T27.AI

Blog

Every t27 course is 9 modules of 3 lessons again

2026-10-07 · 2 min read

On 7 October 2026 the t27 courses were wrongly flattened to 27 modules of one lesson. They are 9 modules of 3 lessons again: courses 1 and 2 get their old modules back word for word, course 0 gets 9 new ones, and the catalog's test holds the shape.

Title card, no illustration yet, for: Every t27 course is 9 modules of 3 lessons again
View the title card at full size
#t27#Course

Every t27 course is 27 lessons in 9 modules of 3 lessons. On 7 October 2026 a change of ours flattened the courses to 27 modules of one lesson each. That was wrong: the owner's shape is 9 modules of 3, and it is back in all three courses, 0, 1 and 2.

What went wrong

The flattening commit set LESSONS_PER_MODULE to 1 in the course catalog, specs/course/courses.t27, and turned every lesson into a module with its own heading. The course map lost its 9 parts, and the blog posts of that day repeated the 27 by 1 shape as if it were the rule.

What changed back

ModuleCourse 0 lessons
A spec file1 to 3
Values and types4 to 6
Arrays, trits and expressions7 to 9
Tests and functions10 to 12
Inside a function13 to 15
Visibility and your own types16 to 18
Modules, rules and gen-ts19 to 21
The compiler and tri22 to 24
Errors, a program, what next25 to 27

How it is held

The course generator reads the catalog and refuses to build a course whose module count or lessons per module differ from it, or whose lesson does not sit in module floor(i / 3). Setting MODULES_PER_COURSE back to 27 turns the course check red; restoring the file turns it green again.

t27c on gft_xornet.t27 -- XOR needs a bend, native

The same move on a lesson spec: the recording of lesson 21 of course 2, xor-needs-a-bend, the last lesson of module 7. t27c test-report passes 4 of 4 on gft_xornet.t27, none vacuous; one sed removes relu from h1 and exactly one test, x00, fails; git checkout puts the line back and sha256sum prints the hash from before the edit. 27.5 s. Open the recording on its own page.

What this does not settle

Receipts

Work with me

Need an FPGA/RTL problem taken to measured hardware?

I work contract and part-time on hardware-AI, FPGA/RTL and ML systems — from specification and open toolchains to reproducible measurements.