical link gives ISOLATOR FAULT, so the gate array core is running

What looks WRONG
1) ADC references are off and non-proportional (measured at the LT1001 outputs):
- +12ref (TP160) = +12.367 V
- −12ref (TP165) = −12.714 V
- +5ref (TP151)   = +5.501 V
Swapping temporarily U165 (LT1001) for an OP07 did not change these — so the LT1001 buffers aren't the cause. My understanding (from xDevs write-ups) is that the reference ratio-setting resistor network is inside U180, and degraded internal resistors skew exactly these voltages. Is that deviation outside spec / diagnostic of a bad U180?

2) The slope-steering command reaches U180 — but U180 doesn't act on it.
During the rundown the minor-slope drivers U150 (74HCT14) are toggling (XS64, pin 13 → pin 12), so U210 is issuing the slope-change commands and they are arriving at U180's steering inputs. Despite that, the integrator (TP112) produces a single uninterrupted ramp — the reference currents are never actually switched. So the digital steering command is present at U180, but the multislope response is absent. That should localize the break to the current-steering switches inside U180, not to a silent/dead U210.

3) Interpolator never starts — but I believe this is a consequence, not the cause:
- RAMP (U404 output, TP405) sits static at 0 V
- HTRIG (U405 output, TP406) sits static at ~0.1 V
- Traced back: LRAMP into U400A is stuck high (5 V) — i.e. U210 never issues the "start ramp" command. Since LRAMP fires only at the end of a completed multislope sequence, this is consistent with the rundown stalling after the first slope.

My conclusion (please critique)
Two independent symptoms point at U180:
(a) the skewed, non-proportional references — which U210 has nothing to do with, and which I understand are set by the resistor network inside U180
(b) the single-slope rundown — and critically, U180 fails to steer even though the steering command demonstrably reaches it (see U150 toggling above). U210 looks alive (comms, clocks, talks to A5), so the non-response points at U180's switches.

Rather than hunt for a scarce ~$2000 A3, I'd like to go the NU180 route, since it replaces exactly the two things I suspect — the precision resistor array (my skewed references) and the current-steering switches (my single slope). I am watching evolution of NU180 project here on eevblog forum.

My questions
1. Does this set of measurements really point at U180, or am I missing a test that would implicate U210 (or something else) instead?

2. Is the reference deviation above (+12.367 / −12.714 / +5.501) actually out of spec and a recognized U180 signature, or is it within normal range (or can be adjusted by successful ACAL) and a red herring?

3. Is there any non-destructive measurement (without probing signals at the bottom of fine QFP/hybrid pins) that would definitively separate U180 from U210 before I commit to removing U180?

4. NU180 fit for my failure mode: the project mostly describes drift; mine is a hard single-slope / no-convergence with the ERR 114. Is a "complete failure + skewed references" case like mine a good NU180 candidate, i.e. should fitting NU180 restore both the references and convergence? Has anyone resurrected a non-convergence unit (not just a drifty one) with it?

Thanks in advance — I've tried to do the legwork rather than just ask "which board do I buy," and any expert steer, before I move to more complicated and costly exercise, would be hugely appreciated.

Thank you.

Vladimir

[TheDefpom]

Your diagnostics process is useful, I also have a unit with bad A3 but is not even trying to generate any slopes, so your testing information is going to be helpful when I dig into my unit.

I have built a NU180 for it but it did not change anything, so I am suspecting an issue with the board somewhere as well as U180, which is why your information is helpful.

Unfortunately I do not know enough about this unit yet to add any information to help you in return, apart from agreeing that your suspicion of the 12V/5V references look incorrect.

This is what I was getting on my bad board, before it failed completely.

Code: [Select]

TP160 12V Ref         11.89970V
TP165 -12V Ref       -11.90050V
TP151 5V Ref          4.958717V
TP142 Comp. Out       4.15649V


[Kleinstein]

The measured reference voltages are definitely wrong. One may want to also measure the voltage from the A9 reference and check if there is some oscillation going on at the reference voltages. Just in case also check the supply to the OP-amps, but a doubt this would be wrong so much.
The symmetry between the +-12 V readings and the 12 V to 5 V ratio should be much better (like < 0.1% from nominal for the ratios).
ACAL can not change anything with the voltages - they are set by the hardware with nothing the software can change.
ACAL can only compensate for the overall voltages, like if all are 1 or 2 % high. AFAIK there is no numerical correction for errors in the ratio and they need a good accuracy to get good DNL.
The resistors for generating the reference voltages are a part of the U180 that could in theory still be replaces with external resistors. However with these resistors bad there is a good change for more defects.

The run-up part should be more than single slope  (single slope runup may happen only for the very short (e.g. < 20 µs)  integration time).
The scope picture (2706) does not looks like the rundown part, more like the start of run-up at about 3 µs before the trigger and the end of the reset phase at about -7.8 µs.
Wider view shows that the run-up is not working correctly - somehow the integrator saturates at the negative side, as if the negative reference is not activated. This ref. part is active at the very start - so the U180 part for this seems to work. So there may be an issue also with the control (U210) or the comparator signal not reaching U210.
It looks like parts of the rundown is working, at least the comparator switches several times, but the resolution is not there to see details.

The XS64 slope part is switched by the HCT14,  U180 only has a switch to connect the fine slope part (for the rundown) and disconnect it (for the run-up).

The interpolator is not directly linked to the ADC. It is more related to the S&H stage used for digital AC measurements. The normal DC mode readings don't use this part.

[vf33184]

Hi Scott,
thanks for jumping in and sharing your readings — and good luck with your own unit!

Interesting thing about your reference numbers: they actually look really healthy, and I think that tells you something useful about your fault.

As I understand now, based on the comment from @Kleinstein, the trick is that the absolute "12 V / 5 V" aren't the real target. The references are derived from the A9 LTZ1000 (~7.0–7.1 V, and every reference reads slightly different) through fixed resistor ratios inside U180. So what matters isn't whether they hit 12.000 exactly — it's whether the ratios between them are tight.

Yours are:
- +12Vref / −12Vref symmetry (should be exactly 1.000, set by the 10k:10k inverter): 11.9005 / 11.8997 = 1.00007 → 0.007 % off
- 12Vref / 5Vref ratio (nominal 2.400): 11.8997 / 4.958717 = 2.3998 → 0.008 % off

Both essentially perfect. All three of your references also sit about 0.83 % below nominal by the same proportion — i.e. they're shifted together with the ratios intact. That's exactly what a healthy resistor network should look like when it's just running off a reference that reads a hair low (or a nominal that isn't quite 12.000). And a uniform offset like that is what ACAL is designed to null out, so it's harmless.

For contrast, mine are off by different amounts (≈+3.7% / +6% / +10% from nominal), so the ratios themselves have drifted — 2.8 % on the +12/−12 symmetry and 6.3 % on the 12:5 ratio — which ACAL can't fix, as confirmed by @Kleinstein. That's the most likely "bad resistors inside U180" signature.

So I think your U180 reference network part is actually fine — the ratios are spot on, only a benign common offset. That's consistent with NU180 not changing anything for you: if your reference die is healthy, your fault is somewhere else (control side / U210 / comparator path rather than U180). Given your unit "isn't even trying to generate any slopes," I'd be chasing whether the run-up phase starts at all, the U210 clocks (CK10/CK20), and whether the comparator edge actually reaches U210 — same chain I've been working through here.

Happy to compare notes as we both dig in.

Code: [Select]

TP160   12V Ref         11.89970V
TP165   -12V Ref       -11.90050V
TP151   5V Ref          4.958717V
TP142   Comp. Out       4.15649V


[Kleinstein]

U180 has more internal resistors than just the one to create the +-12 and 5 V reference. So even if these resistors are good, other resistors could still have a problem. AFAIR TiN had a bad U180 with slightly off references and he tried it with extermal resistors to set the references. The meter still failed.
The current theory behind the U180 failures is that the hermetic seal fails and this makes the resistor drift, and not all at the same speed. So if the reference ratio is bad chances are high that the other resistors (e.g. 80 K for the references) are also drifting.

[vf33184]

Thanks Kleinstein — this is very helpful. Follow-up split into the two points you raised.

1) References: stable, no oscillation

Measured the A9 reference at the A3 input and the buffer outputs:
- A9 reference (ZR_HI−ZR_LO at J1, and across U160/U165 pin 3 inputs): +7.0588 V, stable.
- +12ref (TP160), −12ref (TP165), +5ref (TP151): DC-stable, and on AC coupling at high sensitivity there is no oscillation on any of them — flat, no ringing on the buffer outputs.

So the references are quietly sitting at the wrong (and non-proportional) values rather than oscillating. Combined with your note about the hermetic seal failing and the internal resistors drifting at different rates — that matches what I see: the +12/−12/+5 ratios are each off by a different amount (~3.7 / 2.8 / 3.8 %), not a common shift. And per your point about TiN's unit, I'm treating the reference resistors as just the visible symptom, with the steering resistors (80k etc.) very likely drifted too.

2) Run-up / rundown — re-interpreting the earlier "saturation"

Capture method: the meter halts on the errors and doesn't free-run conversions, so I arm the scope on single trigger and then start a self-test on the 3458A — that produces the conversion attempt I trigger on. All captures below are from that single-shot held in scope memory (so I can re-scale V/div and timebase and zoom on the stored record). CH1 yellow = TP112 integrator, CH2 cyan = TP142 comparator, CH3 purple = TP140 slope amp.

You were right that picture 2706 was not the rundown. Re-reading the whole conversion at progressively finer timebases:

What I read now is that the converter is running a normal multislope cycle, and the earlier "saturates to the negative side" was an artifact of the 10 ms/div view collapsing the fine staircase into a flat line.


[2ms/div]


[1 µs/div]  Clean symmetric staircase: stepped run-up, peak, stepped rundown — not a saturated single ramp.



[2 µs/div]  Same, with the comparator (cyan) toggling around the cycle.



[500 ns/div]  Dense burst of narrow steps on TP112 — the fast multislope ping-pong; the comparator switches many times.



[200 ns/div]  Stepped run-up with the per-step structure resolved.



[50 ns/div]  Individual comparator transition (cyan) alongside single integrator steps.

So: run-up and rundown both step, the comparator switches many times, and U142 looks fast and clean (no lazy/slow edge). My reading is the A/D mechanically works, and the non-convergence is the references/steering ratios being off — consistent with your "other internal resistors drifting too" point — rather than a control-side (U210) fault. Does that match your interpretation?

3) One thing I can't explain — U140 gain

With TP112 and TP140 on the same 200 mV/div and aligned from the single-shot memory, TP140 does not show ~13× the integrator step amplitude — it's closer to ~1–1.5×.



[20 ns/div, CH1=TP112 200 mV, CH3=TP140 200 mV] The two staircases are nearly the same height.


Per the schematic U140 (LM6361) with R141 12.1k / R140 1.0k should be a gain of ~13. Is the near-unity amplitude expected here because of the CR140/141/144/145 clamping around U140 (i.e. it only delivers the high gain in a narrow window around the zero crossing and clamps otherwise), or would you read this as a faulty U140? I'd rather not chase it if it's normal clamp behaviour.

(My 'scope is 50 MHz, so I'm not over-reading edge slopes.)

Thank you.

Vladimir

[Kleinstein]

The new scope pictures show artifacts from zooming in too much with capured data. The steps are the quantization steps from the scope.

The phases run-up / run-down are not from the direction the integrator moves. Run-up is the phases where the input active and for the MS SDC the referenes switched to keep the integrator within bounds. The rundown phase has the input no longer active and going back the zero voltage in steps. The run-up should have a triangle kind waveform and the slopes (~ +-1 V/µs) at around the center in the 2706 picture is about right. So this part looks like the start of run-up, but still not right. AFAIR the phases should be some 3 µs and 200 ns.
However the waveform continues down to saturation all the way to -15 V. So there is something no working with a the run-up.
The slope up and down look like they come from the negative and posive reference setting with likely 0 V at the input.

The picture also shows the gain of the slope amplifier (U140). At around zero the pink curve has the gain and as designed the gain than goes to about 1 for larger voltage. The short peak at about -8 µs (from the center / marker) looks like the end of reset / connection of the input. This can come with some charge injection and thus the small step, that is visible with the slope amplifier, but not yet at the integrator directly.

The 2707 picture seems to still have the run-down part, about 200 µs later, and it seems to converge to zero, the slope amplifier output somewhere in the middle and the comparator switching a few times. However no details visible.  So it may also be the reset getting activated.

[vf33184]

Hello @Kleinstein.

thank you for clearing up the scope artifact confusion, and good to know U140 is as designed.

Quick recap of where I've landed: references stable but ratios drifted non-proportionally (~3.7 / 2.8 / 3.8 %), run-up saturates to −15 V, while U210 (comms + clocks), U142 and U140 all check out — pointing at U180 along the lines of your hermetic-seal theory.


A few questions to you:

1) Would you call this a confirmed U180 failure, or is there still a realistic chance it's broken U210 and/or control side?
 
2) Is there any other measurement that would separate a U180 from a U210 fault? If so I'll happily do it.
   
3) If it is U180 — would you expect a NU180 to restore both the references and the run-up, since it replaces the whole network? Or could the run-up saturation be something NU180 wouldn't address?


I am just trying to reach a go/no-go on NU180, i.e. whether I should order PCB, components, etc.


Thank you very much again for all the input.


Vladimir

[Kleinstein]

The wrong references clearly points to an U180 issue. This is especially with U165 (LT1001) exchanged and no oscillation.
To be more sure, maybe check the supplies for U165, U151.

The integrator going to saturation is a bit odd. There is a chance the U210 also has problems. A point to check would be if the comparator signal really reaches U210. So look at pin 48 of U210. If the comparator signal does not get here, U213 could also have an issue, but this would be an easy to get part at least. If the comparator signal is there, this would make U210 a bit suspect - though it could still be an odd selftest step, that is supposed to run to saturation.

If U210 is OK the NU180 should fix the error messages, would not bring back the full performance. This noise so far is higher and the INL error may be higher - one would at least need a test.

[vf33184]

Kleinstein — on the comparator-to-U210 check you suggested (pin 48): on my Rev D board there's no U213 in that path (it was removed in later revisions), and I've confirmed continuity straight from TP142 (U142 output) to the U210 pin. So the comparator signal is galvanically connected to the gate array with nothing in between to gate it — I'm taking that as "the comparator does reach U210." 
I checked U165 and U151 power supply rails again, they look ok, just U165 has a bit bigger power consumption than U160 and U151. See the table:


With that, here's where I've got to, and where to go next — would welcome opinions.

Summary of the diagnosis: references are drifted with non-proportional ratios (pointing at U180's internal resistor network), the run-up saturates rather than balancing, while U210 (comms + clocks), U142 and U140 all check out.

The honest limitation: I don't have a second, working 3458A. So I can't swap boards to narrow the fault down to A3 with certainty, and I can't run a good unit next to the bad one to compare what's different. That's the main thing shaping my options.

The options as I see them:

1. Build an NU180 and try it. Cheap, and it doubles as a diagnostic step — if the meter starts converting, A3/U180 was the (only) fault; if not, the problem is elsewhere (e.g. U210 or another board). Downside: it won't match original A3 performance, so it's more "prove the fault and get it running" than a final metrology fix.

2. Send it to Keysight. Would definitely be repaired + recalibrated, but very expensive — and there's the risk they find faults beyond A3, the quote becomes silly, I decline, and I still pay the repair-attempt fee. (And U180 itself is unobtainable, so they'd swap the whole A3 anyway.)

3. Independent cal/repair house (e.g. ab-precision.de). Closer (EU) and likely cheaper than Keysight, but I don't know the outcome — they'd probably also want to replace A3, and whether they can source one any cheaper than I can is unclear. Or anyone know other options (company or person within EU) who would be able to perform repair more reasonable than KS?

4. Find another A3 board. I assume anything I can find on sources like eBay is likely to be faulty too, so that's a gamble again. (And both customs — U180 and U210 — are only really sourceable from a donor A3 anyway.)

5. Sell it as faulty and move on.

6. Keep it as a shelf decoration.

My plan is to start with option 1 — it's the cheapest way to actually answer the open question (is it only A3, or is something else dead too?) before I commit money to any of the others. Until I know that, options 2 and 3 are a gamble: I'd hate to pay a big repair attempt only to find multiple faults.

Any other options I'm missing, or experience with the repair routes above? And if anyone has a spare/working A3 they'd part with, I'd be interested.

[IanJ]

My 3458A has a convergence error on startup. No display, other than the units and when I set TARM AUTO I get a display albeit it needs TIMER set to 0 to speed it up (seems to be stuck at 1sec irrespective of low NPLC).
At this point, it fails ACAL AC with a timeout error but passes ACAL DCV and ACAL OHMS.

I looked at the +/-12Vdc ref voltages and recorded the following:
+12.152V
-12.171V
+5.0918V

So, thinking the +/-12V aren't matched close enough (0.156% diff) and just to explore.......I hooked up an external +5V source (PDVS2mini) through a 1K resistor to pin 3 of U165 and effectively sent a few mA down ZR_LO2.
ZR_LO2 has a 0.5ohm resistance down to the main A.Gnd on the A3 (the path back to the A9 board and catching a few friends on the way).

The result is that the -12.171V now reads -12.168V (0.132% diff) and the meter now boots clear of any errors and has a great readout without requiring TARM AUTO or TIMER 0 etc. The +12.152V didn't change IIRC.
I haven't powered it up for any length of time to see how it performs. But it consistently boots just fine.

Next step, I'm building a small daughter board with my own precision +/-12Vdc and injecting that into the U180......and wondering how that will perform.

Making a YT video about this........

Ian.

[vf33184]

Quick update — I tried the external-resistor reference-correction experiment (à la Ian / TiN) to see how far it gets on a heavily-drifted U180.

My starting references:
- +12ref = +12.367 V
- −12ref = −12.714 V
- +5ref  = +5.501 V

So the ratios were off non-proportionally: +12/−12 symmetry ~2.8% out, 12:5 ratio ~6.3% out — much worse than Ian's ~0.16%.

I corrected the two gain stages back to their design ratios with parallel resistors across the internal feedback 10k's:

1. U165 (−12ref inverter, gain −1): added ~368k in parallel with the internal 10k feedback. That brought the gain to exactly −1.000 (12.385 V in, −12.385 V out). Interesting that it only took correcting essentially one resistor — looks like a single drifted 10k, or leakage through the 20k/80k network rather than a uniform shift.

2. U151 (+5ref, ratio 2.4): added ~294k in parallel with its feedback 10k. That set the +12:+5 ratio back to exactly 2.400.

Both reference ratios are now spot-on by design.

Result: the meter still fails self-test / still won't converge.

So on my unit, correcting the reference ratios is not enough — consistent with what Kleinstein mentioned about TiN's unit. It backs up the idea that the other internal resistors in U180 (the ones setting the steering/slope weights, not just the references) are drifted too, which external resistors on the references can't fix.

My takeaway / data point: small drift (Ian, ~0.16%) → external reference tweak can get it running; large multi-ratio drift (mine, ~2.8/6.3%, and TiN) → reference correction alone isn't enough, the whole U180 network is gone. So I'm taking this as confirmation that a full U180 replacement (NU180) is the way forward rather than a reference patch.

Next step: building an NU180 (now considering whether to go for DB4UCH V0.4.2 or wait for the newest V0.6.0) and trying that.

[Zondar]

Next step: building an NU180 (now considering whether to go for DB4UCH V0.4.2 or wait for the newest V0.6.0) and trying that.

V0.4.2 should work for you if you want to proceed right away. Per Kleinstein's comment, the noise is 20-30% higher than spec, and there are some remaining relatively minor (but very puzzling) issues with offsets and linearity.

V0.6 has many details reworked, it's close to ready, and it should be at least as good. If V0.6 is your choice, I'm willing to help with modifications, e.g. per your preferred or available resistor types.

[vf33184]

Thanks, that's really helpful — and thanks for the offer to help with V0.6, much appreciated.

Let me lay out how I'm thinking about it. My priority right now is simply to find out whether an NU180 brings my meter back to life, or whether the fault is somewhere else as well — I don't have a second working 3458A to swap boards against, so building one is my way of narrowing it down. For that diagnostic purpose I'd populate the board with ordinary 0.1% thin-film resistors, since noise and the minor offset/linearity issues don't matter at this stage — I just need to see the errors clear and the meter convert.

If it does revive the meter, then I'd want to go for maximum performance as a second step — and that's where I'd aim at the best resistors (foil etc.). V0.6 looks really nicely done for exactly that: the multiple footprints supporting SMD or through-hole is great for fitting the best parts.

The one thing shaping my decision is timing — I'd like to have some version built by the end of July. So my question is really about V0.6 readiness: do you have a rough feel for when the design will be ready? If it's likely to be ready soon, I'd happily go straight to V0.6 with your help on the resistor types and skip building two boards. If it's more open-ended, I'd lean towards building V0.4.2 now to answer the "is it U180 or something else" question within my timeframe, and then potentially do V0.6 as the proper build later.

Either way I appreciate the input — just trying to pick the path that fits my target.

[IanJ]

The curious thing which I haven't got to the bottom of yet, is that if I adjust the feedback resistor on U165 so that I get:

TP160 = +12.152V
TP165 = -12.154V

ZR_LO2:
With no injection = -0.249V
With injection = +2.64mV

Then I still get the convergence error on startup. It's only by hooking up a +5V source through a 1K resistor into ZR_LO2 that I get an error free power up.

Ian.

[TheDefpom]

Ian, Maybe the voltage is noisy from the flaky resistor, tried scoping it?

[IanJ]

The injected current is correcting a large negative imbalance at the zero/reference low summing node.

ZR_LO2:
With no injection = -0.249V
With injection = +2.64mV

Thinking that the convergence failure is likely because the ADC zero/current-steering loop cannot pull ZR_LO2 close enough to zero by itself.

I need to scope the control signals next and see if it's trying.

UPDATE:
With injection in place the meter now passes ACAL AC.
With injection in place, the meter loses some accuracy, i.e. from my PDVS2mini 1.000000V without injection then becomes 1.0002V with injection.
So, I suspect, my +/-12V has always been fine (because CAL is spot on) and possibly never drifted, it's possibly one of the switches in the U180 that's failing or not operating. More tests required.

Ian.

[Kleinstein]

The rather large change in ZR_lo2 suggest that there is too high a resistance in the path. This could be oxidized connector. The path should go via A3 - A1- A9. The are also connectors for the ground. Overall the resisance to ground should still be more in the 0.1 ohm range.

[Zondar]

So my question is really about V0.6 readiness: do you have a rough feel for when the design will be ready? If it's likely to be ready soon, I'd happily go straight to V0.6 with your help on the resistor types and skip building two boards. If it's more open-ended, I'd lean towards building V0.4.2 now to answer the "is it U180 or something else" question within my timeframe, and then potentially do V0.6 as the proper build later.

"'Finished' is a state of mind." 

For the purpose of demonstrating functionality using less-expensive resistors (a wise move), V0.6 is already fine and "done."

However, most of the important resistor footprints are currently 1206's. If you want to or have to use other sizes, then some of those will need to change. That's easy since virtually all alternates will be smaller. So depending on that, there may be an hour's worth of work to do. Unless you are a KiCad expert, or even if you are, I'm happy to make those changes for you.

You can contact me directly to discuss this further.

[Kleinstein]

From another 3458 replated thread (https://www.eevblog.com/forum/metrology/hp-3458a-repair-convergence-error-with-wierd-workaround/msg6287780/#msg6287780) there is the indication that there is a separate test for the reference ratios (as part of the ADC) as part of the self test / internal test.
This may may well be not a normal conversion, but a more odd sequence. The one "conversion" you see may not be a normal conversion, but that self test part.  Chances are it would still be similar, but can still be different (e.g. different timing) to be a little more sensitive to reference ratios.