Now that the PSU is selected, I'll address a few other aspects.

My initial idea was to reuse the transistor-resistor-capacitor structure of the #Yet Another Electronic Lampyridae but the idea has evolved. I'll reuse the system I developed earlier, with a 74HC245 as a quad-amplifier.

The '245 has 8 outputs but they are paired and paralleled (with one resistor per pin) to provide the required 20mA per SSR. This means 8/2=4 SSR (not 3 or 5).

I have decided to use only 1 SSR per channel. I have wondered if paralleling two SSR was such a great idea. It saved on assembly but electrically, I didn't really test the idea. A single SSR can sustain 2A@220V under "all-weather" operating conditions and a dissipator is required for more. I think that I'll use a dissipator this time, for an additional reason : spacing. The SSR use some room/space, to ease convection cooling.

Back to the 245:

    |
245 |     330
    |]---\/\/\---|     |
    |            |----[| SSR (1.6V, 20mA)
    |]---\/\/\---|     |
    |     330

The enable pin can be left active but can also be used for a global "disable". It's not the best way, though, as we'll see.

The 245 is a simple amplifier, but the original system used the 74HCT245 which has a lower trigger voltage. How can I make a delay line with it ?

Each output goes to the next input through a RC network. I want about 1 second so the RC ratio is about 1/1. For example 1µF and 1M Ohms. But there is something else to consider : the 245 doesn't have Schmitt trigger inputs. They are easy to implement with a 1M feedback resistor for example, so the RC's resistance must be lower. 10µF and 100K Ohms sound good.

The original system used the 74HCT245 which has a lower trigger voltage. I must explicitly choose a 74HC245 to prevent the system from triggering too fast, or to remain active too long after an event.

       |
   245 |     330
       |]---\/\/\---|     |
outputs|            |----[| SSR (1.6V, 20mA)
       |]-*-\/\/\---|     |
       |  |  330
          |              |
<--\/\/\--*--\/\/\--*---[| next 245 inputs
     1M       100K  |    |
                    *---[|
                    |    |
                    |
                   --- 10µF
                   ---
                    |
                    0V

So overall the system is mostly analog. All those resistors will create a mess though...

So the idea now is to use a 74HC595 to properly isolate each output and avoid the weird hysteresis circuit. But the '595 requires a (slow) clock. That's one more circuit.

I can make a clock with a pair of inverters (74HC04) but there might be a better approach: why not get the clock from the mains ?

@K.C. Lee suggested (in a comment) that I "spy" the mains with an optocoupler. This allows the detection of short failures of AC current. This signal should be bandpassed to avoid eventual HF parasites.

The optocoupler's output can then go to RC network, which will keep the /RESET signal low if no input is present, but also a 74HC4040 or 4060 binary ripple counter. In a european 50Hz network we get :

1. 25Hz
2. 12.5Hz
3. 6.25Hz
4. 3.125Hz
5. 1.5Hz
6. 0.7Hz
7. 0.39Hz
8. 0.19Hz

A simple selector (or DIP switches) will select the desired frequency.

The '595 Data input and Data output pins can be chained: an optocoupler will isolate the signal and each board can select between internal and external Din (another DIP switch).

Note : the '595 is rated for a total current of 70mA so the resistors would probably be a bit higher, 390 Ohms probably.

A red LED in series with the SSR's input will show the current flowing through it and give a direct reading of the sequencer's state.

Once again I favour a "no programming" system, where it's so easy to pop a µC. I don't want to lose time with IDEs anymore and it's cheaper...