Four copies of the same lie

The three boot sides are in no catalog at all: no index, no entries, nothing but firmware from sector zero to the end, which is why the first week of July went to counting instead of reading. The same header turns up four times on each side. The distance from one to the next is 498 sectors, or 63,744 bytes, and four of those almost exactly fill a side. Every disk this machine booted from handed it the interpreter four times over.

The four are not the same. I laid the slots against each other and they disagree in patches, which is what a disk looks like after a life in apartments and not in an archive. The headers match, and matching headers are what make four things copies. I counted them, saw four where I expected four, and went on to the bytes.

So I voted. Position by position across all four, take what three of them agree on, write the result out. One copy could have lied to me the whole way down and I would never have caught it. Four copies cannot keep the same lie straight, like witnesses put in separate rooms: wherever they come apart is wherever the disk has been eaten, and the majority puts the word back. The commit went in at half past eleven, majority vote across the four slots, and two masters came out of it that look the way interpreters are supposed to look.

I am pleased with myself maybe twice a year and this is one of them. The notebook says: redundancy is a gift, take it. The tea went cold two hours ago, the laptop is warm on my knees, and for the first time since April I have a file I trust.