Measured ledgers · append only run history
Every ledger, and every time it was published
A measurement published once is a snapshot wearing the clothes of a standing claim. The only
way to tell the difference is to point the same probe at the same population again and show
both answers. This page is that record: one row per publication, never edited, with the
digest that served each one.
This index measures nothing itself · it cannot make a refusal disappear · a ledger that is
wrong is wrong here too, twice, with both digests shown
5 ledgers · 11 publications on record
Index digest 556af54e3c984ee90108eea67a70fb11c1ff2145318793186c9e7b451f3c08b5 generated 2026-08-02 06:09 UTC
Built by ledger_refresh v1.0.0 · raw rows at runs.jsonl on both origins
11
across 5 ledgers
publications on record
Every row is copied from a summary.json that was actually served, and named by the digest that served it.
5
of 5 ledgers
have been re-derived at least once
A ledger with one run is a snapshot. It is labelled as one below rather than left to read as a standing claim.
4
this refresh
headline counters moved between runs
Every one of them on the sampled ledger, and 2 of the rate comparisons behind them still have overlapping intervals. A move is not automatically a finding.
4
of 5 ledgers
held every headline counter
The digests still differ, because timestamps and per row detail differ. Identical counters are not an identical file.
What this index is not
Not a measurement of its own. This index does not fetch any subject, does not score any ledger and cannot make a refusal disappear. If a ledger is wrong, it is wrong here too, twice, with both digests shown.
A row is appended when a ledger publishes a digest this file has not recorded before. Prior rows are never edited or removed, including when a later run makes an earlier one look bad.
A change in a headline counter between two runs is a change in what the world served, in what the probe asked, or in the probe itself. The probe version sits on every row so the third case can be ruled out before the first two are argued.
A counter can move while the published 95 percent intervals of the two runs still overlap. That is a move this measurement cannot distinguish from sampling noise, and it is counted separately here so a re-run is not mistaken for a finding.
Ledgers
ERC-8004 Base refusal ledger
/erc8004/ · one row is one sampled registration · 2 publication
s on record
| Run | Published | Probe | Digest | Headline counters |
| 1 | 2026-08-02 03:11 UTC | v1.0.0 | ab79fc3924f1 | full admit 45, full refuse 455, must admit 135, must refuse 365 |
| 2 | 2026-08-02 05:48 UTC | v1.0.0 | f01dd0580950 | full admit 56, full refuse 444, must admit 143, must refuse 357 |
Comparing two runs of this ledger: Two runs of this ledger are two samples, not two looks at the same set. Each run pins a fresh Base block and draws 500 agent ids seeded by that block hash, from a registry that keeps growing, so the rows compared here overlap only by chance. A count that moves by single digits is sampling noise before it is news. The published rate and its 95 percent interval are the comparison that means something, and they are shown below the counts.
Moved on the latest refresh: must admit 135 to 143 (+8) · must refuse 365 to 357 (-8) · full admit 45 to 56 (+11) · full refuse 455 to 444 (-11)
must miss rate 0.73 [0.689441, 0.767052] then 0.714 [0.672876, 0.751861] , and the two 95 percent intervals overlap, so this run does not establish that the underlying rate moved at all
full miss rate 0.91 [0.88169, 0.932058] then 0.888 [0.857346, 0.912738] , and the two 95 percent intervals overlap, so this run does not establish that the underlying rate moved at all
PYTHONPATH=engine python3 engine/tools/erc8004_probe_run.py --sample 500
SLSA and PEP 740 census of this node's own dependencies
/slsa/ · one row is one resolved distribution · 2 publication
s on record
| Run | Published | Probe | Digest | Headline counters |
| 1 | 2026-08-02 03:44 UTC | v1.0.0 | 1fa8727b43a7 | full admit 0, full refuse 94, must admit 44, must refuse 50 |
| 2 | 2026-08-02 05:44 UTC | v1.0.0 | 0c23b0305508 | full admit 0, full refuse 94, must admit 44, must refuse 50 |
Comparing two runs of this ledger: A census, not a sample: both runs cover every distribution pip resolves for this node's requirements file. A count that moves is a real change in what PyPI served or in what the requirements resolve to, and neither is noise.
the headline counters did not move between the last two runs. The digests still differ because timestamps and per row detail differ.
must refuse rate 0.5319 then 0.5319 , published without an interval
full refuse rate 1 then 1 , published without an interval
/tmp/slsa_venv/bin/python engine/tools/slsa_provenance_run.py run
Real world asset disclosure interface survey
/rwa/ · one row is one issuer · 2 publication
s on record
| Run | Published | Probe | Digest | Headline counters |
| 1 | 2026-08-02 04:34 UTC | v1.0.2 | ce9c3eb42c04 | full admit 0, full refuse 17, must admit 2, must refuse 15 |
| 2 | 2026-08-02 05:43 UTC | v1.0.2 | e4d5d4e38e70 | full admit 0, full refuse 17, must admit 2, must refuse 15 |
Comparing two runs of this ledger: The population is frozen in the expectations file and identical across runs, so a count that moves is an issuer that changed what it serves. Still not a sector rate: the population is what this node already tracks.
the headline counters did not move between the last two runs. The digests still differ because timestamps and per row detail differ.
must refuse rate 0.8824 then 0.8824 , published without an interval
PYTHONPATH=engine python3 engine/tools/rwa_disclosure_run.py run
EDGAR and Federal Reserve grounded claim corpus
/grounded-claims/ · one row is one claim · 2 publication
s on record
| Run | Published | Probe | Digest | Headline counters |
| 1 | 2026-08-02 05:20 UTC | v1.0.1 | 20b2faf73e04 | full admit 0, full refuse 6, must admit 5, must refuse 1 |
| 2 | 2026-08-02 05:42 UTC | v1.0.1 | 5df1c68a86bd | full admit 0, full refuse 6, must admit 5, must refuse 1 |
Comparing two runs of this ledger: The corpus is frozen in the expectations file and identical across runs, so a count that moves is a regulator page that changed under a citation this node already made.
the headline counters did not move between the last two runs. The digests still differ because timestamps and per row detail differ.
PYTHONPATH=engine python3 engine/tools/grounded_claim_run.py run
Transparency log batch monitor
/tlog/ · one row is one log · 3 publication
s on record
| Run | Published | Probe | Digest | Headline counters |
| 1 | 2026-08-02 05:34 UTC | v1.0.0 | ed5ebec4afeb | full admit 22, must admit 22, must refuse 1 |
| 2 | 2026-08-02 05:43 UTC | v1.0.0 | 2fac8d589132 | full admit 22, must admit 22, must refuse 1 |
| 3 | 2026-08-02 05:43 UTC | v1.0.0 | 1e9ae859e742 | full admit 22, must admit 22, must refuse 1 |
Comparing two runs of this ledger: The same 23 logs every batch, so counts are directly comparable. What is not comparable is the cold start: the first batch held no prior checkpoint for any log and every batch after it does, which changes what some limbs can even reach.
the headline counters did not move between the last two runs. The digests still differ because timestamps and per row detail differ.
PYTHONPATH=engine python3 engine/tools/tlog_monitor_run.py run
Machine readable