Documentation
Q-space · public documentation · snapshot 2026-11-17
Q-space is a formal knowledge space for mathematics and an autonomous research system that grows it: the system surveys existing formalized mathematics, identifies what is missing, formulates candidate statements, searches for proofs, rejects false candidates by counterexample, and registers every surviving result with strict machine verification. $$\text{Q-Space} = \bigl(\text{Knowledge Graph}\bigr) + \bigl(\text{Proof Search}\bigr) + \bigl(\text{Strict Verification}\bigr)$$
What the Discovery Stream shows
The stream is a read-only window into the result registry — the single record of everything the system has produced. Each entry carries:
- the exact formal statement;
- a proof excerpt in Lean 4 (abridged for display; full sources are kept in the internal registry);
- the verification record: prover version, library revision, permitted axioms, timestamp, artifact reference;
- an assessment block: formal verification status, novelty, significance — evaluated separately and honestly;
- a research trace: the recorded steps from target identification to registration;
- dependencies: statements and anchor theorems the result builds on.
Failed research is recorded too: refuted candidates appear as FALSIFIED entries with their counterexamples, and open targets appear as HYPOTHESIS entries with what is known about their boundaries.
Verification policy
| Term | Meaning |
|---|---|
| STRICT VERIFIED | The proof compiles in Lean 4 and passes an independent strict check: the set of axioms the proof depends on is extracted and compared against a fixed whitelist — $\text{propext}$, $\text{Classical.choice}$, $\text{Quot.sound}$. $$\text{axioms}(P) \subseteq \{\text{propext},\; \text{Classical.choice},\; \text{Quot.sound}\}$$ Anything outside the whitelist rejects the proof. $\text{sorry}$, $\text{admit}$ and analogous placeholders are rejected. |
| COMPILE ONLY | The proof compiles, but the strict axiom-level check has not yet classified it. These are mostly small auxiliary lemmas. They are never counted as verified results. |
| FALSIFIED | The candidate statement was disproved by an exhibited counterexample, found by automated search and machine-checked. |
| HYPOTHESIS | An open target: a statement the system aims to prove, with the current knowledge of its boundary of validity recorded. |
| IN PROGRESS | An active attempt: statement accepted, proof search under way. |
What we do not claim
Honesty about the limits of formal verification is part of the design.
- Correctness ≠ novelty. A strictly verified theorem is formally correct relative to its statement. Whether the statement is new is a separate question; most entries are marked NOVELTY UNASSESSED until a literature check is performed.
- Correctness ≠ significance. Many current entries are elementary facts and bridging lemmas. They are recorded because they are true and verified, not because they are deep.
- Statement fidelity. Verification confirms the formal statement — that the formal statement faithfully captures an intended mathematical question is assessed separately.
How results originate
Every entry records its origin. Current sources of targets include:
- corpus analysis — systematic comparison of what is proved across mathematical domains, identifying absent counterparts and structural gaps;
- concept demand — when many targets require a notion that does not exist, the concept $C$ itself is introduced: $\quad \text{demand}(C) \geq \theta \;\Rightarrow\; \text{introduce}(C)$, and registered (see the Int.Coprime entry of 2026-11-17), after which a family of statements about it becomes reachable;
- prior results — registered theorems serve as components for new ones.
Counterexample search runs on candidates before and during proof attempts; a refuted candidate is registered as a negative result rather than discarded.
Registry state
| Metric | Value (2026-11-17) |
|---|---|
| Total registered | 259 |
| Strictly verified | 224 |
| Compile-only (auxiliary) | 35 |
| Domains represented | bitwise algebra, div/mod, gcd/lcm, order, powers, coprimality (new) |
The stream shows a curated subset; the registry grows with every research run.
Access & citation
- This page is a static snapshot, generated from the registry; a live read-only endpoint is planned.
- Each entry references a stable artifact identifier of the form qsr://registry/<id>.
- Licensing and citation format: to be announced.
Frequently asked
Why are there “compile only” entries at all?
Transitional honesty: auxiliary lemmas are recorded when they compile, and upgraded to strict status after the independent axiom check. Mixing verified and unchecked results under one label would be worse than showing the difference.
Why publish refuted statements?
A verified counterexample is knowledge: it maps the boundary of what is true. Negative results also protect future work from re-attempting dead ends.
Is this an AI system?
Q-space is an automated research system: it operates autonomously — surveying, formulating, searching, checking — without a language model in its proof loop. Trust in a result never rests on who or what produced the proof: it rests on the strict verification record attached to it.
What is the long-term goal?
A self-growing formal knowledge space: results compound into new results, demanded concepts get introduced, and every fact carries a machine-checkable certificate and full provenance.