A learner's course copy came out missing exactly one node — the exercise workbench — with no error logged at the point of failure. The install correctly never recorded itself as complete, so the Store's sequencing is not at fault; what is missing is any signal naming what did not land.
Evidence
MeshWeaver.Education run 32503778979, shard 2, portal sha256:88eb1afc… (3.0.0-rc5.ci.4547). The copy of ThinkInStreams into e2e-admin runs 16:52:07.7 → 16:52:12.1.
The absent node — raised while rendering the learner's page, 16 minutes later:
17:08:38 DeliveryFailureException: No node found at
'e2e-admin/ThinkInStreams/09-RxVersusIx/Exercise/OperatorForge/Source/OperatorForge'
17:08:38 Area target is gone (routing NotFound) — rendering unavailable placeholder
It is the only one of lesson 09's ten nodes with no Node created at line. Its structural twin in the same lesson, 09-RxVersusIx/Solution/OperatorForge/Source/OperatorForgeSolved, has one — both are the same {Page}/Source/{Id} shape under the same declared root.
No install record. e2e-admin/_Install/Systemorph-MeshWeaver.Education is never created, for the remaining 17 minutes the job runs. That is the design working — the manifest is the last phase, so a partial install cannot record itself as installed.
It is not the normal outcome. On 2026-08-19, the same course copied into the same user landed 11,9,10,10,9,9,9,9,10 = 86 nodes with that node present and rendering (156 references in that run's portal log).
Nothing failed loudly. Grepping the copy window for error|fail|refus|abort|timeout|denied|exceed turns up only unrelated ObjectPolymorphicConverter and SYNC_STREAM warnings. There is no line saying the install stopped, which root it was on, or which node it could not write.
Why this is worth a fix even though the record was withheld
The two states are indistinguishable from outside:
- an install still in flight, and
- an install that stopped partway and will never finish,
and both present to the learner as a course whose pages render prose with the workbench missing. The absent record means "not installed", which is correct but not actionable — nobody can tell which node is missing without diffing the partition against the repo by hand, which is what I had to do.
What would have made this diagnosable in seconds:
- A terminal line naming the failure — the phase, the root, and the node path — when the chain stops. Right now the last thing in the log is an ordinary create, then silence.
- The verify phase reporting the DIFF, not just a pass/fail. It knows the declared roots and what landed; the set difference is the answer.
I have deliberately not guessed at the cause — the log does not support one, and I would rather file the observation than a theory. Happy to hand over the artifacts.
Found while getting MeshWeaver.Education's main green (Systemorph/MeshWeaver.Education#191, #192). Related: #1996 (a hub recycle leaving an open page dead) came out of the same investigation.
🤖 Generated with Claude Code
A learner's course copy came out missing exactly one node — the exercise workbench — with no error logged at the point of failure. The install correctly never recorded itself as complete, so the Store's sequencing is not at fault; what is missing is any signal naming what did not land.
Evidence
MeshWeaver.Education run 32503778979, shard 2, portal
sha256:88eb1afc…(3.0.0-rc5.ci.4547). The copy ofThinkInStreamsintoe2e-adminruns 16:52:07.7 → 16:52:12.1.The absent node — raised while rendering the learner's page, 16 minutes later:
It is the only one of lesson 09's ten nodes with no
Node created atline. Its structural twin in the same lesson,09-RxVersusIx/Solution/OperatorForge/Source/OperatorForgeSolved, has one — both are the same{Page}/Source/{Id}shape under the same declared root.No install record.
e2e-admin/_Install/Systemorph-MeshWeaver.Educationis never created, for the remaining 17 minutes the job runs. That is the design working — the manifest is the last phase, so a partial install cannot record itself as installed.It is not the normal outcome. On 2026-08-19, the same course copied into the same user landed
11,9,10,10,9,9,9,9,10 = 86nodes with that node present and rendering (156 references in that run's portal log).Nothing failed loudly. Grepping the copy window for
error|fail|refus|abort|timeout|denied|exceedturns up only unrelatedObjectPolymorphicConverterandSYNC_STREAMwarnings. There is no line saying the install stopped, which root it was on, or which node it could not write.Why this is worth a fix even though the record was withheld
The two states are indistinguishable from outside:
and both present to the learner as a course whose pages render prose with the workbench missing. The absent record means "not installed", which is correct but not actionable — nobody can tell which node is missing without diffing the partition against the repo by hand, which is what I had to do.
What would have made this diagnosable in seconds:
I have deliberately not guessed at the cause — the log does not support one, and I would rather file the observation than a theory. Happy to hand over the artifacts.
Found while getting MeshWeaver.Education's
maingreen (Systemorph/MeshWeaver.Education#191, #192). Related: #1996 (a hub recycle leaving an open page dead) came out of the same investigation.🤖 Generated with Claude Code