Apex publish broken since #1535: the catalog refresh wiped 40 course lesson specs #1540

What broke

Since merge commit 8e81487 (PR #1535, the t27 world scan catalog refresh, 2026-10-08 09:42Z) the apex publish fails: gHashTag/ghashtag.github.io run 37758579139 dies in npm run build -> prebuild -> node scripts/course-from-spec.mjs with 47 problems.

Why

Course PRs committed lesson specs straight into apps/website/public/t27/files/, the tree scripts/sync-t27-specs.mjs wipes and re-vendors from gHashTag/t27 master. The refresh did exactly that, so:

• 33 course-only specs vanished: specs/basics/01..27 (127ac93, course 0), specs/fpga/coverage.t27 (cb62daf), specs/fpga/{mmcm,mtbf,reset_sync}.t27 (6c745c4), specs/fpga/{eth_crc,rgmii}.t27 (6518d5c, course: buses-and-peripherals -- 27 lessons, 33 casts, 392 pages (Closes #1484) #1516);
• 7 corpus specs the courses had corrected reverted to t27 master, which the vendored wasm compiler reports as not clean: specs/numeric/phi_ratio.t27 (151 discarded), specs/math/phi_split_optimality.t27 (131), specs/math/radix_economy.t27 (59), specs/fpga/uart.t27 (12), specs/fpga/spi.t27 (73), specs/fpga/top_level.t27 (57), specs/fpga/bridge.t27 (6 typecheck errors).

The world scan's catalog gates never ran check:course, and a PR opened with the workflow token starts no website-checks, so nothing caught it before merge.

Where the truth belongs

In gHashTag/t27. Every one of the 40 files has an upstream PR in gHashTag/t27 with the same blobs: t27#7799 (the 27 specs/basics lessons) and t27#7496, t27#7659, t27#7579, t27#7586, t27#7626 (the other 13).

Done when

• node scripts/course-from-spec.mjs --check passes on main and the apex publish is green;
• the world scan refuses a refresh that drops or breaks a lesson spec.