Nine tracked files in this corpus contain the word Langlands. Three of them are index pages, three are inventories recording that the corpus wants the material, and one is WP-82, the ruler that measured the absence. Two carry mathematics. Drinfeld, Beilinson, Harish-Chandra and fundamental lemma occur zero times each. The rung has been inventoried thoroughly enough that the inventory reads as coverage.
WP-82 held this corpus against a circulating graphic, 33 Levels of Mathematical Mastery,
and used it as a ruler rather than a curriculum. Rung 32 is Motivic / Langlands.
At commit d97154e its combined pattern returned 6 files. The table below re-runs
the same method at 6b7918b, one spelling per concept, tracked *.html
and *.md only, with this paper excluded from its own range. DATA
| Pattern | Files | Reading |
|---|---|---|
langlands | 9 | but see §2 — seven are not content |
motivic | 11 | |
automorphic | 5 | |
reciprocity | 4 | |
galois representation | 2 | |
shimura | 2 | |
gaitsgory | 1 | and it is a wants-list |
drinfeld | 0 | |
beilinson | 0 | |
harish-chandra | 0 | |
lafforgue | 0 | |
raskin | 0 | |
fundamental lemma | 0 | |
w-algebra | 7 | was 0 at WP-82's correction |
vertex algebra | 10 | |
frenkel | 9 | via Feigin–Frenkel, not via Langlands |
k-theory | 25 | was 8; rung 28 got built |
connes | 36 | rung 33, unchanged in character |
Two rows carry most of the argument, in opposite directions.
k-theory has gone 8 → 25 and w-algebra
0 → 7. WP-82 named rung 28 as the missing floor and chose the W-algebra route as
rung 32's reachable bridge; both have since been built. The instrument works, and the series
did the work it told itself to do. That is the control for everything below:
when this corpus decides to build a floor, the count moves.
WP-82's own block [3] exists because a file count counts index pages and the ruler itself,
so the composition is printed, not just the total.
Applied to rung 32 that lesson stops
being a caveat and becomes the finding. Here is every file behind the 9:
DATA
| Kind | File |
|---|---|
| INDEX | book32/index.html |
| INDEX | book7/index.html |
| INDEX | series-hub.html |
| INVENTORY | docs/audit-log.md |
| INVENTORY | docs/corpus-inventory.md |
| INVENTORY | docs/part-ii-source-wants.md |
| RULER | book6/wp82-the-missing-floor.html |
| CONTENT | book4/ch14.html |
| CONTENT | book7/ch-feigin.html |
Read as a total, rung 32 rose from 6 to 20 across two vocabularies and looks like progress.
Read by kind, it did not. Seven of nine files are the corpus talking about its own relationship
to the material: three pointing at it, three recording that it is wanted, one measuring that it
is absent. The single mention of gaitsgory is in
docs/part-ii-source-wants.md — a list of sources to obtain.
The shape of the defect. WP-82 recorded three failures with one shape:
a vacuity scan whose grep anchor could never match, so its green meant nothing; a Lean header
claiming five honest admits where the kernel reported six; a theorem whose hypothesis had become
sorry under Mathlib drift while #print axioms returned the same four
axioms before and after the statement stopped meaning anything. In each case the surface reads
correct and rests on something never checked to exist.
This is that shape one level up. The absence has been documented so thoroughly — inventoried, audit-logged, given a wants-list and a ruler — that a grep for the subject now returns the documentation of its absence. A corpus can bury a gap in its own accounting of the gap. The remedy is the one WP-82 used on itself: stratify the range, and print the composition rather than the count. DATA
The gallery slate this paper floors needs one place to state the mathematics, so that four chapters cite it rather than restating it four times. What follows is description, not result. OPEN
The organising image is older than the program and belongs to a man who already has a chapter here. In 1940, writing from a prison in Rouen to his sister Simone, André Weil described an analogy in three columns: number fields, function fields over a finite field, and Riemann surfaces — three settings in which the same statements keep appearing in different dialects. The Langlands program has three strands for the same reason Weil's letter has three columns.
Strand one — arithmetic. For a global field F and a reductive group G over it, a conjectural correspondence between automorphic representations of G(AF) and Galois-side parameters valued in the dual group LG — a construction Langlands introduced. Class field theory is the abelian case, G = GL(1). Modularity of elliptic curves over Q — Wiles, and Taylor–Wiles, 1995 — is a case of GL(2).
Strand two — function fields. Where F is the function field of a curve over a finite field. Drinfeld proved the GL(2) case in 1983; Drinfeld and Gérard Laumon formulated the general GL(n) geometric statement in 1987; Laurent Lafforgue proved GL(n), for which he took a Fields Medal in 2002. The trace-formula machinery the arithmetic strand needs rested on a statement Langlands and Shelstad had conjectured and could not prove — the fundamental lemma — which Ngô Bảo Châu proved, Fields Medal 2010.
Strand three — geometric. Over a curve in characteristic zero, a correspondence between sheaves on the moduli of G-bundles and the moduli of LG local systems; the formulation is Beilinson and Drinfeld's. It was proved in 2024 by a nine-person team led by Dennis Gaitsgory and Sam Raskin — with Dima Arinkin, Nick Rozenblyum, Joakim Færgeman, Dario Beraldo, Lin Chen, Justin Campbell and Kevin Lin — in five papers. Gaitsgory took the 2025 Breakthrough Prize in Mathematics for it and Raskin a New Horizons Prize. Sources disagree on the length: Quanta reports more than 800 pages, Wikipedia more than 1,000. This paper does not pick one. VALUE
One detail belongs in this corpus more than the rest of it does.
Beilinson and Drinfeld's conjecture could not be proved as stated. Quanta's account is that their
best hope
had been a bit too simplistic about how the relationships… should work
,
and that Gaitsgory and Arinkin made it into a precise conjecture in 2012 — twenty-five years
after the formulation and twelve before the proof.
A conjecture whose statement had to be repaired before it could be true is the same object this project has now caught four times in its own Lean: a name that promised more than its statement said. The difference is only that theirs was repaired in public, by the people who would go on to prove the repaired version. That is the standard, not an anecdote. OPEN
WP-82's correction of 2026-09-11 replaced rung 32's Langlands seed with the
vertex-algebra route for Volume XV — Feigin–Frenkel,
z(ĝ) ≅ W(Lg) — on the grounds that the W-algebra
path was the one the corpus could reach. docs/part-ii-source-wants.md still carries
Volume XV as The W-algebra route — BRIDGE
. That judgement stands and this paper does
not reverse it. The w-algebra row above, 0 → 7, is that judgement
being vindicated.
So a Langlands slate in Book 7 goes past what WP-82 judged reachable, and needs a reason that
is not the material is adjacent.
The reason is that Book 7 is a gallery:
its unit is attribution, and its established voice is naming what the corpus does not have.
ch-weil is in the gallery, in its own words, for the half of the sentence about why
his proof does not cross over. ch-grothendieck names K-theory as
a rung this corpus has been standing on without having built.
A chapter in that voice
attributes mathematics without claiming the series reaches it.
ASSUME The distinction between attributing a rung and occupying it is a choice about what a gallery chapter is for, not a finding. It is recorded here so that a later reader can reject it in one place rather than in four chapters. The failure mode it is meant to prevent is the one this series reversed in V5 of the CATGT paper, where seven domains were claimed to share one invariant on the strength of a shared functional form. The Langlands program is described as a unification; this corpus is also described as one; those two facts are not a bridge between them.
Four chapters, in the gallery's existing format, each with a verify.py checking
what is actually checkable. Named here so the slate is a commitment with a shape rather than an
open-ended expansion:
ch-feigin, which already carries the vertex-algebra
floor under it.verify.py has something
concrete to check.ch-feigin rather than opening a
thread. Vincent Lafforgue's description of the preparatory decades as a rising sea, in Grothendieck's sense, points at
ch-grothendieck; the cross-link is already earned.book6/wp130-verify.py recomputes every number in §1 and the composition table
in §2 at the commit they name, and fails if either drifts. It also asserts the zero rows,
which are the rows most likely to stop being true — and which this slate is intended to
falsify.