The institutional loop

Academy

Where the stack teaches, and proves the teaching took. Three layers that compose into a loop: grounded material that can show its sources, practice against real repositories, and proof that accumulates into new claims worth publishing.

The loop

No single layer is Academy. The output of each becomes the input of the next, and the last one feeds the first.

Read BENDSCRIPT · PUBLIC · CRAWLED MOCK Practice RUNEFORT · LOGGED-IN BLOCKED — NO DIST/ Prove WORKBENCH + PRISM EXISTS — NOT WIRED teaches emits new quantity + credential claims — transmuted by alkeyword

Bundle counts and gate pass-rates are exactly the quantity and credential claim types alkeyword identifies as the strongest citation levers. The proof layer feeds the reading layer — that is why this is a loop and not a content calendar.

Three layers

Distinct jobs, distinct technologies, distinct audiences.

What is actually true today

Measured 2026-08-10 by pointing alkeyword at the portfolio's own sites. Zero engine calls.

Atlas · extractable words
786,388 was 6
Atlas · indexable URLs
432 was 1
Docs published
314 / 36 unlisted / 23 private
FAQ pages in portfolio
0 / 20 surfaces
How-it-works pages
0 / 20 surfaces
Engine probes run
0 unmeasured

The last tile is the important one. 786k crawlable words is an infrastructure result, not a citation result. Nothing here yet says "cited 2/3 runs where it used to be 0/3" — and until it does, Academy's central thesis is unproven.

Build order — two lanes

The original order was one chain, which put the step capable of invalidating the Practice layer seventh — after all the work needed to build it. These two lanes do not depend on each other and run in parallel.

Lane A — infrastructure

makes the material real
A0Prerender the atlas350 pages · 786,388 words · deterministic
A0bPublication policyceiling > declared > pattern > global · 41 tests
A2Re-run the self-crawlgate: G5 matrix recomputed against the atlas
A3Mint the first FAQ + How-it-worksgate: the 0/20 forms exist — or are refused in public
A4Semantic extractor beside the deterministic oneharness built · 18 self-tests · semantic arm unrun
A5bendscript.provenance.v1 + renderergate: hover any sentence, see its source span

Lane B — falsification

finds out whether anyone wants it
B2Ask what was missing when they got stuckgate: a named deficit, not a satisfaction score
B3Decide what Practice must therefore begate: the tiled floor is confirmed, narrowed, or replaced
B4Practice surface over one real repositorygate: someone who is not the author finishes a task in it
B5Wire practice → SkillBundle → PRISMgate: a scored bundle without hand-holding
C1Academy nav entrygate: A3 and B4 both live

B1 is the highest-information step here and the cheapest. One person, one task, no new code. If they finish using the atlas and an assistant they already have, that does not kill Academy — it means the scarce thing was never the workspace, and Practice becomes a governed execution and evidence surface rather than a place you work. The tiled floor on the Practice page is drawn because it can be drawn today, not because it has been chosen.