1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
//! Slice 7b's core-level coverage for the **keepalive livelock** (F1).
//!
//! Written from `CONTRACT-7b.md` §4 and `ADVERSARIAL-liveness.md` F1 by an
//! author who never read the fix (CLAUDE.md working rule 6), in a worktree
//! cut at `c131904`.
//!
//! # The defect
//!
//! `transmit_keepalive` refuses to seal when
//! `contested.is_pending() || !amplification.admits(30)`, and returns
//! **without moving `last_send`**. `transmit_keepalive_if_owed` then calls
//! `sync_liveness_timer()` unconditionally, which re-arms `Keepalive` at
//! `last_send + KEEPALIVE_TIMEOUT` and `PersistentKeepalive` at
//! `last_send + interval`. Neither expression consults `contested` or
//! `amplification`, so both land at an instant already passed, the shell
//! `sleep_until`s a past instant, and the driver spins.
//!
//! # Why "the deadline is correct" is not a test
//!
//! **A livelock is not a wrong value — it is the same correct value
//! forever.** `assert_eq!(timer(PersistentKeepalive), last_send + interval)`
//! passes the spinning build; it *is* the spinning build's behaviour. Every
//! test here asserts from the side that separates them:
//!
//! | shape | the broken build | the fixed build |
//! |---|---|---|
//! | **non-retrospection** — the announced deadline is `None` or `> now`, for the `now` just handled | announces `now` | announces the next real deadline, or `None` |
//! | **termination** — a driver loop that sleeps to each announced deadline reaches a fixed point in a bounded number of steps | never advances past the refusal instant | dies at `DEAD_TIMEOUT` in a handful of steps |
//! | **monotone advance** — the deadline after `handle_timeout(d)` is strictly greater than `d` | equal to `d` | greater |
//!
//! `CONTRACT-7b.md` §4.3 makes the first of these binding and adds the
//! second half of the bound: *"against a build that suppresses **the death
//! clock too** it also passes — so a second assertion is required: the
//! `Liveness` deadline is still announced when it was armed."* Every test
//! below carries that companion, because a core that answers `Timeout(None)`
//! to everything satisfies non-retrospection trivially and is an
//! **immortal** connection, which ruling 182's beacon proof calls the worse
//! collapse.
//!
//! # The two guards are pinned separately — this is the working-rule-9 part
//!
//! The refusal has **two** disjuncts and a build may fix one. So:
//!
//! * [`beacon_refused_by_the_budget_does_not_re_arm_in_the_past`] sits at
//! `room == 0` with **no** contested mark. A build whose new arming
//! condition consults `contested` but not `amplification` still spins here.
//! * [`beacon_blocked_by_a_pending_mark_does_not_re_arm_in_the_past`] sits at
//! `room == 30` **exactly** — the budget admits the 30-byte keepalive and
//! refuses the 31-byte probe, so the mark is pending and the *budget is
//! not* the reason the keepalive stays home. A build whose arming
//! condition consults `amplification` but not `contested` still spins here.
//!
//! Neither test alone separates the four builds; together they do. That is
//! the whole reason the contract asks for **one** predicate
//! (`keepalive_can_leave`) rather than two copies.
//!
//! # No clock, so no runtime
//!
//! Sans-io core tests: `now: Instant` is an argument and nothing here reads
//! a clock. Plain `#[test]`, no `sleep`. Every mutating call is followed by
//! draining `poll_output()` to the terminal `Timeout` (§16.4), which is what
//! [`testfix::drain`] does.
//!
//! # What is **not** here, and why — a reported reachability limit
//!
//! See [`the_passive_form`] at the end of this file. The `armed == false`
//! variant `ADVERSARIAL-liveness.md` calls *unbounded* is, on this author's
//! arithmetic, **not constructible**, and the reason is a one-line
//! inequality. It is reported rather than half-tested.
//!
//! **[CORRECTED 2026/08/18 — ruling 265.] It is constructible, it was
//! built, and the state it reaches is worse than the spin this file pins:
//! a connection that announces `Timeout(None)` with no timer armed at all
//! and neither dies nor sends.** The inequality's second step is the one
//! that fails — a pure ACK is *sized to the room*, not ~35 bytes — and
//! [`the_passive_form`] carries the corrected arithmetic. The tests are
//! `tests_park.rs` (core) and `tests/story_park.rs` (shell); reporting
//! rather than half-testing was still the right call, and the report is
//! what got measured.
use ;
use *;
use TimerKind;
use crate;
use crateConnectionLost;
// ═══════════════════════════════════════════════════════════════════════
// 1. Fixture-level scaffolding
// ═══════════════════════════════════════════════════════════════════════
/// §3.4's cleartext header plus the AEAD tag — what every Data packet costs
/// before a single frame byte, and therefore **exactly** the size
/// `transmit_keepalive` asks the budget to admit.
const KEEPALIVE_LEN: u64 = as u64;
/// §7.5's contested probe on the wire: one `FRAME_PING` byte in a Data
/// packet. One byte larger than the keepalive, which is the entire reason
/// [`beacon_blocked_by_a_pending_mark_does_not_re_arm_in_the_past`] can
/// separate the two guards.
const PROBE_LEN: u64 = KEEPALIVE_LEN + 1;
/// §7.5's beacon, at its floor. `PERSISTENT_KEEPALIVE_MIN` is 1 s, so this
/// is the shortest interval the core will accept — and it must be shorter
/// than the first PTO (`K_INITIAL_RTT`-derived, ~1.02 s) or the beacon is
/// not the timer under test. Every test that relies on that asserts it
/// rather than assuming it.
const BEACON: Duration = from_secs;
/// A core installed as the **responder**, i.e. anchored from a msg1 source:
/// §3.2's second arming event, `budget = (107, 196)`, cap 588, room 481.
///
/// **[Integrator, ruling 218 — premise migrated, not weakened.]** This
/// author wrote `(0, 196)` against a tree where the arming charged nothing
/// for msg2. The adversarial review's A2 found that msg2's 107 bytes are
/// emitted endpoint-side and were charged to **nothing**, making the real
/// responder ratio 3.55× against a normative MUST of 3; the remediation
/// implementer fixed it in the same slice, blind to this file.
///
/// So this red was the fixture premise doing precisely its job — the author
/// wrote it to *"make the premise visible rather than silently weakening
/// every bound below it"*, and it caught a deliberate change to the very
/// quantity it pins. Every test here derives its room through
/// [`room`]/[`spend_to`] rather than hardcoding, so nothing below this line
/// needed touching.
/// The budget's two counters, which must be armed.
/// How many more datagram bytes §7.3 will admit right now.
/// Spend the budget down until **exactly** `target` bytes of room remain,
/// with a single unreliable datagram.
///
/// Datagrams are the lever because §11's send seals inside the call (§16.7),
/// so one drain collects the packet and the size is chosen to the byte. The
/// frame is `DATAGRAM_LEN` (`0x31`): one type byte plus an explicit length
/// varint, 1 byte below 64 and 2 up to 16383 (§8.1). The emitted packet's
/// length is **asserted**, so a wrong varint guess fails here, loudly, and
/// never silently mis-calibrates a later assertion.
///
/// The datagram is a *marking* send (§7.4 — fresh application intent), so it
/// pins `last_send` at `now`, which is what both beacon tests arm from.
/// §7.5's passive debt and §7.4's arming bit, read straight off the core.
///
/// Both are `pub(crate)` on `Liveness` and reachable through
/// `Connection::liveness()`. Reading them is what keeps the fixtures below
/// from passing **vacuously**: a construction that failed to reach the held
/// state would otherwise satisfy "no deadline in the past" for the boring
/// reason that no keepalive was ever owed.
/// Assert the terminal `Timeout` of a drain is not retrospective.
///
/// **This is F1's pin.** `Driver::run` does `sleep_until(deadline)`; a
/// deadline at or before the `now` just handled completes immediately and
/// `handle_timeout` re-fires the same timer, forever. `None` is fine — it is
/// a parked connection, not a spinning one.
/// The companion half `CONTRACT-7b.md` §4.3 requires: **the death clock is
/// still announced**.
///
/// Suppressing `TimerKind::Liveness` alongside the keepalives satisfies
/// [`assert_not_retrospective`] trivially and turns a spinning connection
/// into an **immortal** one — the collapse ruling 182's beacon proof exists
/// to forbid. Without this assertion the whole file is passed by
/// `fn deadline() -> None`.
// ═══════════════════════════════════════════════════════════════════════
// 2. F1-a — the budget refuses the keepalive
// ═══════════════════════════════════════════════════════════════════════
/// `room == 0`, no contested mark: the **amplification** disjunct alone
/// blocks the beacon.
///
/// Mutation caught: `sync_liveness_timer` re-arming `PersistentKeepalive` at
/// `last_send + interval` without consulting `amplification`. Also caught: a
/// partial fix whose new arming condition tests only `contested`.
///
/// Not caught by, and deliberately not asserted as, `timer(...) == Some(fire)`
/// — that equality is precisely what the broken build satisfies.
/// The same hold, driven a second time at the **same instant**: the deadline
/// must not be the one just consumed.
///
/// This is the monotone-advance framing, and it is the one that survives a
/// build that clamps rather than suppresses. `CONTRACT-7b.md` §4.2 declines
/// a shell-side `max(d, now)` clamp for exactly this reason — *"`sleep_until`
/// completes immediately for exactly the same set of deadlines"* — and a
/// core-side clamp has the same defect one layer down. A clamped core
/// announces `Some(fire)` at `fire` and this assertion still fails.
///
/// Mutation caught: any fix that keeps the deadline at `last_send + interval`
/// and merely bounds it below by `now`.
// ═══════════════════════════════════════════════════════════════════════
// 3. F1-b — a pending contested mark blocks the keepalive
// ═══════════════════════════════════════════════════════════════════════
/// `room == 30` **exactly**: §7.3 admits the 30-byte keepalive and refuses
/// the 31-byte probe, so the mark stays pending and the **contested**
/// disjunct alone blocks the beacon.
///
/// This is the one-byte gap the whole file is built around. At `room == 30`:
///
/// * `amplification.admits(KEEPALIVE_LEN)` is **true** — 30 ≤ 30;
/// * `contested.is_pending()` is **true** — the probe needs 31.
///
/// Mutation caught: a fix whose new arming condition consults the budget but
/// not the mark. Such a build passes every test in §2 of this file and spins
/// here, which is precisely why `CONTRACT-7b.md` §4.1 requires **one**
/// predicate shared by `transmit_keepalive` and `sync_liveness_timer`
/// rather than two copies — *"a build that states it twice is the build that
/// drifts"*.
/// The mirror of the fixture assertion above, stated as its own test so the
/// one-byte calibration cannot rot silently.
///
/// At `room == 31` the probe **is** admitted, the mark arms, and the
/// keepalive is no longer blocked by `contested`. If this ever stops holding,
/// [`beacon_blocked_by_a_pending_mark_does_not_re_arm_in_the_past`] has
/// stopped isolating the contested disjunct and has quietly become a second
/// copy of the budget test.
///
/// Working rule 9's shape: the bound is only a separation while the
/// neighbouring value falls on the other side of it.
// ═══════════════════════════════════════════════════════════════════════
// 4. Termination — the driver loop, run at the core
// ═══════════════════════════════════════════════════════════════════════
/// The step cap [`drive_like_the_shell`] is given, and the separation it buys.
///
/// **[round 41 item 11 — 2026/08/18]** Both call sites passed a literal `64`,
/// under a comment calling it *"deliberately loose: the point is bounded, not
/// small, and a tight cap would turn an unrelated recovery-timer change into a
/// red here"*. Loose to the point of **vacuous** — working rule 9: *a bound is
/// only a test if the degenerate case violates it*, and nothing violated 64.
/// The three quantities, measured at this commit rather than argued:
///
/// * the shipped build reaches `Timeout(None)` in **1** step. The budget
/// admits no probe, so ruling 249 announces no `Pto` at all; the only timer
/// left is §7.4's liveness, one sleep to `install + DEAD_TIMEOUT`.
/// * the build this loop's own assertion message names — *"a held keepalive
/// re-offered without bound"* — scores `DEAD_TIMEOUT / BEACON` = **25**.
/// Measured, not assumed: driving an **un**held beacon on a validated
/// address is exactly that behaviour, and it takes 25 steps to die. 25 < 64,
/// so the old cap could not fail the one build it was written against.
/// * a build that walked §13.3's ladder here instead (ruling 249's defect, in
/// the variant whose anchor advances, so the retrospection assert above does
/// not catch it) reaches `DEAD_TIMEOUT` in **6** firings at ruling 254's 2³
/// — arithmetic, not measured: a ~1.02 s first interval doubling to 8× sums
/// 1, 3, 7, 15, 23, 31 intervals, and 25 s falls between the fifth and the
/// sixth. At the old 2⁶ it was **5**, a longer ladder having fewer rungs
/// inside the same 25 s — so tightening here is not a consequence of 254 so
/// much as something 254 made worth doing properly.
///
/// The separating band is therefore `1 ≤ cap < 25`, and every shape above
/// sits outside it. 4 keeps a three-step margin for exactly the unrelated
/// recovery-timer change the old comment was protecting, and still fails the
/// 25-step build by a factor of six.
const FIXED_POINT_CAP: usize = 4;
/// The upper end of that band, pinned to the constants it is derived from
/// rather than transcribed: a cap at or above a beacon re-offered every
/// interval until §7.4 reaps the session asserts nothing at all.
const _: = assert!;
/// Simulate `Driver::run`'s steps 4 and 5 against the core alone: sleep to
/// whatever deadline was announced, hand it back, repeat.
///
/// Returns the number of steps taken. Panics on the two failures that *are*
/// the livelock:
///
/// * an announced deadline at or before the `now` that produced it — the
/// spin, caught on its **first** iteration rather than after ten million;
/// * more than `cap` steps without reaching `Timeout(None)` — the spin,
/// caught by exhaustion if some other route produced it.
///
/// This is the shape `CONTRACT-7b.md` §4.3 warns cannot be written against a
/// runtime: *"on tokio's paused clock a livelock is an infinite loop that
/// never advances virtual time, so the test hangs rather than fails."* At the
/// core there is no clock to advance, so the loop is finite by construction
/// and the failure is an assertion instead of a hang.
/// The budget-refused hold, driven to a fixed point.
///
/// Mutation caught: the spin, on the **first** step past the refusal
/// (`next == now`). A build that suppresses correctly announces no `Pto`
/// at all while the budget is closed (§13.3, ruling 249 — before that
/// ruling it walked the PTO backoff here) and dies of §7.4's liveness at
/// `install + DEAD_TIMEOUT`, in a handful of steps.
///
/// The cap read `64` until round 41 item 11 measured what it was separating:
/// nothing. See [`FIXED_POINT_CAP`] — this build takes **1** step, and the
/// re-offering build the assertion names takes 25.
/// The pending-mark hold, driven to a fixed point.
///
/// Same pin, other disjunct. `ADVERSARIAL-liveness.md` calls this variant
/// *unbounded* on the grounds that the mark clears only on a peer send; the
/// death clock still bounds it here, because the shaping datagram is a
/// marking send and §7.4 arms from that. The **timer** is what is unbounded,
/// not the connection, and this test pins the timer.
/// The **control**: an unheld beacon on a validated address must keep firing.
///
/// Mutation caught: over-suppression. A build that solves F1 by never arming
/// `PersistentKeepalive` at all passes every other test in this file — no
/// deadline is ever retrospective if no deadline is ever armed — and has
/// deleted §7.5's beacon. Here the budget is absent entirely (`connect()`
/// anchors a **validated** address, §3.2), no mark is taken, and the beacon
/// must fire, seal a 30-byte datagram, and re-arm **one interval on**.
///
/// This is the assertion that makes the rest of the file mean something.
/// The control for the **other** timer: §7.5's passive keepalive still arms,
/// still fires, and still clears itself.
///
/// `TimerKind::Keepalive` has no test of its own anywhere above, because §5
/// of this file argues its *held* state is not constructible. That makes an
/// over-suppressing fix on this path invisible: a `keepalive_can_leave`
/// wired into the passive arm with the wrong sense, or applied where it does
/// not belong, would stop the passive keepalive dead and nothing else here
/// would notice. So the admissible path gets pinned end to end.
///
/// The route is §7.4's, exactly: an authenticated fresh receive sets §7.5's
/// passive debt (ruling 195 — the flag, not `R > S`) and disarms the death
/// clock; the keepalive is then owed at `last_send + KEEPALIVE_TIMEOUT`,
/// where `last_send` is still the install pin because nothing marking has
/// been sealed since.
///
/// Mutation caught: `sync_liveness_timer` failing to arm `Keepalive`;
/// `transmit_keepalive` refusing on a validated address; the marking seal not
/// clearing the debt, which would re-arm the keepalive forever at a
/// `last_send` that does move — a walking timer rather than a stuck one, and
/// the assertion that catches it is the final `None`.
/// The last control: **`Timeout(None)` is not a free pass.**
///
/// [`assert_not_retrospective`] treats `None` as acceptable, because a parked
/// connection really does announce it. That makes `fn deadline() -> None` a
/// build which passes every non-retrospection assertion in this file. The
/// three [`assert_death_clock_armed`] calls block it at the instants they
/// cover; this test blocks it at the install, which is the instant §7.4 makes
/// load-bearing — *"the handshake is the arming event … without it a session
/// that receives nothing would be held **forever**, since §7.6 is deleted and
/// liveness is the only reaper."*
///
/// Mutation caught: any suppression broad enough to reach
/// `TimerKind::Liveness`, which `CONTRACT-7b.md` §4.1 forbids in terms —
/// *"suppressing it too would turn a spinning connection into an **immortal**
/// one, which is worse."*
// ═══════════════════════════════════════════════════════════════════════
// 5. The passive keepalive — a reported reachability limit
// ═══════════════════════════════════════════════════════════════════════
/// **No test here covers `TimerKind::Keepalive` in its *held* state, and this
/// is the reason.** The admissible path is pinned by
/// [`an_admissible_passive_keepalive_fires_and_then_disarms`]; what is missing
/// is the refusal, and it is missing because it does not appear to exist.
///
/// **[Ruling 265.]** It exists, and it is now covered — in `tests_park.rs`
/// rather than here, because the construction needs a fragmented replay
/// window and a full congestion window and belongs with the assertions it
/// serves. This note is kept whole, with its corrections marked inline, for
/// the reason ruling 220 kept the last one: the argument is sound down to a
/// single step, and deleting it would lose both the step and the flag the
/// author put on it.
///
/// `ADVERSARIAL-liveness.md` F1 names two *unbounded* variants, both of which
/// require §7.5's **passive** debt — `owes_passive_keepalive()` — to be set
/// at the instant of the refusal. On this author's arithmetic that state
/// cannot be built next to either disjunct of the refusal at the core, and
/// the obstruction is one inequality:
///
/// > The debt is set **only** by an authenticated fresh receive
/// > (`Liveness::on_authenticated_fresh_recv`). The same datagram credits
/// > §7.3's budget by `AMPLIFICATION_FACTOR × len`, and the smallest packet
/// > a peer can send is §3.4's empty plaintext at 30 bytes. So **every**
/// > receive that sets the debt also raises the room by at least
/// > `3 × 30 = 90` bytes — strictly more than the 30-byte keepalive
/// > `transmit_keepalive` asks for, and strictly more than the 31-byte probe
/// > whose refusal is the only thing that keeps a mark `Pending`.
///
/// Both disjuncts fall out of it:
///
/// * **amplification.** Room ≥ 90 > 30 the instant the debt is set. To
/// refuse the keepalive the room must be spent back below 30 *without*
/// clearing the debt — and only a **non-marking** send preserves it
/// (§7.4: `seal` clears the debt, `seal_quiet` does not). The quiet set is
/// retransmissions, credit frames, RESET_STREAM, pure ACKs and PTO probes.
/// A pure ACK is the only member that is also non-ack-eliciting, and it
/// costs ~35 bytes against the ≥ 90 its own trigger credited: the room
/// grows monotonically. Every other member sets `armed`, which puts the
/// connection back in the **bounded** case this file already tests.
///
/// **[CORRECTED 2026/08/18 — ruling 265.] A pure ACK does not cost ~35
/// bytes. It is sized to the room, and the room does not grow
/// monotonically.** [`Connection::packing`](super::Connection::packing)
/// clamps the plaintext to `room − 30` (ruling 203/207(c)) and
/// [`ack::derive`](super::ack::derive) truncates newest-first *at that
/// room*, up to `MAX_ACK_RANGES` = 64 pairs — so on a replay window with
/// enough gaps **one** non-marking, non-arming packet takes the whole
/// budget. Measured in `tests_park.rs`: a roam funding `3 × 35 = 105`
/// datagram bytes is answered by a **105-byte ACK carrying 33 range
/// pairs**, leaving `room == 0` with `armed == false` and the debt set.
/// The end-to-end form is `tests/story_park.rs`, where a 39-byte roam
/// trigger funds 117 bytes and the ACK takes all 117.
///
/// The author flagged its own uncertainty here — *"building it needs
/// assumptions about `Packing`'s budget clamp that a blind author would
/// be guessing at"* — and that flag was exactly right: the clamp is the
/// step. Nothing else in this file's reasoning moves, and the
/// `Pending`-mark disjunct below is untouched.
/// * **contested.** A mark is `Pending` only while the budget refuses the
/// 31-byte probe. The same ≥ 90 bytes of credit release it on the very
/// pump the receive triggers, so a pending mark and a fresh receive cannot
/// coexist. [`one_more_byte_of_room_releases_the_probe_and_the_isolation_with_it`]
/// pins that release directly.
///
/// **What I am *not* claiming.** A route through a *large* quiet
/// retransmission — enough unacked stream data that a PTO probe drains the
/// post-receive room below 30 in one packet — would reach
/// `owes_passive_keepalive() && room < 30` with `armed == true`. That is a
/// **bounded** spin (the death clock runs), it is the same defect this file
/// already pins through the beacon, and building it needs assumptions about
/// `Packing`'s budget clamp that a blind author would be guessing at. I have
/// left it out deliberately rather than ship a test whose fixture I cannot
/// verify — and flagged it, because if the implementer's `keepalive_can_leave`
/// is wired into the beacon's arm but not the passive one, nothing in this
/// file catches it.
///
/// **The consequence for the finding itself:** F1's *"nothing bounds the spin
/// at all"* rests on `armed == false`, and `armed` is false only immediately
/// after a receive — which is exactly the moment the room is largest. The
/// unbounded claim looks to me **overstated**; the bounded claim is exact and
/// is what this file tests. Reported, not resolved (working rule 3).
///
/// **[CORRECTED 2026/08/18 — ruling 265.]** The unbounded claim was
/// **understated**, not overstated. `armed == false` *is* the moment the
/// room is largest, and the room is spent to nothing in the very next
/// packet — so the reachable state is not an unbounded spin but something
/// the announce-gate turned into an unbounded **park**: `Timeout(None)`,
/// no timer armed, alive and silent at 2.4 × `DEAD_TIMEOUT`. Ruling 265
/// closes it by announcing §7.4's own deadline for a keepalive that is
/// owed and vetoed.
///
/// Two things worth keeping from this note rather than deleting with it.
/// The author **flagged the exact step it was unsure of** and left the test
/// out rather than shipping a fixture it could not verify — that flag is
/// what the measurement was aimed at, and it was aimed correctly. And the
/// **`contested` half above is untouched**: a pending mark and a fresh
/// receive still cannot coexist, for precisely the reason given.
// ═══════════════════════════════════════════════════════════════════════
// 5. What makes suppression safe — the integrator's addition
// ═══════════════════════════════════════════════════════════════════════
/// A suppressed beacon **comes back** when the hold lifts.
///
/// **[Integrator, ruling 220.]** This author and the implementer closed F1
/// differently, both correctly: the author expected the beacon armed in the
/// future while held, the implementer suppresses it entirely. Suppression is
/// stronger — it makes the spin unreachable rather than survivable — but it
/// moves a burden the author's design did not carry. **An armed-in-the-future
/// beacon is self-healing; a suppressed one is only as good as whatever
/// re-arms it.**
///
/// So this is the assertion that the disagreement created, and neither blind
/// agent could have been asked for it: the author's design did not need it,
/// and the implementer had no test file to write it in.
///
/// Working rule 9: the degenerate build this separates is the one that
/// suppresses and **never re-arms** — a silently disabled keepalive on a
/// connection that has explicitly configured one. Every other assertion in
/// this file passes against that build, including all of section 2's, since
/// "no deadline in the past" is satisfied most easily by no deadline at all.