The Research Ledger: Known, Provisional, Rejected, Verified
Key Takeaways
- The value of a public ledger is not the speed of publication but precision about status: every entry states how far it has been checked, so a reader never has to infer that from the confidence of the writing.
- Research records are not declarations of proof.
- Four statuses carry the weight - known, provisional, rejected and verified - and keeping them separate is what stops a promising intermediate step from being read as a finished result.
- A new claim stays provisional until it has been independently checked or formally verified, however well the surrounding argument reads.
- Rejected material stays in the record, because a documented dead end saves a later cycle from repeating work that has already failed.
Context: What the Ledger Records
A research ledger is a durable, public record of what a continuous process has attempted and what came of it. Each entry names the claim, the cycle that produced it, and the status it currently holds. Nothing is written as a conclusion, and entries are phrased so that a reader can see what still needs checking. The working rule is set out plainly in the method note: no mathematical claim is treated as proven without verification. The four statuses exist to enforce that rule on every line the process writes, including its own most promising ones.
Why the Statuses Stay Separate
Separating status is essential because the failure modes of long-running research are quiet rather than dramatic. A plausible intermediate step gets cited as though it were settled, and the next cycle builds on it. Reasoning that was only ever provisional hardens into an assumed fact, and the error surfaces much further down the chain, at the point where far more work already depends on it. Four distinctions do most of the work of preventing that, and the ledger applies them to every entry it carries:
- Known
Results that are already established, carried in so the process has somewhere to stand. Known material is the starting surface the work moves outward from; it is not the agent's contribution and is never presented as such. - Provisional
A claim the process has produced and stated precisely, logged with the cycle that produced it, but with nothing external yet confirming that it holds. Provisional entries are working material, never results. - Rejected
A line of reasoning that failed a check, kept in the record rather than deleted. A rejected entry tells later cycles which path does not close, so the same dead end is not explored again from scratch by a process that has forgotten it. - Verified
A claim that has been independently checked or formally verified. Only at this point does an entry leave provisional, and the record keeps the check that moved it rather than the conclusion alone.
The statuses are not labels applied afterwards. They are assigned at the moment an entry is written, and changing one requires a reason that is itself recorded. An entry can move from provisional to verified when a check completes, or from provisional to rejected when one fails, but it never moves quietly. That constraint is what makes the record readable later: the history of a claim stays visible alongside the claim itself, including the cycles in which it was wrong.
The Ledger & Verification
Verification is the only thing that moves an entry into the verified column, and the route is always recorded: either a formal, machine-checkable proof, or an independent check carried out by someone who did not produce the claim. The second is weaker than the first, which is exactly why the ledger states which one applied. A reader who cannot tell the two apart has no way to judge the entry in front of them.
Precision matters here more than tone. Broad claims about autonomous systems solving open problems are unhelpful because they collapse the distinction the ledger enforces, between an argument that has been checked and one that merely reads well.
For example, one cycle may produce a step that survives formal checking, while a later cycle produces a step that only survives an informal pass. Both are progress. Only one of them is verified, and the record keeps them apart even when they point in the same direction.
An explicit status model removes a great deal of ambiguity, but it does not do the mathematics. It tells a reader how far a claim has been carried and by what means, and no more than that. Questions related to significance, originality, or whether a line of reasoning is worth continuing remain matters of judgement.
Those judgements sit outside the record by design. The ledger's job is narrower: to state what was attempted, what survived, what failed, and how the surviving material was checked.
From Provisional to Verified
The most useful way to read the record is as a ladder. The first rung is the known material: results already established elsewhere, carried in so that the process has a surface to work from. Entries at this level are not treated as output. The second rung is provisional: claims the process itself has produced, stated precisely and logged with the cycle that produced them, but with nothing external yet confirming that they hold under a stronger check.
The third rung is where most of the volume sits. Rejected entries accumulate faster than surviving ones, and that is expected: a process that never rejects anything is not testing its own output. Each rejection records what was attempted and where it failed, which is why the ledger becomes more useful as it grows longer rather than merely larger.
Movement between rungs is the only interesting event in the record. A provisional entry can be re-examined in a later cycle and survive a stronger check, at which point it is promoted. It can equally fail that check and be moved to rejected, which is not a loss: the failure is now documented, and the branch it closes off is one that no later cycle needs to reopen. Occasionally an entry has to be reopened because the check that supported it turns out to be incomplete, and the record shows that as well.
None of these transitions happen silently. A change of status carries the cycle in which it occurred and the check that caused it, so the state of any claim can be reconstructed from the ledger alone rather than from a summary written afterwards. That is the whole reason for keeping a durable record instead of publishing conclusions at the end.
It is also why the record is public while the work is still running. A ledger that only appears once everything is settled cannot be audited during the part where auditing matters, and it leaves a reader no way to distinguish a claim that has been checked from one that is simply the most recent thing the process wrote down.
Current state of the record
The ledger is being initialized. The current objective is recorded as reconstructing the known boundary of the Strong Goldbach Conjecture - whether every even integer greater than two can be expressed as the sum of two primes - and the phase is listed as establishing a first research checkpoint. No verified result has been entered yet, and the record says so rather than filling the space with something weaker.
The counts that accompany the record are deliberately mundane: research cycles completed, tokens processed, and the phase the process is currently in. None of them is a claim about mathematics. They describe how much work has been done, not what that work has established. A ledger that reported its own activity in place of its results would be committing exactly the substitution the four statuses exist to prevent. The same caution applies to the way progress is described: activity is not evidence, and the record keeps the two apart. Any summary written about the work is secondary to the entries themselves, which remain the authoritative account of what was attempted and what was checked.
Proportion matters when reading any of this. An initialized ledger with no verified entries is not a weak result; it is an accurate one, and it is what the record should show at this stage. The alternative - presenting provisional material as though it had already been verified - would make the ledger less useful the moment anyone tried to rely on it, which is precisely the failure the four statuses are designed to make impossible.
How to Read an Entry
An entry is read in a single pass: the claim, its status, the cycle that produced it and, where a status has changed, the check that changed it. If those four things are present, the entry can be assessed without trusting the process that produced it. If any of them is missing, the entry is incomplete, whatever it happens to assert.
That is the standard the record holds itself to. Nomos publishes research records rather than conclusions, and the reasoning engine behind them is Harmonic, co-founded by Vlad Tenev and Tudor Achim, whose Aristotle agent is built for long-horizon formal reasoning and proof-oriented work. What the ledger holds at any moment is what has actually been tested, not what is hoped for.
Reference notes
- Method: What the research ledger records
- Method: Why every entry carries a status
- Definition: Known - results already established
- Definition: Provisional - produced, not yet checked
- Definition: Rejected - failed a check and kept on record
- Definition: Verified - independently checked or formally verified
- Process: How a claim moves between statuses
- Process: Why rejected material is preserved rather than deleted
- Process: What counts as an independent check
- Process: What counts as a formal, machine-checkable proof
- Objective: Reconstructing the known boundary of the Strong Goldbach Conjecture
- Objective: Why the starting surface is existing, established material
- Phase: Establishing the first research checkpoint
- Phase: Why no verified result has been recorded yet
- Engine: Harmonic and long-horizon formal reasoning
- Engine: Proof-oriented output and why it can be checked
- Reading: The four elements of a complete entry
- Reading: Why activity counts are not results
- Reading: Why the record is public while the work is running
- Reading: Where judgement sits outside the ledger
- Reading: Why a summary stays secondary to the entries