Docs · Science
Methodology and evidence
Evidence kinds, response statuses, computation methods and the rules every number follows.
Principles
- Every number carries an evidence kind. A theorem value, an SDP output, a Monte Carlo estimate and a photon measurement are never interchangeable and never shown without their label.
- Monte Carlo is a lower bound on worst-case attack success, always with its interval. It is never presented as an upper bound or a proof.
- Numerical SDP is not a proof unless residuals are rigorously accounted (that is
CERTIFIED_NUMERIC, re-verified in exact arithmetic by lane D). - Simulation is not physical security. No QPU result exists in this build (gate G6 is skipped).
- Unknown is null, never 0. An unknown bound is
null/UNCERTIFIEDand renders as text. - No theorem crosses games. W1 ≠ W1_CV ≠ W2 ≠ S1 ≠ S2 ≠ D1 ≠ A1. ε is a maximum adversary win probability, never an error bar.
- Sources are cited to an anchor (URL plus theorem, equation or table). If a primary source could not be read, the entry says so instead of reconstructing it from memory.
Evidence kinds
The glyph shape encodes origin — theory ■, numeric ◩, simulated □, measured ●, none ×, API ◇ — and its colour encodes tone. The table is generated from qmint/ui/evidence_badges.py via pnpm gen:api, the same source the API uses.
| Label | Key | Meaning |
|---|---|---|
| CERTIFIED-NUMERIC: Numeric bound with rigorous residual accounting | CERTIFIED_NUMERIC | Numeric bound with rigorous residual accounting |
| DIGITIZED-FIGURE: Read off a published figure; carries digitization error | DIGITIZED_FIGURE | Read off a published figure; carries digitization error |
| MEASURED-LATENCY: Latency from real measurements | MEASURED_LATENCY | Latency from real measurements |
| MODELLED-LATENCY: Latency from a model (light-cone, queue model); not a measurement | MODELLED_LATENCY | Latency from a model (light-cone, queue model); not a measurement |
| MONTE-CARLO (lower bound): Sampled attack success: a LOWER bound on worst-case attack success, never an upper bound | MONTE_CARLO | Sampled attack success: a LOWER bound on worst-case attack success, never an upper bound |
| NUMERICAL: Floating-point SDP solver output; not a proof without residual accounting | NUMERICAL_SDP | Floating-point SDP solver output; not a proof without residual accounting |
| PHOTON-EXPERIMENT: Real recorded photon events with dataset hash | PHOTON_EXPERIMENT | Real recorded photon events with dataset hash |
| PUBLISHED-AGGREGATE: Number quoted from a paper table or text | PUBLISHED_AGGREGATE | Number quoted from a paper table or text |
| THEOREM: Bound transcribed from a peer-reviewed proof | PUBLISHED_THEOREM | Bound transcribed from a peer-reviewed proof |
| QPU: Actual quantum hardware job with job id | QPU | Actual quantum hardware job with job id |
| SIMULATOR-IDEAL: Noiseless circuit/statevector simulation | SIMULATOR_IDEAL | Noiseless circuit/statevector simulation |
| SIMULATOR-NOISY: Simulation under a declared noise model | SIMULATOR_NOISY | Simulation under a declared noise model |
| SYNTHETIC: Generated fixture or hypothetical scenario | SYNTHETIC | Generated fixture or hypothetical scenario |
| UNCERTIFIED: No applicable evidence | UNCERTIFIED | No applicable evidence |
Certificate statuses
Only CERTIFIED is shown in mint, and only when the frozen C5 contract allows it: assumptions met, proof reference, checker id, evidence hashes and every target bounded.
| Label | Key | Meaning |
|---|---|---|
| CERTIFIED: Every target met by an applicable, independently checked bound | CERTIFIED | Every target met by an applicable, independently checked bound |
| INFEASIBLE: Applicable bounds exist but targets cannot be met | INFEASIBLE | Applicable bounds exist but targets cannot be met |
| OUT-OF-MODEL: Request violates the proof's trust/physical model | OUT_OF_MODEL | Request violates the proof's trust/physical model |
| UNCERTIFIED: Missing bound, assumption or independent check | UNCERTIFIED | Missing bound, assumption or independent check |
| UNSAFE: An attack exceeds the claimed bound or targets | UNSAFE | An attack exceeds the claimed bound or targets |
Response statuses
Every API response has a status. OK means a merged lane's code path produced it. The other two always carry a reason:
| Label | Key | Meaning |
|---|---|---|
| DEMO ADAPTER: Demonstration computed from analytic / published formulas or merged lane-F/G code while the owning lane is not merged; not that lane's output | DEMO_ADAPTER | Demonstration computed from analytic / published formulas or merged lane-F/G code while the owning lane is not merged; not that lane's output |
| NOT AVAILABLE: Nothing honest to show yet; see the reason | NOT_AVAILABLE | Nothing honest to show yet; see the reason |
Computation methods
Quantity.method disambiguates how a value was computed — for example an exact analytic value that the frozen contract labels SIMULATOR_IDEAL.
| Method | Meaning |
|---|---|
| closed_form | a transcribed or derived formula evaluated exactly |
| exact_enumeration | exact model value from enumerating outcomes or Kraus operators |
| aer_sampling | shot-sampled local Aer simulation; carries a confidence interval |
| event_simulation | lane G discrete-event timing simulation |
| published_value | a number quoted from a paper |
| measured_dataset | a statistic of a committed real measurement dataset |
| not_available | no value; shown as text, never as 0 |
Research verification statuses
Research-library entries keep the verification status from their own file — VERIFIED_FROM_SOURCE and NOT_VERIFIED per transcribed theorem field, FULL_TEXT_READ / ABSTRACT_READ / METADATA_ONLY per paper, proof statuses such as PUBLISHED_REPRODUCED per obligation. Definitions and their source files are listed under “Verification status definitions” in the research library.