Transparency

What happened

The 6 September fixing was captured at its scheduled time and the day’s evidence bundle was preserved at 16:15 Europe/London.

The publication run then created and froze the record for the date at 16:30, and failed moments later, before the public surfaces were delivered. The step that stores the run’s own copy of the market reading had not written it, and reported that incomplete state as already stored. The independent check that runs afterwards found the missing custody and stopped the run.

The scheduled backup run at 16:36 saw that the day’s evidence had been preserved and stood down. It did not complete the publication.

Recovery used the evidence the original run had already preserved. No source was called again, no second observation was taken, and the frozen record was not rewritten. The 6 September record reached the public surfaces carrying the same bytes and the same hash it was frozen with.

What it means for the record

There is one record for 6 September and there has only ever been one. It was delivered late in the day rather than replaced, corrected or reconstructed.

It is a level-only record: a valid supply level and benchmark were frozen, and constituent attribution is unavailable for the date. That was determined before any of this, by the day’s own reading, and the recovery did not change it.

No other date is affected. Nothing was carried forward from another day, estimated, or filled in afterwards.

What changed

The publication path now stores and verifies its copy of the market reading before it freezes the day’s record, so a record cannot be published without the evidence custody that belongs to it.

The custody step can no longer report success on an incomplete local state. It reads the published state fresh and compares the bytes it requires, rather than treating the presence of a reference as proof.

A failed publication now begins recovery inside the same run, instead of waiting for the next scheduled trigger to arrive.

These changes were merged on 6 September 2026. They are proved by a regression suite that needs no network access, and by rehearsals that replay the day’s preserved evidence through the real publication code, including a rejected custody write followed by recovery within the same run. Those establish that the path completes when it is exercised. They are not an unattended publication observed in production, and Stable Tape will describe unattended publication as demonstrated once it has run unattended in production, and not before.

What happened

From 3 September 2026 the Daily Tape is fixed at 16:00 Europe/London. The record dated 3 September is the first one fixed under the new time.

The fixing time is the only thing that changed. The constituent universe, the sources, the calculation, the rounding and the freezing rules are all unchanged.

The capture software still targeted the old time when 3 September was observed, so the evidence preserved for that date records an intended instant under the old regime. That evidence has not been edited. The day’s designated observation was admitted against the new methodology instant, which it reached inside the required window.

What it means for the record

Records dated 2 September 2026 and earlier were fixed at 12:00 UTC and are unchanged. Records dated 3 September 2026 onward are fixed at 16:00 Europe/London. Every record states the cutoff it was fixed under, so the convention in force on a date is read from the record rather than assumed.

The two conventions are not the same point in the trading day, so daily movement is withheld across the change rather than measured between a reading taken under one and a reading taken under the other. The first movement measured from one 16:00 London fixing to the next is 5 September 2026.

16:00 in London is 15:00 UTC through British Summer Time and 16:00 UTC through Greenwich Mean Time. The fixing follows London, not a fixed offset from UTC, so it does not drift by an hour twice a year.

What changed

The fixing schedule is held as a dated rule rather than as a single constant, so each record resolves the schedule that was in force on its own date and no historical record can be restated under a later one.

Public surfaces that state a reading’s cutoff now also say when that cutoff belongs to a regime that is no longer the current one, so a frozen reading cannot be read as a statement of current policy.

What happened

No numerical Daily Tape edition was published on 28 August 2026.

The scheduled fixing process did not start in time to capture an admissible observation at the 12:00 UTC fixing moment. No designated observation was recorded for the date.

Because no admissible observation exists for 28 August, there is nothing from which a reading for that date could be assembled.

What it means for the record

There is no numerical Daily Tape edition for 28 August, and there will not be one later.

Stable Tape did not carry another day’s value forward, substitute a later snapshot, estimate the daily movement, reconstruct the reading afterwards, or publish a Signal for the date.

The date remains NO READING in the record and in the archive.

What changed

A single scheduled trigger was the only path into the fixing window. When it did not run on time, there was no second route to the observation and the day was lost.

Independent triggering, a standing watchdog and an early failure alert are being added, so that one trigger failing is no longer enough to lose a day. Stable Tape will describe them as in place once they have run unattended in production, and not before.

What happened

A genuine fixing-window observation was preserved, but the normal process did not complete the path from that observation to a published reading.

What it means for the record

The recovered 27 August reading is now the published numerical result for the date.

The governed recovery used the exact observation preserved within the fixing window. No later observation was substituted and the fixing was not retrospectively reconstructed.

It is a level-only record. Constituent attribution and Signal remain unavailable rather than being inferred.

This recovery establishes the result for 27 August; it does not by itself prove that the unattended fixing system is reliable.

What changed

Stable Tape added a governed recovery path that verifies the exact stored observation and fixing rule before recovery.

What happened

On 25 August, after that day’s 12:00 UTC Stable Tape fixing, FRED published its July 2026 M2SL observation.

The new observation was available before the 26 August fixing. Stable Tape’s monitoring recorded the newer M2 figure that evening, but the Daily Tape measures against a separately governed M2 reference that only changes when it is explicitly admitted.

Stable Tape did not advance that governed reference after the new FRED release became available, so the 26 August reading was published against the previous June M2 reference.

What it means for the record

The 26 August stablecoin supply reading is unchanged. So is the movement figure for that day, which is measured from supply levels and does not involve M2.

What is affected is the denominator used for Onchain Dollar Share, and the reference month stated alongside it. The reading used June M2 where the July observation was already available.

The rounded figures Stable Tape headlines, the share to two decimal places and the whole-dollar expression, are the same against either reference. The precise figure is not: the record states 1.3393%, and against the July observation the same supply would be 1.3356%.

The original 26 August edition stays exactly as published, with its original values and its original identity. Stable Tape does not quietly rewrite a reading it has already published.

Whether to publish a corrected reading for 26 August is a separate decision that has not yet been taken. Stable Tape’s existing correction process was built for a date that produced no reading at all, and it cannot describe this case honestly, so it is not being used here.

What changed

Until the July 2026 M2 observation has been admitted as the governed reference, the Daily Tape will withhold the share rather than measure it against the superseded June reference. Admission is a reviewed step and applies only to a fixing that has not yet happened; no earlier reading changes.

The check that stands between the two has been strengthened. A release that Stable Tape has durably observed, but has not yet admitted, now stops the next reading from quietly reusing the previous reference.

Regression coverage now includes the case this turned on: a release arriving between two fixings, tested in both directions, so that a release landing after a cutoff leaves that day alone and the same release stops the next one.

What happened

Following verification of the preserved 12:00 UTC observation and applicable M2 reference, Stable Tape has published the corrected fixing. The earlier No Reading status is superseded.

What it means for the record

The corrected level-only record is the primary and effective historical result for 24 August 2026.

Constituent movement and attribution remain unavailable, and no Signal is published.

What changed

The earlier No Reading event remains preserved and discoverable in the technical correction history; it is not a co-equal current result.

What happened

The day's source evidence was captured and preserved.

An internal publication check found that the proposed record did not describe its own evidence consistently. Stable Tape rejected the proposed record rather than publish an internally inconsistent reading.

No Daily Tape reading was published for 21 August 2026.

What it means for the record

No earlier figure was carried forward. The date remains an explicit gap in the record.

The underlying evidence remains preserved for audit.

What changed

The specific software defect was identified and repaired, and regression coverage was added for the failure it caused.

What happened

Stable Tape froze and published the supply level and benchmark for 20 August 2026.

A usable market-wide supply total was available, but the supporting constituent readings did not pass the checks needed to determine the constituent breakdown.

What it means for the record

This is a valid level-only record, not a no-reading date.

Constituent movement, attribution and Signal output are unavailable, not zero. Market change and ODS change are separate aggregate comparisons and are available only when the immediately prior calendar day is also an accepted record.

What changed

The frozen supply level and benchmark remain part of the published series. Outputs that depend on constituent evidence were withheld.

What happened

No numerical Daily Tape edition was published on 19 August 2026.

The day's source evidence was preserved, but the checks needed to support a valid record were not resolved by the decision deadline.

What it means for the record

No reading was published, no earlier figure was carried forward, and the date remains an explicit gap in the record.

What changed

The public record identifies 19 August as no reading rather than implying a numerical edition.

What happened

Stable Tape froze and published the supply level and benchmark for 18 August 2026.

A usable market-wide supply total was available, but the supporting constituent readings did not pass the checks needed to determine the constituent breakdown.

What it means for the record

This is a valid level-only record, not a no-reading date.

Constituent movement, attribution and Signal output are unavailable, not zero. Market change and ODS change are separate aggregate comparisons and are available only when the immediately prior calendar day is also an accepted record.

What changed

The frozen supply level and benchmark remain part of the published series. Outputs that depend on constituent evidence were withheld.

What happened

No numerical Daily Tape edition was published on 17 August 2026.

Required source-continuity checks did not support a valid Daily Tape record for the date.

What it means for the record

No reading was published, no earlier figure was carried forward, and the date remains an explicit gap in the record.

What changed

The public record identifies 17 August as no reading rather than implying a numerical edition.

What happened

No numerical Daily Tape edition was published on 16 August 2026.

The day's source observations were preserved as intended. Stable Tape recorded what each source reported before the fixing moment, at it, and after it, and kept that evidence.

The reading did not go ahead because the composition of the sources had shifted. Eleven token identities appeared that the series had not been reading before, and four it had been reading could no longer be accounted for. A change of that size has to be understood before it is allowed into a published reading, not after.

Stable Tape stopped there. No reading was assembled for the day, and none was estimated or carried over from another day.

What it means for the record

There is no numerical Daily Tape edition for 16 August, and there will not be one later.

Stable Tape did not carry another day's value forward, substitute a later snapshot, estimate the daily movement, reconstruct the reading afterwards, or publish a Signal for the date.

The date remains NO READING in the record and in the archive.

Whatever Stable Tape decides about those identities afterwards governs later readings only. A decision taken after the fact does not reopen 16 August or turn it into a published reading.

What changed

Nothing about the day's evidence needed to change: it was preserved, and it remains available. What the day showed is that the series needs a written account of how sources that appear and disappear are admitted, so a shift of this size is a decision taken in the open rather than a run that stops.

That account is being prepared. Until it exists, a day whose sources move this far will continue to stop rather than publish.

What happened

No numerical Daily Tape edition was published on 15 August 2026.

The day's source observations were preserved. This was the first day under the preservation practice described in the 14 August note, and that part held: each source was recorded before the fixing moment, at it, and after it, and the evidence was kept and checked.

Processing then stopped before a reading was completed. The evidence was in hand, but the day's reading was never assembled from it, and the fixing moment had long passed by the time that was settled.

Stable Tape did not assemble one afterwards.

What it means for the record

There is no numerical Daily Tape edition for 15 August, and there will not be one later.

Stable Tape did not carry another day's value forward, substitute a later snapshot, estimate the daily movement, reconstruct the reading afterwards, or publish a Signal for the date.

The date remains NO READING in the record and in the archive.

The preserved observations stay preserved. They are evidence of what the sources reported that day, not a reading, and Stable Tape will not turn them into one after the fact.

What changed

The evidence promise made on 14 August held. What 15 August exposed was the step after it: a day can preserve everything it needs and still end without a reading, and nothing said so plainly at the time.

A day now ends in one of two published states, a reading or an explicit no-reading outcome, and the public record is written either way rather than only when a reading succeeds.

What happened

No numerical Daily Tape edition was published on 14 August 2026.

The checks introduced after the previous day's withdrawal could not establish that the sources being read were the same ones the series had been reading. Without that continuity, Stable Tape could not publish the day's reading with enough confidence, so it published none.

The detailed evidence from that run was not durably kept. Stable Tape cannot now show exactly what each source returned at the time, and so will not reconstruct or backfill a numerical 14 August edition afterwards.

The missing evidence is not itself proof that anything further was wrong. It means the day cannot be settled after the fact, which is why it stays open rather than being filled in.

What it means for the record

There is no numerical Daily Tape edition for 14 August, and there will not be one later.

Stable Tape did not carry another day's value forward, substitute a later snapshot, estimate the daily movement, reconstruct the reading afterwards, or publish a Signal for the date.

The date remains NO READING in the record and in the archive.

What changed

From 15 August onward, every observation a daily reading depends on is preserved durably and checked before the reading can go ahead. Each day's sources are recorded before the fixing moment, at it, and after it, and no determination proceeds unless those observations are present and agree.

This is what was missing on 14 August: not the reading itself, but the durable evidence needed to stand behind it.

What happened

Stable Tape published a Daily Tape edition on 13 August 2026 and withdrew it after identifying an error in how one source value had been handled.

One of the supply figures had already been adjusted for market value by the source it came from. Stable Tape applied a market-value adjustment to that figure a second time. The double adjustment inflated the holding, and with it the stablecoin total for the day.

The error was in Stable Tape's own processing of the source, not in the underlying holding.

What it means for the record

The original 13 August edition is preserved. It remains published at its own dated address, with its original figures unchanged, as part of the historical record.

It should not be relied upon as a valid Daily Tape reading. It is marked READING WITHDRAWN wherever it appears and is excluded from the benchmark series.

No assessment, Signal, selection, share card or social post is published for a withdrawn edition.

What changed

Stable Tape strengthened the checks that sit between reading a source and publishing a reading, so that a value already adjusted upstream is not adjusted again, and so that an implausible total stops publication rather than reaching the page.

These checks close a known way of getting this wrong. They do not prevent every error.