THE FOLD / CHEAT / THE EXPLOIT / THE ORDER-BLIND HASH
THE ORDER-BLIND HASH
a digest that forgot where it had been
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
There is a tempting shortcut for hashing a collection: hash each item, then XOR the results together. It is fast, it parallelises, and the order of the items does not matter. That last property is the shortcut and the hole. XOR is commutative and self-inverse, so a digest built this way cannot see a reordering, cannot see a swap, and cannot see any item added an even number of times.
LIT verified live on the same eight-file ledger: across 600 sampled orderings, the XOR digest takes exactly 1 value while an order-sensitive digest takes 600 — and offline, over every one of the 40,320 orderings, the same one-against-all-of-them holds. Swapping the first and last file leaves the XOR digest bit-for-bit unmoved. Adding one entry to the ledger twice also leaves it unmoved — a two-file forgery the digest is structurally unable to notice — while the chain changes immediately.
LIT verified live on the same eight-file ledger: across 600 sampled orderings, the XOR digest takes exactly 1 value while an order-sensitive digest takes 600 — and offline, over every one of the 40,320 orderings, the same one-against-all-of-them holds. Swapping the first and last file leaves the XOR digest bit-for-bit unmoved. Adding one entry to the ledger twice also leaves it unmoved — a two-file forgery the digest is structurally unable to notice — while the chain changes immediately.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) did not build this one; he built the thing it is a foil for. His
AVAN (AI) should state the counterpoint rather than leave the XOR digest looking merely broken, because it is not. Order-blindness is correct when order carries no meaning — a set of permissions, a bag of tags, a commutative merge — and the page checks that too: two orderings of the same unordered set agree, exactly as they should. The fault is never the tool. It is reaching for a commutative digest to seal something whose order is part of what it says.
seal.py carries the line “a hash per file, chained into a root… reordering the ledger changes the root”, and the accumulator in that sentence is the entire difference. This sphere exists to measure what he avoided.AVAN (AI) should state the counterpoint rather than leave the XOR digest looking merely broken, because it is not. Order-blindness is correct when order carries no meaning — a set of permissions, a bag of tags, a commutative merge — and the page checks that too: two orderings of the same unordered set agree, exactly as they should. The fault is never the tool. It is reaching for a commutative digest to seal something whose order is part of what it says.
3 ONE DIMENSION
The same eight files, shuffled again and again.
4 TWO DIMENSIONS · INTERACTIVE
Three forgeries. One digest sees them, the other does not.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: 40,320 arrangements collapsing to a single point.
AVAN’s addition (the inverse-companion): the forward reading is “a commutative digest loses information.” The inverse is that losing it is the whole service being bought. A hash that ignores order is a hash that says two things are equal as sets, and there is no other way to get that answer cheaply — you cannot both quotient by a symmetry and still detect it. Read backwards, this is not a weak hash but a correctly aimed one pointed at the wrong question, and the design error lives entirely in the sentence “seal these files,” which never said whether the order was part of the thing.
LIT across 600 sampled orderings of the same eight-file ledger the XOR digest takes exactly 1 value while an order-sensitive digest takes 600, and offline over all 40,320 orderings the same one-against-all-of-them holds; swapping the first and last file leaves the XOR digest bit-for-bit unmoved; adding one entry to the ledger TWICE also leaves it unmoved, a two-file forgery the digest cannot see, while the chain changes immediately; and two orderings of a genuinely unordered set agree, which is the same blindness being the correct answer
FIG David did not build this one - he built the thing it is a foil for. seal.py carries the line 'a hash per file, chained into a root ... reordering the ledger changes the root', and the accumulator in that sentence is the whole difference. AVAN states the counterpoint rather than leaving XOR looking merely broken, because it is not: order-blindness is CORRECT when order carries no meaning - a set of permissions, a bag of tags, a commutative merge - and the page checks that case too. The fault is never the tool. It is reaching for a commutative digest to seal something whose order is part of what it says. The page's order-sensitive comparison is a cheap mixer, not SHA-256; the chain itself is measured at full strength in [[the-chained-root]].
FIG David did not build this one - he built the thing it is a foil for. seal.py carries the line 'a hash per file, chained into a root ... reordering the ledger changes the root', and the accumulator in that sentence is the whole difference. AVAN states the counterpoint rather than leaving XOR looking merely broken, because it is not: order-blindness is CORRECT when order carries no meaning - a set of permissions, a bag of tags, a commutative merge - and the page checks that case too. The fault is never the tool. It is reaching for a commutative digest to seal something whose order is part of what it says. The page's order-sensitive comparison is a cheap mixer, not SHA-256; the chain itself is measured at full strength in [[the-chained-root]].
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE EXPLOIT · David Lee Wise (ROOT0), with AVAN