20 claims, each with the documents that will settle it written down before the deadline. How the criteria may change, who adjudicates, where disputes are logged, and the letter we have not sent.
The short version
A verdict is decided from named documents against criteria fixed in advance. Those criteria can be clarified in public, by adding definitions, while a claim is still open and before its deadline, and each clarification keeps the old text on the page and in the integrity log. Anything that would change a stated condition is not a clarification: it removes the claim from scoring and says so. No adjudicator has been appointed, so the desk decides alone, and every verdict page says that.
The documents that settle each claim
Each claim’s criteria say what VERIFIED, PARTIALLY_VERIFIED and NOT_VERIFIED require, and name the sources the verdict will be written from. A source is a link to a document, so anyone can open and archive it. Where a source is a class of documents rather than one link, a written reason is on file (resolver-reviewed.json). A build check fails if a claim’s criteria lack an outcome or name a source that is neither a link nor reviewed.
Two kinds of change are allowed, both only while a claim is open and before its deadline:
A clarification adds definitions to terms the criteria left open. Every word of the earlier criteria stays, and wins if a definition seems to conflict with it.
A locator adds a link to a document the criteria already named only by citation. The criteria text is untouched.
Each one is entered in the integrity log as an amendment with the old and new hashes and the reason, and the earlier text is kept on the claim page under “How the criteria changed”. A build check recomputes the hash of the kept text and fails if it does not match what the log recorded. Anything that changes a stated condition, such as an outcome, a threshold or a deadline, is moving the goalposts. The Ledger’s rule is that this removes the claim from scoring and leaves a public note, and a change of that kind is not made here.
Amendments so far: 3.
D-Wave Quantum Inc., 19 October 2026 (locator). Two resolver sources were named by citation, not by link (arXiv:2503.05693 and an EPFL preprint). Each now has its arXiv link, so the documents that settle the claim can be found and archived. Adds a link to each resolver source that was named only by citation. The same two documents are meant; no condition is changed.
IBM, 19 October 2026 (clarification). The criteria used undefined terms (an independent party, a standing reproduction, an open dispute) and named a resolver source, the Quantum Advantage Tracker, with no link, and the Tracker is run partly by IBM. Definitions and the link were added so two readers would score the same evidence the same way. Written after the 30 July 2026 IBM and Qedma announcement was logged, and without scoring it: it is not confirmation under the definitions either, since IBM and Qedma are the authors. Adds definitions only. Every sentence of the criteria as first written is kept, and takes precedence if a definition seems to conflict with it. No outcome that the original text decided clearly is changed.
QuNorth, 19 October 2026 (clarification). The criteria used undefined terms (installed, first workloads, independent reporting). Definitions were added so two readers would score the same evidence the same way. Written after the 2 September 2026 report that the building was complete and installation was only beginning was logged, and without scoring it: under the definitions, as before, that report is not an installed machine. Adds definitions only. Every sentence of the criteria as first written is kept, and takes precedence if a definition seems to conflict with it. No outcome that the original text decided clearly is changed.
These were written by the editor, one person, and each states what the editor knew when it was written. A clarification written after evidence exists can still, without anyone meaning it, favour one reading, so the reason for each clarification says what had already been logged when it was written. You can see on the claim page exactly what was added.
Who adjudicates
Every claim has an adjudicator field. It is empty (null) when nobody has been appointed, and when it is filled it must carry a name, a role, an appointment date and declared conflicts. 0 of 20 claims have an adjudicator. Until one is appointed, the editor decides alone, and each claim page says so. Named reviewers will appear on the reviewers page when there are real ones; none is made up here.
Conflict log
Disputes about a claim’s criteria or evidence, objections from a claimant, and conflicts of interest are logged in ledger-conflicts.json, opened when raised and closed with an outcome. People are named by role only.
None yet. The log began on 19 October 2026. Nobody has raised a dispute, an objection or a conflict of interest, as far as the desk has been told.
Evidence snapshots: how a verdict is reproduced
Every source a verdict cites is kept three ways. An independent archive copy (the Wayback Machine), whose date is shown next to the link on the claim page and said plainly when it was taken after the verdict. Our own fetch, with the HTTP result and a SHA-256 of the exact bytes. And a record, in data/evidence/, of who drafted the verdict and who reviewed it. The bundle’s hash is chained in the integrity log, so it cannot be swapped quietly. The fetched pages themselves are kept in the private desk repository, not published here, so this site holds hashes and archive links and not other people’s pages.
drafted by agent:claude-opus-5; reviewed by no one
What this does not give you. The verdicts above were published on 22 September 2026, and our own snapshots were taken later, on 19 October 2026, so they are labelled retroactive: they show what the sources say now, not what the drafter saw, and 1 of the 8 archive copies were taken after the verdict. 1 source(s) refused or failed our fetch; each failure is recorded in the bundle, not hidden. The prompt and transcript of the drafting sessions were not kept, so the drafting cannot be replayed; what can be repeated is the check of each verdict against its sources. From 19 October 2026 a verdict cannot be published without a snapshot taken on or before the day it is published: a build check fails otherwise.
Telling the claimant
The Ledger’s standing rule is that any claimant can answer a verdict at any time. This is the letter the desk would send before a claim’s deadline if it chose to. It is a template. Nothing has been sent to anyone: the date a claimant was told is recorded on a claim page only when it is set, and it is set on 0 of 20 claims.
Subject: SymbiQ Ledger: a claim of yours is scored on {resolves_by}
To the {claimant} press or investor-relations team,
On {source_date}, {claimant} said:
"{verbatim}"
SymbiQ (https://symbiq.dev) is an independent, unfunded teaching site. Its Ledger tracks dated, public claims and scores each one on its deadline. Yours is here:
{claim_url}
It is scored on {resolves_by}, with a {grace_days}-day grace period, against criteria written down on {criteria_frozen_at}. They are on that page, with the documents we will use to settle it. {amendment_note}
What happens next:
1. After the deadline we write a verdict from those documents only, and publish it with an archived copy of every source it cites.
2. You may send us, at any time, a document that bears on the criteria, or a reply. A reply is published in full beside the verdict. If it shows the verdict is wrong, the verdict changes and the change is logged in public.
3. Nobody pays for a claim to be tracked or for the wording of a verdict, and we do not show a verdict to a claimant before it is published.
Reply to the address on https://symbiq.dev/about, or use the correction form at https://symbiq.dev/corrections#report.
The desk, SymbiQ