test: make the retirement notices police their own precondition
The retirement form deliberately keeps a retired arm PASSING with a message naming its successor, so a red is never hidden. That form has an UNSTATED PRECONDITION -- the successor is live -- and nothing enforced it. Chain two retirements and the guarantee degrades silently: a reader follows the pointer and lands on prose explaining what used to be there, while the arm goes on printing a pass. Observed live: gate-sd-p1 arm 5 names gate-sd-p3 arm 3, which this branch retired, so that chain is now two hops with a notice in the middle. Filed as t-a16a6c; these three notices are mine, so they are fixed here.
A retired arm is a PASSING arm, not a no-op -- it still executes -- so the assertion belongs immediately before the pass it guards.
It pins the successor's ARM, not just its file: a file-existence check would pass today, since gate-sd-p3.sh still exists and it is the arm inside it that would be gone. It pins the CALL SITE rather than the runtime marker, because the call site is the static artifact a reader greps.
The failure names its own remedy, including "do NOT delete this check to make the gate green; that is the failure it exists to catch" -- a notice that fails without saying what to do just moves the puzzle.
gate-sd-p7 uses grep, not rg: its tool preflight requires grep and NOT ripgrep, so an rg here would have been an absent-tool false negative inside a check whose whole purpose is to not be quiet.
SABOTAGE-TESTED AS A 2x2, because one plant can only demonstrate half of it -- and the first run did exactly that, leaving gate-sd-p7's guard shown able to stand down but never to fire:
gate-sd-p3 (succ arm 5) gate-sd-p7 (succ arm 6)
arm 5 removed RED exit 1 green exit 0
arm 6 removed green exit 0 RED exit 1Each gate fires on its own successor and stands down on the other's. Both sabotage runs restore from a saved copy under trap EXIT INT TERM and reported RESTORED-CLEAN; an earlier attempt without a trap hit the tool timeout mid-plant and left a live sabotage in the tree.
test/gate-sd-p3.sh | 29 +++++++++++++++++++++++++++++
test/gate-sd-p7.sh | 26 ++++++++++++++++++++++++++
2 files changed, 55 insertions(+)test/gate-sd-p3.shmodified
test "$located" -ge 6 || die "locations verified at only $located nodes"pass 2 "equality over $texts text / $codes code / $refs reference / $urls url-link / $paras paragraph nodes; locations at $located nodes"# THE SUCCESSOR MUST STILL EXIST, AND SOMETHING MUST ASSERT IT.## The retirement form deliberately keeps a retired arm PASSING, with a message# naming its successor, so that a red is never hidden. **That form has an# unstated precondition: the successor is live.** Chain two retirements and the# guarantee degrades silently -- a reader follows the pointer and lands on prose# explaining what used to be there, while the arm goes on printing a pass.# Nothing checked it, and this repository has three retirement notices in flight.## A retired arm is a PASSING arm, not a no-op: it still executes, so the check# belongs here, immediately before the pass it guards.## It pins the successor's CALL SITE (`pass N `) rather than its runtime marker,# because the call site is the static artifact a reader greps and the marker only# exists once the gate has run. A file-existence check would NOT do: this file's# successor lives in a file that still exists; it is the ARM that would be gone.successor_arm_exists() { # $1 = file, $2 = arm number test -f "$1" && rg -q "^pass $2 " "$1"}# ---------------------------------------------------------------- ARM 03# RETIRED 2026-08-22 BY THE LEADER, AT SD-P8, UNDER FABLE'S RULING OF# 2026-08-11. This arm has NOT silently disappeared and its red has NOT been# before/after differential scoped to your own base, embed the base SHA in the# arm, and make a superseded base fail with "baseline superseded -- see ruling"# rather than with a bare diff. `test/gate-sd-p8.sh` arm 5 is a worked example.successor_arm_exists test/gate-sd-p8.sh 5 \ || die "arm 03: this retirement names test/gate-sd-p8.sh arm 5 as its successor, and that arm no longer exists. The retirement form keeps a retired arm PASSING so that a red is never hidden, which assumes the successor is LIVE. With the successor gone this arm asserts nothing and says so to nobody. REMEDY: find what superseded gate-sd-p8 arm 5, re-point the notice above at it, and update this check -- or restore the arm. Do NOT delete this check to make the gate green; that is the failure it exists to catch."pass 3 "RETIRED 2026-08-22 at SD-P8 under Fable's 2026-08-11 ruling; superseded by test/gate-sd-p8.sh arm 5 (see comment)"# ---------------------------------------------------------------- ARM 04# `test/gate-sd-p8.sh` arm 5 additionally shows that the D8 migration moved# NOTHING in the lowering of the manuscripts SD-P3 cared about, header aside,# which is the closest standing equivalent of what this arm certified.successor_arm_exists test/gate-sd-p8.sh 5 \ || die "arm 04: this retirement names test/gate-sd-p8.sh arm 5 as part of what covers its property, and that arm no longer exists. See the remedy under arm 03 above; the same reasoning applies."pass 4 "RETIRED 2026-08-22 at SD-P8 with arm 3, whose changed-set was its only input; property covered by arm 1, arm 2 and test/gate-sd-p8.sh arm 5 (see comment)"# ---------------------------------------------------------------- ARM 05test/gate-sd-p7.shmodified
|| die "arm 01: $witnesses core witnesses for $disjuncts block-node? disjuncts"pass 1 "section-profile suite ${suite_total}/${suite_total}, ${sections_built} sections built through a profile, ${witnesses} core witnesses = ${disjuncts} block-node? disjuncts"# THE SUCCESSOR MUST STILL EXIST, AND SOMETHING MUST ASSERT IT.## The retirement form deliberately keeps a retired arm PASSING, with a message# naming its successor, so that a red is never hidden. **That form has an# unstated precondition: the successor is live.** Chain two retirements and the# guarantee degrades silently -- a reader follows the pointer and lands on prose# explaining what used to be there, while the arm goes on printing a pass.## A retired arm is a PASSING arm, not a no-op: it still executes, so the check# belongs here, immediately before the pass it guards.## It pins the successor's CALL SITE (`pass N `) rather than its runtime marker,# because the call site is the static artifact a reader greps and the marker only# exists once the gate has run. A file-existence check would NOT do: the# successor lives in a file that still exists; it is the ARM that would be gone.# `grep`, not `rg`: this gate's tool preflight requires grep and not ripgrep.successor_arm_exists() { # $1 = file, $2 = arm number test -f "$1" && grep -qE "^pass $2 " "$1"}# ---------------------------------------------------------------- ARM 02# RETIRED 2026-08-23 BY THE LEADER, AT SD-P8, UNDER FABLE'S RULING OF# 2026-08-11. This arm has NOT silently disappeared and its red has NOT been# EVIDENCE PRESERVED: `test/transcripts/sd-p7/construction-before.txt` stays in# the tree unmodified, and the `test -f` preconditions still assert it is# present. It is SD-P7's reviewed evidence.successor_arm_exists test/gate-sd-p8.sh 6 \ || die "arm 02: this retirement names test/gate-sd-p8.sh arm 6 as its successor, and that arm no longer exists. The retirement form keeps a retired arm PASSING so that a red is never hidden, which assumes the successor is LIVE. With the successor gone this arm asserts nothing and says so to nobody. REMEDY: find what superseded gate-sd-p8 arm 6, re-point the notice above at it, and update this check -- or restore the arm. Do NOT delete this check to make the gate green; that is the failure it exists to catch."pass 2 "RETIRED 2026-08-23 at SD-P8 under Fable's 2026-08-11 ruling; superseded by test/gate-sd-p8.sh arm 06, which asserts on structure rather than offsets (see comment)"# ---------------------------------------------------------------- ARM 03