Reports · September 2026

Every incident report, and what happened.

The record is continuous since 2026-09-05 22:10:00 UTC. Before that instant there is no record, and a missing turn means nothing.

By month: October 2026 · September 2026

2026-09-30 12:20:00 UTC

anchor_retried · resolved · incidents/2026/09/30/1220-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-30 04:10:00 UTC

anchor_retried · resolved · incidents/2026/09/30/0410-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-30 02:30:00 UTC

anchor_retried · resolved · incidents/2026/09/30/0230-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-25 14:40:00 UTC

anchor_retried · resolved · incidents/2026/09/25/1440-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-25 00:10:00 UTC

anchor_partial · unresolved · incidents/2026/09/25/0010-01.jws

Missing in gitlab

0010-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-25 00:08:05 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the provider said it did not have a file it did have

"Missing in gitlab" is incorrect. `emissions/2026/09/25/0010-commitment.jws` is anchored on GitLab, in commit 9182949b9ddcf75401cb35d564a04b63738772f1, and GitLab stamped it at 00:08:55 UTC. That date can be read with no credentials by asking for the file's commits. This case differs from the seven of the previous day. There the emitter never got an answer to the question "is the file there?". Here it got one, and it was false. The timeline, which can be followed in the log below. The second attempt went out at 00:08:35 and was given up at 00:08:53 with no answer; GitLab created the file two seconds later, at 00:08:55. The third attempt went out at 00:09:08 — thirteen seconds after the file already existed— and on asking about it was told it was not there. With that answer, the emitter declared it missing and closed the commitment's window at 00:09:23. That is two things, not one. The provider answered that it did not have a file it did have, in the middle of a disruption it had declared itself; and this system took that answer as a fact, with no room to doubt it. And there is a third, which is the one that could have been avoided: this report was signed at 00:11:37, two minutes and fourteen seconds later. At that moment GitLab was answering in one second — the last attempt in the log shows it— and it was not asked again. A stale conclusion was signed while the connection was open and being used, in that same second, to publish this file. ⚠️ When reading the log, bear in mind that it mixes the attempts of two different files. Those before 00:10:00 belong to the commitment; those after, to the emission, which cannot be published until the round exists. The attempt that says "ya estaba" (it was already there) belongs to the emission, not to the commitment. Nothing about the emission is affected: both of the turn's files are anchored in both repositories, signed, and they verify. The commitment was anchored before its round existed. The signed report is not rewritten; this note sits beside it.

2026-09-25 00:00:00 UTC

anchor_timeout · unresolved · incidents/2026/09/25/0000-01.jws

0000-commitment.jws could not be confirmed on gitlab

The write went out and the answer never came back. Asking gitlab afterwards whether the file had landed got no answer either, so it is not known whether it is there. It is not declared missing: not knowing is not knowing that it is not.

This can be checked with no credentials and without asking us for anything: ask the provider for that file's commits. If a commit shows up, the file is anchored and the provider stamped the date.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:50:00 UTC

anchor_retried · resolved · incidents/2026/09/24/2350-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:40:00 UTC

anchor_timeout · unresolved · incidents/2026/09/24/2340-01.jws

2340-commitment.jws could not be confirmed on gitlab

The write went out and the answer never came back. Asking gitlab afterwards whether the file had landed got no answer either, so it is not known whether it is there. It is not declared missing: not knowing is not knowing that it is not.

This can be checked with no credentials and without asking us for anything: ask the provider for that file's commits. If a commit shows up, the file is anchored and the provider stamped the date.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:30:00 UTC

anchor_retried · resolved · incidents/2026/09/24/2330-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:20:00 UTC

anchor_timeout · unresolved · incidents/2026/09/24/2320-01.jws

2320-commitment.jws, 2320-emission.jws could not be confirmed on gitlab

The write went out and the answer never came back. Asking gitlab afterwards whether the file had landed got no answer either, so it is not known whether it is there. It is not declared missing: not knowing is not knowing that it is not.

This can be checked with no credentials and without asking us for anything: ask the provider for that file's commits. If a commit shows up, the file is anchored and the provider stamped the date.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:10:00 UTC

anchor_timeout · unresolved · incidents/2026/09/24/2310-01.jws

2310-commitment.jws, 2310-emission.jws could not be confirmed on gitlab

The write went out and the answer never came back. Asking gitlab afterwards whether the file had landed got no answer either, so it is not known whether it is there. It is not declared missing: not knowing is not knowing that it is not.

This can be checked with no credentials and without asking us for anything: ask the provider for that file's commits. If a commit shows up, the file is anchored and the provider stamped the date.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 23:00:00 UTC

anchor_timeout · unresolved · incidents/2026/09/24/2300-01.jws

2300-commitment.jws, 2300-emission.jws could not be confirmed on gitlab

The write went out and the answer never came back. Asking gitlab afterwards whether the file had landed got no answer either, so it is not known whether it is there. It is not declared missing: not knowing is not knowing that it is not.

This can be checked with no credentials and without asking us for anything: ask the provider for that file's commits. If a commit shows up, the file is anchored and the provider stamped the date.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 17:00:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1700-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 16:20:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1620-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 16:10:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1610-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 15:40:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1540-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 15:30:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1530-01.jws

Missing in gitlab

1530-commitment.jws, 1530-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 15:28:09 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1530-commitment.jws in commit 3f0b9af83758c13cf8745272425a0036266b9a3e, stamped at 15:28:13 UTC and emissions/2026/09/24/1530-emission.jws in commit 5c02412758886f6128938aa6e2a9c5d7e9b2014a, stamped at 15:30:07 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 15:20:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1520-01.jws

Missing in gitlab

1520-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 15:18:16 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1520-commitment.jws in commit 511fb5daf34abb422b4079538902234eae39b9c0, stamped at 15:18:23 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 15:10:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1510-01.jws

Missing in gitlab

1510-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 15:08:24 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1510-commitment.jws in commit cc90a4998de9a74c55b92c0e1f0c6a847a7c2a0c, stamped at 15:08:33 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 15:00:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1500-01.jws

Missing in gitlab

1500-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:58:15 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1500-commitment.jws in commit e0bc37db9fd253f65e6c145b37b54e91ea184dcb, stamped at 14:58:25 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 14:50:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1450-01.jws

Missing in gitlab

1450-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:48:12 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1450-commitment.jws in commit b54f4b143e5e4e87b9dd9c18f8beec85fb2cd418, stamped at 14:48:20 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 14:40:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1440-01.jws

Missing in gitlab

1440-commitment.jws, 1440-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:37:57 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

Part of this is correct: one file is anchored on GitLab, the other is genuinely missing

Half of this is correct. emissions/2026/09/24/1440-commitment.jws is genuinely absent from GitLab and is held by GitHub alone. But emissions/2026/09/24/1440-emission.jws, in commit 7970c2de3e0f62dc6505afd0c47388b43bb6b787, stamped at 14:44:20 UTC is on GitLab. That write went through and the answer never came back, and unlike the other turns of the day no retry ever reached GitLab's "a file with this name already exists", so nothing contradicted the first conclusion. This was during GitLab incident 23015. What the turn lost is redundancy on one file, not proof: both files are anchored, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 14:30:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1430-01.jws

Missing in gitlab

1430-commitment.jws, 1430-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:28:05 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 14:20:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1420-01.jws

Missing in gitlab

1420-commitment.jws, 1420-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:18:17 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 14:10:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1410-01.jws

Missing in gitlab

1410-commitment.jws, 1410-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 14:08:05 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 14:00:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1400-01.jws

Missing in gitlab

1400-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:58:14 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:50:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1350-01.jws

Missing in gitlab

1350-commitment.jws, 1350-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:47:56 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:40:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1340-01.jws

Missing in gitlab

1340-commitment.jws, 1340-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:37:55 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:30:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1330-01.jws

Missing in gitlab

1330-commitment.jws, 1330-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:27:58 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:20:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1320-01.jws

Missing in gitlab

1320-commitment.jws, 1320-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:17:57 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:10:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1310-01.jws

Missing in gitlab

1310-commitment.jws, 1310-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 13:07:57 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 13:00:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1300-01.jws

Missing in gitlab

1300-commitment.jws, 1300-emission.jws is in github and missing in gitlab. It was anchored on 2026-09-24 12:58:09 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1300-commitment.jws in commit 9de9d6382fcdd5410a3ecad965ebdb2edbbefc29, stamped at 12:58:18 UTC and emissions/2026/09/24/1300-emission.jws in commit 53b84b2978a066b687e8ad6edf4919808faa21af, stamped at 13:00:18 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-24 12:50:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1250-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 12:30:00 UTC

anchor_retried · resolved · incidents/2026/09/24/1230-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-24 11:10:00 UTC

anchor_partial · unresolved · incidents/2026/09/24/1110-01.jws

Missing in gitlab

1110-commitment.jws is in github and missing in gitlab. It was anchored on 2026-09-24 11:08:17 UTC, and that date is stamped by github, not by us: it can be read without credentials by asking for the file's commits.

What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

"Missing in gitlab" is incorrect: the files are anchored there

"Missing in gitlab" is incorrect. It is on GitLab: emissions/2026/09/24/1110-commitment.jws in commit 7aa39b7f762da518f7a681b6a46cf326329380fd, stamped at 11:08:25 UTC. That date is stamped by GitLab and can be read with no credentials by asking for the file's commits, the same way it is done for the other anchor. The write did go through; what never came back was the answer, and the emitter gave it up when its deadline expired. To decide, it then asks for a second confirmation through a separate read, and during GitLab incident 23015 the repository was answering those reads more slowly than their deadline allows, so with no confirmation it recorded the anchor as missing. For the same reason, no redundancy was lost either: this turn's files are in both repositories, signed, and they verify. The signed report is not rewritten; this note sits beside it.

2026-09-17 09:30:00 UTC

anchor_retried · resolved · incidents/2026/09/17/0930-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-16 20:50:00 UTC

anchor_retried · resolved · incidents/2026/09/16/2050-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-14 16:40:00 UTC

anchor_retried · resolved · incidents/2026/09/14/1640-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-13 10:00:00 UTC

unknown_state · resolved · incidents/2026/09/13/1000-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-13 09:30:00 UTC

unknown_state · resolved · incidents/2026/09/13/0930-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-13 09:10:00 UTC

unknown_state · resolved · incidents/2026/09/13/0910-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-13 09:00:00 UTC

unknown_state · unresolved · incidents/2026/09/13/0900-01.jws

The commitment file is missing in GitHub (HTTP 500 and 502)

GitHub rejected this turn's commitment file five times in a row, between 08:57:58 and 08:59:22 UTC, with three HTTP 500 and two HTTP 502, and after that there was no time left for another attempt: the commitment has to be anchored before its drand round is born, so the emitter stopped trying. GitLab accepted the same file on the first attempt and stamped it at 08:57:55 UTC, 125 seconds before drand round 6461966 existed, so emissions/2026/09/13/0900-commitment.jws is in GitLab and missing in GitHub. The emission is not affected and verifies as usual, and the complete record of this turn can be verified in GitLab. What this turn lost is redundancy, not proof: the commitment sits in one provider instead of two, and it cannot be replaced afterwards, because a file uploaded later would carry a date after its round and would read as a commitment that arrived late. The gap was detected by the sweep of this same turn and reported here.

2026-09-13 08:50:00 UTC

unknown_state · resolved · incidents/2026/09/13/0850-01.jws

It reached every anchor, but it cost

The file is byte-for-byte identical in every anchor: the same signed bytes were re-sent, so only the commit date each provider stamps differs.

What was attempted

This log predates each attempt naming its file: the rows may belong to any of the turn's files.

2026-09-12 02:10:00 UTC

unknown_state · unresolved · incidents/2026/09/12/0210-01.jws

Commitment file missing on GitHub (HTTP 500)

GitHub returned HTTP 500 when the beacon published the commitment file for this turn, at 02:07:55 UTC. GitLab accepted the same file at the same moment, so emissions/2026/09/12/0210-commitment.jws is present on GitLab and absent on GitHub. The emission itself is unaffected and verifies normally: the commitment was anchored 124 seconds before drand round 6458266 existed, and the complete record for this turn can be checked on GitLab. What this turn lost is redundancy, not proof — the commitment is held by one provider instead of two. The emitter does not retry the commitment after a failed write, because it must be anchored before the drand round is born, so the file was not sent again. The gap was detected automatically 64 seconds later and reported here.

By month: October 2026 · September 2026