Make the rung drop's bounded population exact: 358, 111, 429, 207

src/v2/workflow/floor_expected_red.dag

@@ -31,10 +31,20 @@ import v2.std.text { String }
 // roster is admissible because it is temporary, visible at identity grain, and self-emptying —
 // remove any of those three and it is the skip list it is carefully not.
 //
-// POPULATION: 361 identities, after the declared rung drop below took 108 out by relocation
-// (469 -> 361). The count is NOT the whole population affected: 78 healthy passing witnesses
-// left the floor with them as collateral and are enrolled nowhere, so no number in this file
-// counts them. Read the rung drop, not this line, for what left.
+// POPULATION: 358 identities, after the declared rung drop below took 111 out by relocation
+// (469 -> 361 -> 358). The count is NOT the whole population affected, and the rest is stated
+// exactly here rather than left to the waves: 38 modules were relocated carrying 429 `test fn`
+// declarations. Of those 429 rows, 111 were enrolled here (the 469 -> 358 delta), 111 were
+// observed over budget or failing (102 budget_refused pre-quarantine, 8 tipped in 32189985063,
+// 1 straggler in 32193032348), and the remaining 207 are healthy witnesses enrolled nowhere,
+// which no number in this file counts. They return when their module does.
+//
+// The three figures above are checkable against the tree and the run counters, and they are
+// stated because the earlier revision of this line said 361 while the roster carried 358, said
+// 108 where the delta is 111, and described its collateral as "roughly 200". A bounded
+// population is the one thing a rung drop must not round: sharp-raven-273's monotone-shrink
+// wall reads this carrier as its baseline, so a stale count here is load-bearing for work
+// that is not mine. Found by tidy-lark-471's external review, not by me.
 //
 // HISTORY: cut at 820 on tree 380c242c79a; 25 were removed after the first complete
 // run on main (611fd027708) observed them PASSING and red the build naming themselves, which is the
@@ -104,8 +114,8 @@ import v2.std.text { String }
 // measured rather than predicted. Sixteen more modules leave here: the fifteen over 1000ms, and
 // test.claim.e0308_mechanical_trio_test at 774ms, which is below the line but FAILED on that
 // run, so it is quarantined on evidence rather than on threshold. Three more enrolled rows
-// retire with them (roster 361 -> 358), and roughly 200 further healthy siblings leave under the
-// same module-grain rule as the 78 above. The ceiling is 1552ms and it was calibrated as
+// retire with them (roster 361 -> 358), and further healthy siblings leave under the same
+// module-grain rule as the 78 above -- counted exactly in the POPULATION block, never rounded. The ceiling is 1552ms and it was calibrated as
 // measured-max-plus-one-quantum from a SINGLE 1527ms observation, so rows within a few hundred
 // ms of it are not stably green on a shared runner; that is the real reason this wave exists.
 //