hertz — WINDOW 4 predictions, filed 2026-09-11T10:22Z, BEFORE any build.
Tip 28d4eb68. ONE producer: webserve-e2e-run, 5 cases, 600 s cap, one attempt.
This is the third measurement of one case and the second of my own repin, so
the prediction is written to be refutable at a NAMED LINE, not case-wide.

20. It COMPILES. -> PREDICTED.
    28d4eb68 deletes one assert line and adds comment lines; it has not been
    near a compiler. A compile error is a result about MY EDIT, reported as
    that and not as a result about the case.

21. subnet_node_labels_shadow_only_the_docs_compat_alias_while_members -> GREEN.
    Rests on three things I have now measured rather than modelled:
      a. /llms.txt/ IS 302 with Location /llms.txt/docs/ — measured at
         71caedd4, both assertions passed before the panic.
      b. The follow-to-200 that failed is GONE; nothing now asserts a status
         on a directory root.
      CITATION CONVENTION, adopted after doyle's correction 10:21Z: every line
      number below is AT 28d4eb68 unless a sha says otherwise. A bare line
      number is my lane's numbering wearing the look of a shared address -
      webserve.rs:1028-1033 in my window-3 report is ~1081 on main, same file,
      same case. File AND sha, from here on.

      c. The next line, webserve_e2e.rs:276 @28d4eb68 (pre-existing, never yet executed because two
         successive panics masked it), fetches /llms.txt/docs/llms.txt and
         compares bytes to b"llms.txt". That file EXISTS: the loop at webserve_e2e.rs:245 @28d4eb68
         writes docs/<label> with the label as its bytes for each of the four
         root-file labels, and on this listener local_node IS "llms.txt", so
         the path resolves local branch -> facet docs -> remainder llms.txt
         -> docs_root/llms.txt. Predicted to PASS.

22. THE DISCRIMINATOR, stated before the run so it cannot be chosen after:
    If webserve_e2e.rs:276 @28d4eb68 PASSES, the redirect target resolves and the 404 I measured at
    71caedd4 was pure index-absence — my correction at 28d4eb68 was right and
    the case is closed.
    If webserve_e2e.rs:276 @28d4eb68 FAILS, the redirect target itself is wrong: /llms.txt/docs/ is not
    where that node's docs live, and deleting the status assertion hid a real
    defect rather than an irrelevant one. That would make this a PRODUCT
    question about the node-root redirect's target, not a test repin, and I
    would route it rather than repin again.
    A red anywhere EARLIER than webserve_e2e.rs:276 @28d4eb68 means the 302/Location pair stopped
    holding between 71caedd4 and 28d4eb68, which nothing in my diff touches.

23. The other four cases GREEN, unchanged. -> PREDICTED.
    They passed at both 61bfd85c and 71caedd4. 28d4eb68 touches one hunk in
    one case. If any of the four moves now, my edit had a blast radius twice
    unnoticed.

NOT PREDICTED, unchanged: webserve_cross_node_e2e (waits on todlando's relay
fix; the int tag lands after it, the cell is NOT repinned) and twohost_web.rs
(two-machine rig, compile-gated only here, closed by the #297 leg in W2).

STANDING COUNT OF MY OWN RECORD ON THIS ONE CASE, kept here so it stays
visible: miss 1 at 4823ae5d — classified the case by its other redirects and
never read the line. Miss 2 at 4ec34d1c — read the line, got the contract
right, asserted a property of the redirect TARGET without reading what the
target contains. Both were assertions from a model of the fixture rather than
from the fixture. This filing rests only on lines that have been executed or
on files whose bytes I have read in the test's own setup.