Skip to content
Protocol STAMP
Latest Calibration

Publications

The CalibrationOur own measurements, pre-registered and signedThe LogWhat ran, what it cost, what movedThe BenchWhat we read and argue with · Mon and ThuArchiveEvery issue, by number and dateThe technical reportThe program in one paper, with its dataSubscribeOne email per issue

Method

MethodologyHow a rate is producedDefinitionsWhat each term means hereThe protocol, v1.25The instruction layer as measuredRun filesEvery draw we can publishfigures.jsonEvery citable number and its fingerprints

AI policy

The DocketRules that bear on how AI is testedThe ChronicleDated events in AI governanceSuper Intelligence (SI)The US rename of AI, tracedMilitaries and AICountry by country

Trust

The Ten House ProtocolsHow we keep AI from making things upIndependenceWho pays, and the rules on itCorrections + changesWhat we got wrong, datedAI Safety + GovernanceWhat our agents may and may not doSupportHow the program is paid forFootprintWhat our tokens cost in power, and who pays itPrivacyWhat we keep, and for how longTermsThe terms of use

About

About Protocol STAMPWhat this is, who makes it, how to check itColophonHow the site is made, and the seven finishesOther STAMPsUnrelated things with the same nameLatest Calibration

Protocol STAMP is teamed with AI and can make mistakes. Verify before you implement anything. What we got wrong · Free to read, paid for by readers: Contribute

The Log · issued document

The Log Vol 002 · Thu20261001

Issued Oct 01, 2026 19:28 from Houston · fills the missed Sep 19, 2026 slot · build a85771944770
PDFSourceExact copyArchive
Oct 01, 2026issued at 19:28 America/Chicago, from Houston
a85771944770build: the first 12 characters of the source’s SHA-256

Below is the issue exactly as it was issued; nothing in it has been edited. This frame, the menu and the footer are the site’s, not the issue’s. To check it, download the Source and hash it: the first twelve characters of its SHA-256 are the build above. The Exact copy in the public repository is the same bytes.

The Log Vol 002 · Thu20261001
The STAMP Protocol · The Log
The Log Vol 002 · Thu20261001 Jayden Lee, AI Reporter
A derived record. Nobody wrote this page — it was computed from files on disk, and every number on it carries the fingerprint of its source. It reports what ran and what moved. It does not say what any of it means; that is what a Calibration is for.

Standing position

candidatefailure rateitemsdrawsnoise floorreliabilityICCdesign effect
luna_v122.7%
[21.0%, 24.3%]
636260.9%0.9410.7387.6×
nemo_v143.3%
[41.2%, 45.5%]
636161.6%0.9450.6957.1×
nemo_v1_t042.6%
[39.8%, 45.4%]
63122——0.8981.8×

Each rate is measured on bank v1 across both halves, under the grader and provider recorded with it in figures.json. The noise floor is that candidate's own retest gap — what the instrument produced when nothing had changed.

Under what conditions

candidategradertempdrawsmax tokprovider
anchor_olmobytedance-seed/seed-2-1-turbo——32768unpinned
luna_v1bytedance-seed/seed-2-1-turbo0.7532768unpinned
nemo_v1
config reconstructed after the fact — asserted, not recorded
bytedance-seed/seed-2-1-turbo———unpinned
nemo_v1_t0bytedance-seed/seed-2-1-turbo0.0116384DeepInfra

All on bank_v1.jsonl. A rate is not a property of a model — it is a property of a model measured a particular way, so the conditions travel with it. A dash is a value that was never recorded, not a default.

What was lost

runusablelostwhyserved by
anchor_qwen A251/31520.3%1 unparsed, 63 truncatedSiliconFlow 185, Venice 29, Parasail 7, DeepInfra 25, Together 5
luna_v1 A315/3150.0%noneOpenAI 315
luna_v1 B311/3151.3%4 unparsedOpenAI 311
nemo_v1 A308/3152.2%6 unparsed, 1 truncatedDeepInfra 308
nemo_v1 B308/3152.2%5 unparsed, 2 truncatedDeepInfra 308
nemo_v1_t0 A61/633.2%2 unparsedDeepInfra 61
nemo_v1_t0 B61/633.2%2 unparsedDeepInfra 61
luna_v1: 0.0% against 1.3% between halves. Attrition is reported per half rather than pooled because pooling hides the case that matters: draws vanishing more often in one half than the other. Whether that difference means anything is a question for a Calibration, not for this page.

What moved

Nothing cleared its noise floor since 2026-09-05. Every rate above is unchanged within the range this instrument produces when nothing has changed.

That is a complete entry, not a missing one. There is no red on this page, and that is the result.

What it cost

candidatemodelgradergrader ÷ modeldraws costed
anchor_qwen$0.0000$0.0000—0/315
luna$0.0000$0.0000—0/44
luna_v1$0.0000$0.0000—0/838
luna_v1a$0.0000$0.0000—0/105
nemo_v1$0.0000$0.0000—0/911
nemo_v1_t0$0.0000$0.0000—0/126
nemotron$0.0000$0.0000—0/105
nemotron_pinned$0.0000$0.0000—0/111
nemotron_stamped$0.0000$0.0000—0/105
seed$0.0000$0.0000—0/13
total$0.0000$0.0000

Charged cost as OpenRouter reported it per call, never tokens times a price. Model and grader are kept apart because in measurement one the grader cost multiples of the models it was grading.

2,673 draws carry no recorded cost. Spend capture was added to the runner after those draws were taken, so the totals above cover only the instrumented ones. They are a floor on what was spent, not the whole of it — stated here rather than folded silently into a number that would look complete.

What ran

jobdatewall clockstatedetail
panel_nemo2026-08-2932 mindonepanel report
nemo_t0_A2026-08-2949 minpartial61/63 draws
nemo_t0_B2026-08-2939 minpartial61/63 draws
fever_nemo2026-08-29—readyneeds both temperatures
luna_A2026-08-29—partial311/315 draws
luna_A2026-08-293 mindone315/315 draws
luna_B2026-08-29222 minpartial311/315 draws

Wall clock as the queue recorded it. A job listed twice was resumed; the later row supersedes the earlier one.

Source of this page, byte-exact: 002-issued-thu20261001/source · sha256 a85771944770… · 12,831 bytes · PDF

Issued from Houston, Texas, USA on Oct 01, 2026 at 19:28 America/Chicago. This issue fills the slot of Sep 19, 2026, which was missed, and it is not a Saturday issue: its name reads THU so the departure can be seen without opening it. The next Log is due on Oct 03, 2026. The time is the moment the release ran, not the moment it was due. The dateline names where the work was done; it moves with the investigator, and each issue keeps its own.

To cite: The STAMP Protocol. The Log Vol 002 · Thu20261001. Budday Budderson Studio LLC, 2026. https://protocolstamp.com/log/002-issued-thu20261001/ · build a85771944770. See Definitions for the form.

The STAMP Protocol can make mistakes. Verify before you implement anything. An issued document is never edited, so a correction to this one is issued under a new date and this page stands as it was. See Corrections + changes.