[ §1 · resonator ]

What an SSTC topload does, what it costs, and what a QCW changes

Not an upgrade, a trade. You pay in volts, and volts are what make the arc long. Whether it is worth it depends entirely on how much your arc detunes you.

CommonGnd4 August 202634 min read

A topload is not an upgrade. It is a trade, and the thing you pay with is voltage, which is the thing that makes the arc long.

What it is and why

Z = sqrt(L / C),   and volts at a given power go as sqrt(P · Z)

More capacitance, lower impedance, fewer volts. What you buy with those volts is a resonant frequency that does not run away when the arc grows.

So the whole question is how much your particular coil detunes, and that is not a property of toploads. It is a property of the machine underneath one.

What you decide

Whether to fit one at all

Criterion: how far the arc drags the resonance.

A slayer exciter at a few watts has nothing to detune. A point, and nothing else.
A 4 MHz HFSSTC at 500 W is stable enough that Richie Burnett runs his resonator with no toroid at all, having "not found large toroids to have a beneficial effect on CW coil designs". His experience is at megahertz; carry it to a 300 kHz CW coil carefully.
A DRSSTC detunes noticeably, and wants one.
A QCW throwing a metre detunes catastrophically, and goes the other way entirely: capacitance is added, sometimes inside the secondary, so that the arc is a smaller fraction of the total.

One toroid or a stack

Small below, large above, and the two do different jobs. The small one screens the top turns of the secondary, where the potential is highest and the gradient sharpest. The large one carries the capacitance and holds the breakout, which decides where the discharge goes. The lower one takes the induction heating, being nearer the primary.

The spacing is the knob, and it does not behave the way a calculator says: see the next section.

How much clearance it gets

Three clearances, and they are not the same requirement.

A gap to the primary of about the secondary's radius. This one is about heat: the toroid sits in the primary's field and heats inductively, and the gap keeps that down. Slot the toroid as well, so it is not a shorted turn. This page used to say Steve Ward slots his; six of his own build pages have been read since and none of them mentions slotting a toroid, so the attribution is withdrawn and the reason stands on its own.

Height above the winding of about the same. This one is about where the discharge starts. Sit the toroid too low and its own field falls back onto the top turns, where the potential is highest, so the breakout begins there instead of off the point.

No other metal closer than half the secondary's diameter, anywhere. This one is about a stranger in the field: a conductor in the near field loads the resonator and drags its Q down, wherever it sits.

And the fitting that joins them is a fourth object, not part of the third. The clamp or lug at the top of the winding sits at the highest potential on the coil and is the sharpest thing there, so it breaks out before anything you designed to. The one builder in this corpus who chased it reports the fix as a distance rather than a shape, "No flashovers due to increased distance between topload and connection clamp", and davekni's suggestion in the same thread is to remove it as an object instead: "That connection could be insulated, or even an entire horizontal plastic layer (big washer) could fit (and seal to) the primary coil form" (HVF 1988). Either way, measure the clearance from the metal that is actually there and not from the toroid's underside.

Bigger is harder to break out of, and throws longer

A larger toroid resists breakout, and the arc it throws when it finally goes is longer:

Steve White, posting as MRMILSTAR on HVF 1937: "As the radius of the toroid gets larger it becomes harder for the streamer to break out but when it does it will be longer."

Why is not settled, and he offers it as speculation rather than as a result. The thread leaves two readings side by side. His is field shape: a larger radius spreads the surface field, so nothing breaks down until the voltage is higher. Marek Novotny, first in the same thread, gives the other, stored energy: a larger toroid holds more charge, which helps the air break down and pushes more current into the spark as it loads. He adds his own doubt about it, that coil streamers are far from the DC discharges the argument is true of. This is a different axis from the stack not adding up. That was about the number a calculator gives; this is about the field and the stored energy, which no capacitance figure captures.

What will get you

This is an argument against the calculator, not against the stack. Stacks work and are common. Measure the assembled secondary's resonance instead, and expect the calculator to read high, the more so the closer the bodies are.

And the skin effect argument about seams is right for the wrong reason. It is widely repeated that the seams of a home-made toroid need not be properly joined, because of skin effect. What actually carries the current across a seam is the capacitance between the overlapping sections. While that is large compared with the topload's total capacitance the voltage across the seam is small; when it is not, you get arcs between the sections and less stability than you had. In practice, fold one or two millimetres of tape over the edges so foil meets foil directly rather than through the adhesive.

The third job: how long the arc can get

It is not an accessory. It is most of the resonator

On one coil, 150 mm of winding on a 90 mm former:

Ces total             14.031 pF
of which the toroid   10.879 pF   =  78 %
the winding itself     3.152 pF   =  22 %

That 22 per cent is not a constant, and the builder who tried hardest to pin it says why it cannot be. davekni needed a topload-only figure to turn his charge captures into a top voltage. His coil's frequency before breakout matches 32 pF in total; he first guessed 28 pF for the topload alone, then revised to 27 on a measured top to bottom current ratio of 84 per cent. That leaves 4 to 5 pF in the winding, 12 to 16 per cent [derived], against the 22 here. His caveat is the useful part: measuring the frequency without the topload does not give you the winding's capacitance with it, because that capacitance is partly to ground, partly to the topload and partly between turns, and once the arc forms some of the topload's own share moves from ground to arc. He says outright that he does not know how to estimate it. So the split is a modelling choice on every coil that reports one, the 22 above included.

So the frequency, the impedance and the coil's tolerance of arc loading are all set by the part on top, and the winding contributes 22 per cent. Which is why "tuning the secondary" means "choosing the toroid" in practice, and why a rewind is almost never the right answer to a frequency you do not like.

The demonstration that used to sit here has been withdrawn, and it is worth saying what it claimed. It printed two versions of one published build, an 8 by 2 inch toroid at 406 kHz against an 11.25 by 3.4 inch at 343 kHz, with impedances of about 48.1 and 40.7 kΩ fitted in JavaTC, under the words "same winding, same wire, same turns". The build is Gao Guangyan's QCW 1.5 and his own log denies most of that. Experiment one ran a new secondary of 3.5 by 6 inches, about 840 turns of AWG 34, under the small toroid; for experiment two he went back, in his words, to the original 32 AWG 5.55 inch secondary from his QCW 1.0, and built a new primary as well, so the winding, the wire, the turns and the primary all moved with the toroid. His measured secondary resonances then run the other way from the table: 302.7 kHz standing alone with the 8 by 2 inch toroid and 339.3 kHz with the larger one, because the shorter winding underneath more than paid for the bigger part on top. Neither 406 nor 343 kHz is a secondary resonance anywhere in that log; the figures near them are pole and operating frequencies, and the ones near 408 to 413 kHz in that log belong to the fat-toroid setup rather than to the small one, which is the toroid the table put 406 against. What this note said next was too strong. It read the two rows as a different machine each, and that holds only if the small-toroid row came from this log. It did not. It came from QCW 1.0, whose 32 AWG 5.55 inch winding is the one he goes back to here, and which he measured standing alone at 408 kHz in 2014 against the 339.3 kHz the fat toroid gives it in 2015. So the toroid is isolated on one resonator after all, across two of his pages rather than inside this one: a drop of 16.8 per cent [derived], with a year and a change of room in it. The secondary is a resonator carries that pair and its caveats, and this page's rows are still gone, because a fitted 406 and 343 with impedances beside them are not those readings. The section above stands on the 78/22 split alone. That split is JavaTC's.

Capacitance and breakout are the two jobs everybody knows. The third decides whether the coil is still working at the end of the ramp.

An arc is a load, and a growing arc is a growing load. Past some conductance it damps the resonator enough that the two poles merge, and the drive loses the thing it was tracking. The topload sets where that happens. The Z in it is the secondary's own characteristic impedance, which the secondary's page works out at 26.6 kΩ for this coil, and it is that page's number rather than this one's:

the poles hold while   R_arc  >  Z / k
here                   Z / k  =  26.6 kΩ / 0.421  =  63.2 kΩ

Feed in Anders Mikkelsen's arc model, about 75 kΩ·m in parallel, so that a longer arc is a lower resistance, and the threshold becomes a length. Read the note under the table before the metres: that constant is the weakest thing on this page.

Toroid	`C`	`Z`	Threshold	Collapse at	Frequency
×0.5	8.6 pF	34.0 kΩ	80.7 kΩ	0.93 m	545 kHz
×1	14.0	26.6	63.2	1.19 m	426
×1.5	19.5	22.6	53.6	1.40 m	362
×2	24.9	20.0	47.4	1.58 m	320
×3	35.8	16.7	39.6	1.90 m	267

Twice the toroid, and the arc reaches 1.58 m instead of 1.19 before the structure gives way. That is the third end of the trade: a topload is not only capacitance and not only a field shape, it is how much arc the resonator will tolerate.

Those metres are derived, not measured, and the constant under them is one builder's. L_collapse = 75000·k/Z is our algebra over Anders Mikkelsen's arc model, 8 pF per meter in parallel with 75 kohm*meter, fitted to his own measured phase angles, 50 degrees near 340 kHz and 60 near 460 (HVF 3140). This page said the constant had no source. It has one, and finding it does not rescue the metres, because the source is the outlier: 75 kΩ·m puts a 1.5 m arc at 50 kΩ, which is below every other published arc resistance once they are converted to the same parallel form. There is no secondary frequency does that conversion and gets 77, 79, 103, 225 and 258 kΩ at 1.5 m on this coil, all five of them above the 63.2 kΩ threshold, and says so in as many words: all five leave the poles alive at a metre and a half. On the harshest of the five the collapse falls at 1.83 m rather than 1.19 [derived, 77 kΩ at 1.5 m is 115.5 kΩ·m, and 115500 · 0.421 / 26600]. The 2.81 m this page used to offer as the soft end of "the models in circulation" is not one of them either; it is this corpus's own figure from the pole-structure calculation on that same page. So take the ratios from the table and not the metres. The whole band moves by one factor, and where it sits is unsettled until this page's constant is either sourced or replaced by the converted one.

Total secondary capacitance, where the winding's share is fixed, so this slider is the toroid

3.2 pF, bare winding14.0, this coil40 pF

Drag the toroid. The winding's few picofarads never move; everything above them is the topload, and it carries the frequency and the impedance down together.

Assumptions. Les 9.925 mH and Ces 14.031 pF from JavaTC, the toroid supplying 10.879. The winding's 3.15 pF is held constant across the slider. The slider is capacitance, not toroid size. Two real toroids of quite different dimensions came out within one per cent of each other in JavaTC, so a size axis would have been an invention. The drawn toroid is scaled from the capacitance and is indicative only.

The same lever, without the aluminium

Read the formula again and it says C. It does not say toroid.

Z = sqrt(L / C)

Anything that adds capacitance between the top of the winding and ground moves Z, and moves the collapse length with it. A string of ceramic capacitors from topload to ground does exactly that. It is on the largest coils currently being designed, where a secondary capacitor bank of C0G ceramics does the job a toroid would otherwise have to. Electrically the two are the same lever, with the same Z, same frequency, same collapse length:

Added	`C` total	`Z`	Frequency	Collapse at
none	14.0 pF	26.6 kΩ	426 kHz	1.19 m
+3 pF	17.0	24.1	387	1.31 m
+6 pF	20.0	22.3	357	1.42 m
+10 pF	24.0	20.3	326	1.55 m
+14.3 pF	28.4	18.7	300	1.69 m

The 300 kHz edge is the same edge. Capacitors do not get around it; the frequency falls for the same reason either way. What changes is what you pay:

	Buys	Costs
Bigger toroid	field control and breakout come with it	space, and space between primary and secondary is coupling
Capacitor string	no extra size, `k` untouched	current, and the string has to survive it

On a QCW the topload is not free real estate: it needs clearance from the primary and height above the winding, and a coil built for coupling has a short, fat secondary with the primary right against it. A toroid big enough to matter pushes the geometry apart, and geometry is k. A capacitor string lowers Z without touching a dimension.

It is not a free toroid. The string has no breakout, shapes no field and does not screen the top turns. The toroid can shrink to a small field-control ring, but it does not leave.

This part is our own reasoning, not a cited source. A toroid's capacitance is its field in the room: the current that leaves it returns through the floor, the walls and whatever else is there. A capacitor string carries the same current down a wire instead. So a smaller toroid plus a string gives the same total C with less of it coupled to the room, and the return path becomes something you chose rather than something the room decided, the same idea as the capacitance that does not care about the room. What that does not buy is quiet: the topload still sits at the same voltage, so its near field is unchanged; what changes is where the return current goes, and how repeatable the capacitance is. And the string alone changes nothing until you shrink the toroid to match: put it beside the same toroid and you have only added capacitance.

The current is real. Six picofarads at a hundred kilovolts and 357 kHz is about 1.35 A through those ceramics [derived, I = 2πfCV]. The practice this page describes, C0G only, 3 kV parts, derated to about half for AC, carries no citation here and is recorded as unsourced rather than as published. On those figures it is roughly 67 parts in series to stand off the topload [derived, 100 kV over 1.5 kV a part], so a 330 pF string comes to about 5 pF [derived, 330 / 67], and useful capacitance means several strings in parallel. Mount on rigid board, pot each in its own tube so a failure is separable, and degas the epoxy.

Coupling carries a contradiction of its own across the ramp, worth setting down even though it is not settled. Worked at one bridge voltage, the gain each pole draws changes sign with the load: ``

 k =         0.2    0.3    0.4    0.5    0.6 light 1 M   104.6  105.0  105.5  106.1  106.8   higher k, higher gain heavy 50 k   92.8   90.0   88.3   87.5   87.3   higher k, lower gain

  Strong coupling helps at the start of the ramp, where the load is light, and hurts at the end, where it is heavy, because it reflects the heavy load into the primary and cuts the

Q`. This corpus otherwise treats coupling as pole-structure margin, a different axis that pulls the other way. One reading, ours and not yet checked against a source, is that a QCW wins on the duration of the energy it pours in rather than on peak gain, and coupling is what sustains that duration. But that reconciliation is a guess until it is measured or found stated somewhere, so it stands here as the open contradiction it is. The table is our own computation.

On a QCW, where the same cost buys something else

The metal ring on top of the secondary. It does the same three jobs here as on any coil, and then one more that only shows itself once an arc is a metre long: it sets how much arc the resonator will carry before its two poles damp together into one and the driver loses the structure it was tracking. Every coil that grows a long channel meets that limit. What is different here is what crossing it costs, and that difference is the shape of the whole department.

The three jobs

It stores charge. A discharge is charge leaving, and a topload is the reservoir it leaves from. Without one, the resonator has only its own winding capacitance, which is a small part of what a topload adds.

It sets a good part of the frequency. That capacitance and the secondary's inductance are the resonator. Change the topload and you have changed the coil's frequency, which means you have changed where the primary has to be tuned.

It holds the breakout and keeps the discharge off the winding. The field on a smooth toroid is spread out, so nothing breaks down until you put a point on it. Without a topload the discharge leaves from the top turns of the secondary and burns the wire down.

And why this department buys it anyway

A metre and a half of arc adds about 6.2 pF to the secondary derived, 4.14 pF per metre times 1.5 m, and that rate is [one coil's calibration rather than a property of arcs], against the toroid's own 10.879 and a cold Ces of 14.031. So the arc puts another 44 per cent of the cold resonator's capacitance on top of it [derived, 6.2 / 14.031], and the resonance moves a long way while the arc grows.

How far, exactly, depends on which frequency you mean, and there are two:

The secondary's own resonance falls 16.7 per cent across the ramp.
The upper pole, which is what the bridge is actually commutating, falls from 519.1 kHz to 463.3, which is 10.7 per cent. It moves less because the primary sits where it was tapped, at 349.9 kHz, and does not move at all, so a pole made out of both resonators is anchored at one end.

The number the tracking has to follow is the 10.7, and it is not a rounding of the other. Both, and the argument for keeping them apart, are the frequency slides down the ramp.

The bigger the resonator's own capacitance, the smaller a fraction of it the arc represents, and the less the frequency moves. Double the toroid, taking Ces to 24.9 pF, and the same 6.2 pF of arc is 25 per cent rather than 44, and the secondary's fall drops from 16.7 per cent to 10.5 [derived, sqrt(24.9 / 31.1)]. So a QCW takes the voltage penalty on purpose, and then goes further: capacitance gets added inside the secondary as discrete parts, so that it arrives without the interlayer voltage stress that winding for capacitance would bring.

The attribution for that last move is Mads Barnkob's:

Mads Barnkob (HVF 1914, 31 January 2022): this is why Steve Ward added a series of small capacitors inside his secondary coil. His first QCW versions simply needed more self-capacitance to withstand detuning from the long sparks.

He is answering Uspring in the post directly above, who had just written that once you have breakout the arc adds capacitance to the secondary tank, which lowers the secondary's resonance.

Where each kind of coil sits

Slayer exciter, a few watts: nothing to detune. A point, and nothing else.
Burnett's HFSSTC, a 500 W demonstration unit at 4 MHz: the frequency is near enough stable. A breakout bolt, no toroid at all. He writes that he has "not found large toroids to have a beneficial effect on CW coil designs", and that a coil this small does not seem to detune by very much even at full power, which is the criterion rather than the toroid (richieburnett.co.uk).
QCW throwing a metre: catastrophic detuning. A large toroid and capacitors inside the secondary.

The topload is a trade in every case. You pay in volts, and volts are what make the arc long. What differs is how much frequency stability you are buying with them, which is what the first half of this page counts.

And that 4.14 pF per metre is one coil's calibration, not a measurement. Two arcs have been measured and both come out above it, but neither builder published a rate: what each measured is a total, 16 pF on Anders Mikkelsen's arc and about 7 pF on Weston's, and the per-metre figures of 5.03 and 7.66 are this corpus's own division of those totals by those arc lengths [derived, 16 pF over 3.18 m and about 7 pF over 3 ft]. Calling them published measurements hands our arithmetic to them, and it hides that the 3.18 m in that division is the one input in the row that has not been read back out of the thread. Both sit above the 4.14 used here, so every line on this page moves one way across that range. What the spread does not touch: the arc adds capacitance, that capacitance pulls the secondary down, and the pole you are switching follows it. Only the decimals are soft.

The job that decides how long the arc can get

The jobs above belong to every coil, and so does this one. The first half of this page counts it as the third, folding charge and frequency together as one job; this section splits those two because on a QCW they are separate decisions. The count differs, the job does not.

An arc is a resistance hung across the resonator, and a growing arc is a falling resistance. Past a point it damps the two poles together into one, and the driver loses the pole structure it was tracking. The toroid sets where that happens, because the toroid sets Z:

the poles hold while   R_arc  >  Z / k
here                   Z / k  =  26.6 kΩ / 0.421  =  63.2 kΩ

The 63.2 kΩ is derived, not measured: it is the secondary's characteristic impedance over the coupling, both of which are above.

A longer arc is a lower resistance, so the threshold reads as a length. The full table, and the widget for it, are above: 1.19 m as built, 1.58 m with twice the toroid. A bigger toroid lowers Z and pushes the length out.

It is not that only a QCW gets there

A DRSSTC reaches a metre and a half too. It gets there by stacking about fifteen 300 microsecond bangs on a channel that survives between them, each burst laying another ten centimetres on the end of the last, which is what a QCW arc is and an arc grows as fast as charge arrives. Every arc model in this corpus sets the arc's resistance by its length alone, so a metre and a half of hanging channel presents the same resistance to the resonator whichever machine grew it. The threshold is therefore not out of a DRSSTC's reach, and any argument that starts by saying it is has started wrong.

Where the threshold actually sits at that length is not settled either, and the corpus is open about the spread. The 1.19 m above comes from the harshest arc model in circulation, and the models spread to 2.81 m for this same coil. Put the five published arc resistances into one form and they all leave the poles alive at 1.5 m, with margins of 1.22 to 4.08 times and this coil's own at 1.85, which is there is no secondary frequency.

What differs is the price of crossing it

A DRSSTC's event is about 300 microseconds long into a flat bus, and its current feedback loop is started from nothing at the beginning of every one of them anyway, by a startup oscillator or by noise or by a comparator modified to free-run, which is every bang starts open loop. The next interrupter pulse is the next bang regardless. So losing the pole structure costs one flash, and the machine that comes back comes back to the conditions it left.

On a ramp there are no such conditions to come back to. The buck voltage has moved on, the arc has grown, and the pole structure is not the one the driver left. That is the same argument what a QCW arc is makes about pulse skip, and it holds here for the same reason: a QCW's event is 6 to 25 milliseconds of one continuous acquisition tracking a frequency that slides the whole way. Give way two thirds of the way along and you have lost the last third, which is the part with the highest bus on it and the part where the arc was going to get longest.

So the collapse margin is not a threshold only a QCW reaches. It is a design input only a QCW has to buy, and the reason is the length of the event rather than the length of the arc.

And past the threshold is a place you can work rather than a place you have failed. Anders Mikkelsen took 3.18 m operating beyond the collapse, and the diagnostic he gives is better than any threshold calculation because it needs no capacitance model: a frequency that stops responding to arc length has left the two-pole regime. That is on there is no secondary frequency as well.

And it pulls against the frequency

A QCW has to stay in the band where the arc grows as a straight sword instead of branching, which what a QCW arc is puts at 200 to 500 kHz and this page draws a practical lower edge on at 300 kHz, past about 28 pF of total secondary capacitance on this coil. A bigger toroid buys collapse margin and drops the frequency towards that edge, and the two walls are closer together than the cold numbers make them look.

Take the doubled toroid again. Ces of 24.9 pF, Z down from 26.6 kΩ to 20.0, collapse out to 1.58 m, and 320 kHz cold, all of which is on that page's table. Then put the arc on it: 6.2 pF more gives about 286 kHz by the end of the ramp [derived, 320 kHz times sqrt(24.9 / 31.1)], which is under the edge.

The department lives in the gap between the two: enough capacitance that the arc's damping does not merge the poles, not so much that the end-of-ramp frequency falls out of the sword band. That gap is the whole shape of it. The short, fat secondary is high coupling, which widens the pole split and stiffens each pole against the arc, carried on a big toroid held just off the floor. The poles themselves are there is no secondary frequency, and how far they walk down a ramp is the frequency slides down the ramp.

The move that is not a move

Doubling the toroid to halve the detuning does not simply work, and the reason is the paragraph above: the absolute frequency at the end of the ramp falls with it, and the absolute frequency is what keeps arcs straight.

Compensate with turns, then. The doubled toroid puts Ces up by a factor of 1.775 [derived, 24.9 / 14.031], so holding the cold 426.49 kHz wants the inductance down to 0.563 of what it was [derived, 1 / 1.775], which is about 25 per cent fewer turns [derived, turns go as sqrt(L) on a fixed former, and sqrt(0.563) is 0.750]. And that lands on the impedance:

Z = sqrt(L_sec / Ces)
  as built           26.6 kΩ
  doubled, retuned   15.0 kΩ   [derived, 26.6 · sqrt(0.563 / 1.775)]

A low secondary impedance means a higher peak current for the same voltage and a shorter ontime, which on a machine built around a long ramp is movement in exactly the wrong direction. Three things move and only one of them is the one you wanted.

And still not by winding

Whatever a QCW needs, it does not get it from the secondary's winding, and the reason is one sentence: capacitance won in the winding comes with impedance lost, and low impedance needs more current for the same voltage. Two layers wound to get capacitance end up with more interlayer capacitance than the topload has, and hundreds of volts per turn across thin varnish at the layer transitions. That failure belongs to the secondary is a resonator, which owns it.

Capacitors placed inside the secondary as discrete parts, which is what Ward did, are a different thing from winding for capacitance: they add capacitance without adding interlayer voltage stress. The general form of that lever, a string of ceramics doing the job a toroid would otherwise have to, is above.

The breakout has a harder life here

On a coil that bangs, the breakout point fires in bursts of a few hundred microseconds, hundreds of times a second. On a QCW it carries a full ramp of 6 to 25 milliseconds out to a metre or more, five to seven times a second, which is the interrupter is the note. Each event is twenty to eighty times longer than a bang [derived, 6 to 25 ms against 300 µs] and finishes at the top of the bus, so the tip is a consumable in the way what a DRSSTC breakout point is for means it: performance drifts across an evening with nothing changed, because a spike that has burned back changes where the arc starts and whether each ramp lights at all.

How far it should stand out is on its own page, and the answer is clear of the field that hugs this surface rather than tucked inside it.

The breakout point's own capacitance is the same trap here as anywhere, and it is stated with its measured case at the top of this page.

The numbers

The trade: Z = sqrt(L/C), volts as sqrt(P·Z).
Two equal touching spheres: 1.386 times one, not twice.
Breakout capacitance in the failed case: 12.6 to 20 pF, against a topload of the same order. It wants to be far larger.
Primary to toroid gap: about the secondary's radius.
Metal clearance: further than half the secondary's diameter.
The impedance lever: Z = sqrt(L/C) = 26.6 kΩ, and Q under arc ≈ R_arc / Z.
Pole collapse: the poles hold while R_arc > Z/k = 63.2 kΩ. As a length that is 1.19 m as built on Anders Mikkelsen's 75 kΩ·m, and 1.83 m on the harshest of the five published models once converted. The threshold in ohms is the settled half; the length is not.
The wall: about 28 pF total, where this coil drops below 300 kHz.

The measured figures here are other people's; the impedance and collapse-length numbers are ours, derived as noted above.

What goes wrong

Arcing inside the structure rather than off the point. The breakout has too little capacitance to the topload. Measure it.
The top turns of the secondary are burnt. It was run with no breakout point.
The measured resonance is well below the calculated one on a stack. The calculator added the capacitances. They do not add.
The toroid gets hot. Induction heating from the primary. Increase the gap and slot it.
Fitting a bigger toroid made the arc shorter. It bought stability you did not need. On a coil that was not detuning, the volts are the whole loss.
Branching and a drifting frequency after removing the toroid. Put it back; that coil needed it.

Where next

Winding and measuring a secondary, where the other half of sqrt(L/C) is set, and why capacitance won in the winding is worse than capacitance won here.
What a DRSSTC breakout point is for.
Is it CW or not, the criterion that decides whether a topload earns its keep at all.

The topload on the department diagrams answers for itself if you point at it.

toploadbreakouttuning