Why
A curriculum encodes an assumption that is false for anyone experienced: that ignorance arrives in order. Day 1 to day 100 is a sensible shape when the reader knows nothing, because the ordering is doing real work — each item is a prerequisite for the next. For someone eight years in, the gaps are not a prefix of the sequence. They are scattered, load-bearing, and revealed by whatever was attempted this week. The sequence stops being a scaffold and becomes a queue, and a queue you are behind on is a source of guilt rather than of learning.
Which changes what the material is for, not whether it is worth having. Read front-to-back, a 100-item track competes with the work; opened at the item that explains today's block, the same track is a reference with an unusually good table of contents. The difference is entirely in the access pattern, and it costs nothing to switch — no reorganisation, no deletion, just a different first move.
Today's example is exact. irreversible-switch-design rests on a fact about BLS12-381 — that signatures aggregate, that n of them still occupy 96 bytes, that verification finishes in a pairing — and that fact is what makes the post-quantum migration hard. Anyone who reads that card and does not already hold the aggregation property is blocked on one page, not on a curriculum. Reading the page is thirty minutes. Reaching it in sequence is weeks, and by then the card is stale.
And there is a measurement, which is what keeps this from being a preference. Log the blocks for two weeks and look at where they land in the sequence. Clustered means the ordering is earning its place; scattered means it is not. That is a real answer either way, and it is cheap.
How it works
Two access patterns for the same material
| Walk the sequence | Open at the block | |
|---|---|---|
| Assumes | Ignorance arrives in order | Ignorance is revealed by work |
| Competes with | The work | Nothing — it is the work |
| Failure mode | Falling behind, then abandoning | Gaps that never get surfaced because nothing touched them |
| Fits | Someone starting out | Someone eight years in |
The measurement, and it settles the argument
For two weeks, record two columns: what blocked me, and which curriculum item covers it. Then look at the distribution.
- Clustered in one region — the ordering is doing real work; reorder around that region and keep walking.
- Scattered across distant items — the sequence is not earning its place, and the material should be indexed rather than walked.
This repository has algorithms at 100 items and math at 52. That is 152 rows built on the walking assumption, and the log above is the cheapest way to find out whether the assumption holds for its one reader.
The honest counter-argument
Opening only at the block never surfaces what you do not know you are missing. Walking a sequence catches unknown unknowns; jumping to the block never will, because nothing has yet failed in a way that names them. So the two patterns are not rivals — the block-driven read is the default, and the sequence is what you walk deliberately in a quiet week, precisely because it finds what the work has not asked for yet.
Today's page
BLS12-381 and signature aggregation. Not because it is next, but because irreversible-switch-design assumes it and the assumption is doing all the work: aggregation is why the current scheme is hard to replace, and without it the post-quantum draft reads as a size problem rather than a structural one.