Making the repair on the core stack - part 1

REALIZED EXISTING TWISTED WIRES WERE LONG ENOUGH

A very fortunate coincidence is the two failed connections are on the jumper block connectors closest to the twisted wires hooked to Sense/Inhibit lines 7 for low 4K and 9 for high 4K. That meant that I could simply snip the wires from the S-clip connector. That connector is where the signals should have been routed across the bottom board to the pins. 

Blocks with bad connections are nearest top

The wires can be brought to the base of the pins directly, since they are just long enough, bypassing the S-clips and bottom board entirely. This eliminates the need to use a substitute twisted triplet of wire as I preserve the characteristics such as impedance of the original IBM wiring. 

Bit 9 high 4K clips are leftmost

Bit 7 lower 4K S-clips are third set of wires from right

The S-clips on those connectors fit over the bottom PCB board and make contact with a pad, which in these pictures would be on the top of the board. They seemed secure but there was no connectivity from them to the jumper block pins. 

Bit 7 wires can be reached and detached

Bit 9 wires can be detached after plastic coating is removed

VERIFIED ALL WIRES AND DIODES FOR X AND Y ADDRESSING

Before I invested time in a repair of the two sense connections, I had to verify that all the other wiring in the core was in good shape. In particular, I wanted to test the continuity of each X and Y addressing wire through the core stack - 192 of them - as well as the array of diodes that are part of the path. 

X and Y wires are able to be accessed by a smaller number of driving circuit through the use of diodes. 

X address wiring scheme

There are eight Read Gate/Write Drive connections as seen on the lower right, plus eight Read Drive connections to the lower diodes at the upper left and another eight Write Gate connections to the upper diodes at the upper left. Thus 24 wires allow selection of current through 64 X wires and the current can flow in two different directions for reading and writing. 

To test all 64 X address wires I hook the diode tester to each of the eight connections on the lower right, while hooking the other lead to eight Read Drive and to eight Write Gate connections on the upper left. A good test shows continuity with a diode drop in the appropriate direction and no flow in the opposite direction. 

This meant I did sixteen checks for each of the eight lower right connections, confirming 64 wires conducting in one direction and the same 64 wires conducting in the other direction. The Y checks were similar except that the lower right equivalent has 16 connections rather than 8, since the Y axis supports 128 wires. A bit more work than confirming the X axis but I was satisfied that all 128 Y wires and all 64 X wires worked and that the steering diodes that supported dual direction currents were all working properly. 

NOTCHING OUT ROOM FOR THE WIRE

The plastic block that has the pins embedded fits directly against the rear of the backplane when the core memory is attached. The backplane is a glass fiber that is brittle enough that it might crack if subjected to bending or other stress. It is mounted in a plastic stiffener before it is installed into a computer frame. 

Closeup of the glass backplane in its stiffener, with the holes for core visible

If I had soldered (or done wire-wrap) on the pins as they were, the wire and solder would stick up above the surface of the plastic block. Those would be stresses applied at small points, therefore potentially cracking the board as I tightened down the core stack mounting screws. 

To protect against that, I notched away plastic material to let the wires and any solder sit down below the plane of the plastic block, removing the high points that could cause damage. I would then solder the wires to the side of the gold plated pins, down in the recess I notched out. 

SOLDERING AND TESTING FIRST REPAIRED SET - BIT 9, UPPER 4K

I cut the wires for bit 9 from the S-clip connector and tested their fit to the H4 jumper block pins. 

Wire ends near H4 block

I completed the notching with the help of my stereo microscope, continuing to remove material until it was deep enough for the wire to fit below the surface of the block.

Beginning the notching

I then stripped the ends of the wires and placed them against the side of the pin barrel in the dug out area. They were easier than I expected to solder. I used the ohmmeter to verify the connectivity to the sense/inhibit wires and the proper resistance readings at the jumper block pins after my repair. 

Wiring complete for bit 9

NOTCHED AND ROUTED WIRES FOR SECOND SET, TO BE FINISHED TOMORROW

The wires for bit 7 must be connected to pins B12, B13 and D13, the pins on the F4 block furthest from the edge. The bit 9 wires connected to the nearest pins, thus the wire length was less critical.

After disconnecting the wires from their S-clip connector, I tested whether they could reach the pins with enough margin to make a safe repair. As you can see from the pictures below, these are long enough. 

I did the notching, so all that remains to do tomorrow is to strip the wires and solder them on. Testing will be done afterwards to ensure that the sense/inhibit wires are now properly connected with low resistance. 

Switched gears after finding second continuity fault, removed core stack for repairs

SINCE I HAD PULLED TWO JUMPER BLOCKS, I TESTED THE SIX SETS OF WIRES

Each jumper block connects three sets of sense/inhibit wires, three wires each. I used the ohmmeter since the wiring was isolated due to the jumper removal. In addition to the continuity break  of one sense winding for bit 7 of the low 4K of memory, I found another break! Bit 9 for the high 4K of memory also had a high resistance in one wire. One was in F4 and another was in jumper slot H4. 

I put one side of the ohmmeter on the pin from the jumper block area (F4) and verified that the loss of connection was from that pin up to the S clips that anchored the three wires running to the sense/inhibit windings for the low 4K of bit 7. The sense/inhibit wires were intact as were the three wires running to the S-clips. 

The same was true for the broken connection on jumper block area H4 - no connection from jumper block pin to the S-clip but good connectivity over the three wires and inside the core plane for bit 9. The same failure mode. 

WHERE TWO ARE BAD, CHANCES ARE MORE WILL BE BAD

If I found a bad connection in each of the only two jumper blocks I had removed, it would be difficult to believe that none in the other eight jumper blocks were bad. I now was committed to pulling the jumper blocks off, rather than proceeding with the repair plan I discussed yesterday using the T-3 connector. 

I created a tool to help me pull off the jumper blocks that were wedged side by side leaving almost no room to slip anything between them. I got them all off so that I could test all the other sense/inhibit wire sets. At the same time, I pulled the jumper blocks that carried the read/write gate and driver wires up to the diode board and the X/Y wires. 

All jumper blocks removed

Bottom rows of each slot have core plane pins sticking out

REMOVING THE CORE STACK FROM THE COMPARTMENT AND BACKPLANE

Since I had the jumper blocks removed, the remaining work to remove the core stack was relatively minor. I decided to pull it off so that I could either repair the traces that were broken or make a more straightforward connection to replace it. 

The core stack is attached to the backplane with four screws, then it would be pulled away from the backplane which pulls all the pins from the core stack out of the holes in the backplane. The screws had been glued into place, so that I couldn't get most of them to turn. I decided that I could remove the entire backplane with the core stack still attached, then on the bench I could separate the two more readily. 

The backplane is held onto the frame by some bristol head socket screws and plastic holders. I removed these, disconnected the power connectors to the board, and lifted the entire assembly out of the frame. Putting it on my workbench, I could get better access to remove the screws holding the core stack onto the backplane. 

Corestack still mounted to backplane

Wire wrap/pin side of backplane

Card slot side of backplane

With the core stack removed, I saw that I wouldn't have access to the traces that ran from the S-clips to the pin blocks unless I unbolted the entire stack and pulled the layers apart. The risk of damaging connections was far, far too high so I didn't do that. 

Bottom board of core stack with jumper pins

Top side of core stack with S-clips at top left and top right

REPAIR PLAN IS SIMPLE

The pin blocks for the jumper pins were right there and it was obvious that I could run a wire up to the S-clip area and solder it on. That is the repair I will do, running new wound triple wires from the pin blocks to the S-clip area instead of relying on the traces on the core stack bottom board which were defective for two of the lines. 

Leaning to use of spare top board connector to bring sense bit 7 signals from core stack to the SLT backplane side

CAREFUL STUDY OF THE UNUSED CABLE CONNECTOR TO FIND FREE PINS

I need three pins for the twisted wire cable from the sense/inhibit winding for bit 7, first 4K words, which is where we have the continuity break between the core stack and the SLT backplane. One of the three is a ground, the other two are sense/inhibit wires for the first 2K and the second 2K of cores. 

The connector I want to use, T-3, runs across the top of the backplane in row 1. It starts partway in column G and ends partway in column K. It is oriented at 90 degrees to the normal SLT card pins. In a card, we have two vertical columns of 12 pins. The top connector has two horizontal rows of 12 pins. 

Just to be confusing, the naming of locations on an SLT backplane has numbers and letters for the major locations, such as where a card is placed. There are eight horizontal rows numbered 1 to 8 from top to bottom. There are eight vertical columns labeled A through N (skipping the letter I, an IBM practice). Rows 2 through 7 are full height and can have an SLT card plugged in to the spot or a connector which uses the same pin layout as a card. 

The top and bottom rows are not full height. They are used exclusively for connectors, never cards. The same pin layout is used for these as for connectors (or cards) in rows 2 to 7, but these are rotated ninety degrees. There are four connector spots at row 1 and another four spots at row 8; these are labeled as T-x and B-x connectors. 

Within each intersection of a row and a column, e.g. E5, IBM also uses letters and numbers for a finer pitch grid. Thus, each intersection has five columns labeled A to E and 14 rows numbered 1 to 14. The pins for an SLT card or connector are placed in columns B and D on this fine grid and are installed only in rows 2 thru 13. Thus, the pins for an SLT card in an intersection are generally labeled D02 to D13 and B02 to B13 using the fine grid names. 

One can therefore have a card at B2 slot, which has a pin B02. The terminology can therefore be confusing unless you mentally switch between the fine grid inside a slot and the gross grid that labels the slots. The reason I had to share this is because of the labeling of the T-1 to T4 slots in row 1 of the gross grid. 

Row 1 and Row 8 are not full height, they don't have fine grid rows 1 through 14 as do the gross grid rows 2 through 7. They do, however, have the very same fine grid columns A through E. I will use the actual pin locations for T-3, the connector location I want to use for my repair, to make this clear.

T-3 is a connector with two horizontal rows of 12 pins. The top row runs across fine grid row 9 and the bottom row runs across fine grid row 11. Thus we need at least 14 fine grid spaces horizontally to fit the pins plus a space, but a gross grid column has only five spaces A through E in it. Therefore T-3 is going to need two full gross grid columns plus a bit of the column before and after it.

For T-3, this means it uses all of H and J gross grid columns, fine grid columns A to E of the first and then A to E of the second. T-3 also uses the last fine grid column, E, of gross grid column G as well as the first fine grid column, A, of gross grid column K. I will use capital letters for gross grid columns and lower case letters for fine grid columns. We know this connector sits in gross grid row 1 so I don't need to repeat that in the list; all numbers therefore are fine grid rows.

The pins of T-3 are, for the top row of 12 pins:

• G e09
• H a09
• H b09
• H c09
• H d09
• H e09
• J a09
• J b09
• J c09
• J d09
• J e09
• K a09

The bottom row of 12 pins are :

• G e11
• H a11
• H b11
• H c11
• H d11
• H e11
• J a11
• J b11
• J c11
• J d11
• J e11
• K a11

Thus when you look at the diagram of the pins used on a backplane, the listings for pins in row 1 and row 8 always look weird. Here is a small sample from the core memory backplane:

Gross grid slot G1 lists four pins, but we now know that the A09 and A11 pins belong to the other connector T-2 and only pins E09 and E11 are part of T3.  In the same way the right end of T-3 sits in gross grid slot K1 but only pins A09 and A11 belong to T-3. The other two are part of connector T-4.

I looked at the backplane listing and saw that three of the pins are not listed, therefore presumably not in use. These are H1 e09, J1 a09 and J1 b09, again using lower case letters for fine grid. Every SLT backplane grounds certain pins. Fine grid d08 pin in rows 2 through 7, all columns. It also grounds pin J1 fine grid a09, something not shown on the backplane listing but key to understand. 

I therefore have just enough open pins to hook up my sense wires for bit 7, 0 to 4095 addresses. The black common wire gets hooked to J1 a09 since it is connected to ground on the backplane when it enters through the jumper block F5. The white and blue wires will be hooked to H1 e09 and J1 b09. 

I will verify this with the ohmmeter and other tests before I begin the installation of a replacement tightly wound three wire cable. I will use a spare SLT card connector from my parts salvage to make the connections to the two pins, so that my cable plugs into T-3 the same way that a cable would have.

ERRONEOUS NUMBERING OF CABLES - VERY DANGEROUS

IBM has a page in the ALDs which shows the cabling required between the SLT compartments. They use the names of the gross grid card slots for connectors that install in row 2-7 columns A, N and sometimes B or M. They use T-1 to T-4 for the top connectors, which should correspond to the official SLT designation of those names. 

This excerpt from the cabling diagram shows the memory compartment on the upper right B-C1 which is used for 4K or 8K sized memory machines using the original 3.6 microsecond core storage. The machine I am restoring is an 8K 3.6us machine so memory is installed there in B-C1.

IBM uses the term Expanded Storage (ES) for machines with more than 8K of memory. An add-on compartment is bolted to the left of the 1130 (called the 'blister' by IBM) and one or two gates are inside with compartments holding core memory. Each compartment holds an 8K core memory, with two compartments per gate and therefore with two gates up to 32K can be configured. 

The red cables in the diagram above area installed only when the machine is bigger than 8K (or with the faster 2.2 us memory but lets ignore that for now). Compartment B-C1 does NOT have core memory inside when the blister is used, but instead has a few SLT cards supporting the extra address bits needed to select above 8K locations. Therefore the signals that come into the B-C1 when it is configured with core memory have to be routed out through A2, A3 and A4 cables to run over to the gates in the blister. 

The additional address bits and control signals for addresses over 8K are connected to B-C1 through connector T-3. If ES is not configured in the machine, then that cable is not installed and T-3 is open. 

The problem is that IBM has labeled this as T2 on the diagram. The signals going into this position over that cable are clearly connected to the pins called T-3 on the backplane listing and the correct name for that position is T-3 according to SLT rules. If they are not using the SLT names for the locations, then they should NOT have used naming like T2 that could be confused, instead using a different naming scheme. 

The pins from gross grid G1 to K1 are part of T-3 as can be seen above. I had to triple check when I found this mismatch. I have claimed in the past that IBM documentation was always accurate, even if it was often unusable or obscure, but this situation disproves that adage. 

Bit 7 connectivity issue for 1130 memory appears to be on base board of core stack

PHYSICAL STRUCTURE OF CORE STACK

The core stack consists of a sandwich, having a bottom printed circuit board, eighteen core planes in the middle and a diode board on the top. As the core stack sits in the IBM 1130 memory compartment, the diode board is facing you when you look inside the compartment and is at right angles to the floor and parallel to the backplane of the compartment. Side A is facing the ceiling and side C is facing the floor. Therefore D is to the left and B is to the right when you are looking in the compartment at the core stack. 

The core planes are parallel to the backplane of the compartment. Their long edge (B and D) is vertical to the floor and has 128 wire connections for addressing the with bits 9 to 15 of the memory address. The shorter edge (A and C) has 64 wires for addressing with bits 3-8 and is parallel to the floor. In addition to the addressing connections on the sides of the core plane, there are connections for the sense/inhibit wires on the A and C sides of the plane. 

Wiring from the addressing and sense/inhibit lines are attached to the bottom board with a clip, contacting the copper trace of the printed circuit on that bottom board. This core memory uses the S type clip. 

Pins stick out of the bottom board. These fit through holes drilled in the compartment backplane and the connection of those pins to the backplane traces is via the jumper blocks I mentioned previously. 

The path from sense/inhibit wires for a particular bit (core plane) begins on the A or C side of the core stack. Each core plane has four sense/inhibit wires, each handling 2K of the 8K bits on that plane. The core plane is divided vertically, with the B side of the plane consisting of addresses 0 to 4095 and the side near D consisting of addresses 4096 to 8191. 

Two sense wires further divide each half of the core plane, so we can think of it has having four quarters divided vertically. Each pair of sense wires are tied together at one end, called the common end and the other side of each wire is brought out separately. 

Each 4K half of the core plane has three terminals, the two sense wires plus the common wire between them, which are soldered to a trio of wires tightly wound as they are routed away from the core plane. The wires are blue, white and black with black used for the common connection. 

 

In the closeup above from side A, most of the contacts you see are the ends of the X addressing wires which are daisy chained from one core plane to the next, so that each X wire runs through all 18 core planes. The blue and white wires are soldered to the two ends of the sense/inhibit lines for the B side of the core plane and the black wire going to the common terminal of those lines which is beyond the right edge of this picture. The red oval shows the connections for the core plane of bit 7. 

The diagram above corresponds to the picture of the wires to bit 7 but the diagram is rotated 180 degrees from the perspective we see looking at the core plane from the front. The bottom of this diagram is the bottom board of the stack, which is furthest away from our view in the pictures. The ovals show where the white and blue wires are soldered in the picture above this. 

The other end of the white, blue and black trio of wires runs to a connector on the left side (from our perspective). The other red oval indicates where the three wires are soldered. These numbers positions are on the bottom board and connect via the S clip to a printed trace. 

This diagram shows where those connections fit onto the bottom board. The blue oval shows the connectors where the wires are soldered, and the picture below shows our wires on the connector as viewed from my perspective working on the core stack. 

That connector has the S clips behind it, as is a bit more visible in the picture below:

We don't know the exact route of the trace from the wire but it ends up on the pins of the bottom board where they will fit through at card slot F4 of the backplane. This is where the jumper connects those bottom board pins to the backplane pins at F4. From there, traces on the backplane route the signals to card slot B3 where the sense/inhibit circuit for bit 7 is implemented. 

We found that the pin D13 at F4 card slot, from the bottom board of the core stack, is not connected to the sense/inhibit wire on the A side as shown in the first photo. The question was where the break occurred and then how best to rectify it. 

LIKELY CONTINUITY BREAK LOCATION IS ON THE BOTTOM BOARD OF STACK

I used the ohmmeter to check continuity between the blue and white wires where the connect to core plane 7 and the other ends where they are soldered to the bottom board connector. I had a solid low resistance connection, and further more when I touched the S clip I had good continuity to that. 

Therefore, I presume that the break is somewhere from the S-clip to the pin D13 for F4 on the bottom board of the stack. It might be the trace on the bottom board, it might be a cracked connection at the pin, or it might be a failed connection with the S-clip to the pad of the bottom board. 

WAYS TO CORRECT THIS BREAK IN CONTINUITY

The core stack is designed with a nice impedance matched connection from the sense/inhibit wire ends all the way to the SLT card at B3. I don't want to throw that off by routing a single wire to bridge around the defect, unless it is done with full regard for the impedance of the circuit. 

The straightforward way to repair this is to remove the core memory stack from the memory compartment. This is done by pulling all the jumper blocks off the backplane, then removing four screws that hold the core stack to the backplane. I would then lift it out extremely carefully and move it to the workbench. There I could find the connectivity fault and make the most organic repair. 

However, this poses some practical problems. There is quite a bit of force on the jumper connections. I was able to remove two of the jumper blocks using a chip puller but the side by side placement of the remaining jumper blocks won't allow the use of that tool. 

IBM mentions a special tool that the engineer should use to pull these. Of course I don't have such a tool, but might have to manufacture one. It would need to fit over the top and bottom of the blocks, not need too much of a gap since they are still spaced closely in the vertical direction, but impossibly close horizontally. 

The other possibility is to route a similar tightly wound triple wire cable from the top (side A) of the core stack and somehow get the signal around to the backplane. There is an unused connector slot T2 at the top of the backplane, thus I have a choice of 24 pins to use. I would have to choose pins that have NO connection to any other part of the backplane, then use wirewrap to bring the signals to card slot B3 which is not far away. 

The new wire from the side A connections would have to run up between SLT cards probably in column E to reach the back of the connector 2 where I would terminate them. 

I have to mull over the relative strengths and weaknesses of the two repair methods, then execute. 

Confirmation of the broken mainspring in the 1130 console printer

REMOVED THE EXISTING MAINSPRING AND SEPARATED BROKEN PART FROM REST

You can see the part of the metal tape that broke inside the mainspring container in the picture below:

The inside of the mainspring attaches to this hub visible with the mainspring cage removed:

Bit 7 has failed again, as I feared

CAME INTO THE SHOP TO DEBUG THE CPU, NOTICED BIT 7 WAS FAILING AGAIN

I put the memory compartment covers back onto the system, now that memory was working (I thought). Previously I had the cover and side rails removed to give easy access to all cards and the core stack.

I intended to check a number of issues I was seeing with various instructions executing. I set up the scope for the first situation and toggled in instructions to execute. One I entered was going to use index register 1, thus it would have bit 7 set. I stored the instruction but when I executed it, there was no use of the register. I displayed the memory location and noticed bit 7 was zero. I tested by trying to store all ones and got the expected 0 in bit 7 and a parity error.

WILL HAVE TO FIND AND REPAIR THE CONNECTION PROPERLY

Touching the connection with my ohmmeter probe had temporarily restored connectivity, but that went away. I need to zoom in much closer and figure out the exact failure so that I can permanently repair it. 

Typewriter work - adjusted front pushbuttons, freed stuck parts, diagnosed broken mainspring

CONTINUING RESTORATION OF SELECTRIC BASED CONSOLE PRINTER FOR 1130

The IBM 1130 comes with a console with a keyboard and a typewriter, along with some switches and controls. The keyboard is not connected to the typewriter at all, it is the keyboard from an IBM 029 keypunch mounted into the 1130. The typewriter is an "I/O Selectric" printer only device. This is based on the original IBM Selectric typeball typewriters, but with solenoids, microswitches, pushbuttons and mechanisms added to allow this to be driven by a computer. 

Because of the additions to the basic Selectric design, many components in the machine are blocked from adjustment or direct access by the layers of additional hardware that were added to make it an Input-Output (I/O) version, in this case a print only type. That often requires substantial disassembly of parts to gain access which on a regular consumer typewriter would be easily reached immediately. 

The oil and grease used in the 1960s will slowly absorb dust from the air and become solidified. The foam sound insulation inside the case looses its plasticizer and reaches one of two states - either a crumbly powder or a tarry goop. In many cases, it does both in different parts of a machine. The result is a dirty device which has most moving parts frozen in place. They rarely work if they have been sitting for years, requiring substantial labor intensive cleaning to restore operation. 

FRONT PUSHBUTTONS WERE MISADJUSTED

The operator of an IBM 1130 has to be able to space up the printout in order to tear it off. They also need to be able to set and clear tab positions. Finally, a means is needed to move the carrier over to a chosen line or to activate the tab function, when the operator is setting up the typewriter to print a form or designed output. 

IBM added three buttons on the front panel - Tab, Space and Return - to trip the operational section functions that accomplish those tasks. The buttons pull on a wire inside a metal jacket since the selectric operational unit triggers are at the right front of the machine and too close together to have buttons in between them. The wire cable routes the pull from the centered pushbuttons on the panel over to a plate in front of the operational unit triggers, ultimately rotating and moving a pad down to push on a trigger and trip it. 

The current adjustment had those pads moving down in front of the triggers, failing to activate them. The parts that must be adjusted to correct this are behind a plate and behind all the parts anchoring the cables, holding the arms with the pads and attaching springs to control their motion. I had to disassemble the plate, detach the wire cables, and disassemble the pushbuttons in order to get inside and adjust the moving arms with the pads. 

Once they were adjusted, I reassembled everything and verified that the pushbutton will now reliably trigger its operational function. Of course, I had to free up the gummed oil to allow all the parts to move freely weeks ago, before I began to address the failure to trip the functions when pushed. 

ITERATIVE PROCESS OF FINDING PARTS THAT SHOULD MOVE AND FREEING THEM

The machine has several thousand parts, many of which have to move. It takes patience to examine everything and find each part that must rotate or slide. I use an extremely fine clock oil to soak in under the frozen lubricants so that I can move the parts. Exercising the parts after application of the clock oil is necessary until the normal range of motion and ease of movement is restored. 

The margins on a Selectric are set by sliding two parts left or right on a bar. These parts have a long blue plastic indicator on the front which shows through a plastic window on the front panel, to indicate the current column for the left and right margins. They should slide easily when the blue part is pushed back to disengage a tooth from the rack they slide upon. It took quite a bit of work to unstick these but now they move freely.

MAINSPRING BROKEN

The Selectric typewriter receives power from an electric motor which turns a shaft in the machine. Various clutches will engage with the shaft to move parts when a function such as tab is required. Another clutch will engage with the left side of the shaft to produce one print cycle - selecting a character on the typeball, striking the ribbon with the ball, advancing the ribbon and tripping the space to move right one column. 

However, the rightward movement of the carrier is not driven by the motor or shafts. A mainspring is wound up as the motor pulls the carrier all the way to the left in a carrier return operation. The tension in the spring pulls on cords running to the right side of the carrier, trying to move it rightward.

The carrier is held in place by a tooth that engages a rack running across the typewriter behind the carrier. The tooth is temporarily pulled out of the rack when a space operation must take place, allowing the mainspring to begin moving the carrier rightward. The tooth drops into the next spot in the rack so that the carrier has advanced just one column.

The tab operation pushes the tooth away from the rack and latches it way. This allows the mainspring pressure to slide the carrier rightward rapidly. A projection on the tab mechanism on the carrier will bang into a metal point if the tab is set for that column, releasing the latch and stopping the carrier. A cylinder runs along the rear of the machine with a metal point at each column position. The point is moved by a tab set or tab clear. 

The clear symptom here is the lack of tension and no movement of the carrier to the right when the tooth is released for a space or tab request. I had rotated the spring housing when I installed it but with a break in the spring metal inside the housing, it provides no force. Carrier return does not wind it either because of the break. The spring itself is a long band of metal which should be fastened to the outside of the housing on one end and bound by a notch in the axle on the inside. 

I have a new mainspring on order. When it arrives, I will install it and pretension it.