Skip to content

Releases

Leanbook 1.0.0 — verified library and Quantum v4.20

Choose a tag to compare

@solanaqubits solanaqubits released this 06 Nov 10:25

Leanbook 1.0.0 consolidates the public library and the Quantum v4.5–v4.20 development. The Lake package version is now 1.0.0; its identifier, import paths, Lean 4.33.1 and pinned dependencies are preserved.

  • 377 catalog modules, 381 library Lean files, 331 direct MasterSuite imports.
  • 31 new quantum modules and 619 named theorem declarations, including 32 in the final mixed-environment / BSM refinement pair.
  • Fresh strict project build: 383 modules/roots, including both executables; shared dependency caches were mounted read-only.
  • Full live suite: 672 tests, no skips. Native store profile tests were enabled.
  • Full verifier and pinned independent audit: 42589 / 42589 declarations respectively (overlapping scopes, not additive). Only propext, Classical.choice, Quot.sound; no forbidden axioms.
  • Registry audit, knowledge links, verification scope and whitespace checks passed.
  • Fixed 15 quantum test launchers to inherit project compiler options; fixture formatting was corrected without weakening assertions. The initial full-suite failure and final rerun are recorded in the evidence.

Release scope and migration · New quantum theorem inventory

The attached source archive and verification receipts bind the checked bytes to commit 5276529c6c1010722cc96c39217892b9e874f6a8. SHA256SUMS covers the three other assets. Candidate hosted CI passed on this exact commit. A host restart interrupted a local test attempt; completed build/audit receipts were revalidated against unchanged sources and runtime, and the full test suite was restarted. The interruption and resume validation are included in the evidence.

Claims concern the actual Lean statements and hypotheses. This release does not certify physical devices, experimental constants or production readiness. The newest mixed-environment results assume an initially independent environment; arbitrary initial correlations and adaptive process tensors remain outside that model. Specification-conformance checks cover registered contracts, not every prose interpretation. Novelty is not assessed.

Quantum v4.20 — mixed environments and Bell instruments

Choose a tag to compare

@solanaqubits solanaqubits released this 05 Nov 20:19

This scoped component update publishes the quantum development series through v4.20.
It does not replace or claim a new whole-library beta checkpoint.

  • 31 quantum modules: four-phase Pauli matrix semantics, Born-derived Bell instruments, classical readout noise, gate-noise composition, retained environments and memory/reset counterexamples.
  • The final v4.20 step adds 32 named theorems for arbitrary fixed mixed independent environments, CPTP reduction, ensemble-independent semantics and branchwise BSM refinement.
  • Public-tree strict build and transitive axiom audit, 39 quantum/hierarchy regression tests, metadata and knowledge checks. The allowed axioms are propext, Classical.choice, Quot.sound.
  • Existing public leaf proofs and pinned dependencies are preserved. New-source license headers were synchronized with the private source without changing definitions or proof bodies.

Scope and proof boundaries · Machine-readable verification summary

The environment is initially independent of the system. General initial correlations,
adaptive interventions and process tensors remain separate obligations. This is a
mathematical formalization, not hardware calibration or experimental validation.

v0.10.0 — verified β checkpoint

Choose a tag to compare

@solanaqubits solanaqubits released this 26 Oct 20:13

A full verification checkpoint of rank β, with a separately qualified SimLab companion. Candidate, main and separate tag CI runs completed successfully for both exact SHAs. This is the new qualified checkpoint after v0.8.0. The failed v0.9.0 tag and its historical evidence remain unchanged.

Leanbook scope

  • Exact checkpoint: 99f90e813607452b0b441daa5689c355e37def7b.
  • 352 freshly compiled project modules, zero project module replay: 349 Verification modules, the library root, and two CLI entry points. Nine external package caches were preserved; jobs from eight packages were reused.
  • 642 local tests passed, no skips. Both axiom auditors checked 40,337 declarations, limited to propext, Classical.choice, and Quot.sound.
  • Strict configured Lean invocations, clean-project receipt validation and mandatory native SQLite preparation now form the release gate. The first failed qualification attempt is retained; the entire gate was rerun after the final fix.
  • 300 direct MasterSuite imports. The additional cached-normalization pilot is checked without replacing production ALCC normalization.

SimLab companion

  • Exact checkpoint: 4c52ddc5e64c0e3455208ea35fe55f4317ef47f5.
  • 152 unittest, 45 Quantum-core pytest, 46 metadata unittest and 4 Rust tests passed locally. Both SimLab auditors checked 2,357 declarations.
  • 16 native ALCC scenarios, 16 receipt audits (15 native plus one direct Python API invocation), 198 archived full outputs rechecked by the bounded Rust oracle, and six Quantum CLI/analytic/receipt scenarios passed.
  • Four live host-watchdog scenarios were run on this unchanged SimLab SHA during local qualification; their receipts were revalidated at closeout. Watchdog is optional and is not automatically integrated into production runners.

Evidence and boundaries

The attached public validation summary binds exact candidate SHAs, source hashes, the accepted local evidence digest, binary/PIN digests and hosted CI run URLs. Candidate and annotated-tag CI are separate runs. Hosted Leanbook tests intentionally exclude target-qualified SQLite tests; those tests passed in the full local gate. Hosted CI is not claimed to reproduce every local qualification environment.

Compiled execution is a runtime result, not a new Lean kernel proof per input. The Rust oracle is a bounded additional reference. Quantum results are floating-point simulations. Alternative normalization equivalence is proved, but production speedup and cross-machine reproducibility are not established. Receipt integrity does not authenticate physical execution or suppliers. Operational authorization settings were not changed or separately inspected.

Repository documentation is frozen at the qualified candidate SHA; historical status statements describe the pre-publication state. This release record and attached CI evidence state the checkpoint's publication status.

Hosted CI evidence

Public validation asset SHA-256: 25b93217fa313159432ca90a4265ac8795d7b0534939e4118971c0dbc19a997b.

v0.9.0 — 300-module snapshot (strict-build failure)

Choose a tag to compare

@solanaqubits solanaqubits released this 25 Oct 16:20

Verification correction (2026-10-25): this checkpoint has a confirmed strict-build failure. Both hosted CI runs failed on Verification/DeFiAMMIntegerSwap.lean:6:7 because the configured header linter rejects the broad Mathlib.Tactic import. The failure was reproduced from published source. The previous cached local gate is insufficient to qualify this release as a successful beta checkpoint. The original tag and validation asset are retained for audit; v0.8.0 remains the last successful beta. A corrective branch is under investigation; no replacement release is claimed here.

CI: main, tag.


Original publication record (qualification above supersedes its success claim):

A full beta checkpoint reaching 300 direct MasterSuite imports, based on v0.8.0.

  • 13 additions: 8 cryptographic algebra modules (Feldman/Pedersen VSS, share repair/refresh, threshold BLS, qualified-dealer sharing, KZG/Groth16 equations) and 5 finance modules (exact-output AMM, integer rounding, stateful routing, finite route selection, lending utilization transitions).
  • 606 local tests passed, zero skips. Both auditors checked 40,333 declarations with only propext, Classical.choice, Quot.sound.
  • 106 new theorem declarations including suite constructors, 106 Lean regression examples, and explicit scope cards for all additions.
  • Strict incremental build and full source verification; dependencies and toolchain preserved. The final documentation-only changes were checked separately against the accepted source hashes.

Full validation source: ec290046edcd9a345bdf4dbab1456e892c47f0ea. The attached validation-public.json binds the publication commit, validation digest, checks and source hashes. The 349-source metric covers the library; CLI, auxiliary tools and test fixtures are separate.

These results establish the stated algebraic contracts, not complete deployed protocol security, SNARK knowledge soundness or machine-overflow equivalence. Hosted CI runs separately with its documented target-specific SQLite exclusions; the full local test gate included that scope. SimLab and private proof-mode branches are unchanged. No operational authorization setting was changed.

See release scope and module contracts.

v0.8.0 — ALCC byte verification and full validation

Choose a tag to compare

@solanaqubits solanaqubits released this 25 Oct 13:03

Public ALCC integration with a full Leanbook validation checkpoint.

  • Eight formal modules and the optional alcc-verify CLI; canonical bytes → typed schema → SHA-256 binding → certificate verdict.
  • Full local gate: 592 live tests, zero skips; 39939 declarations audited by the project CLI and 39939 by the pinned independent auditor. Only propext, Classical.choice and Quot.sound.
  • Companion SimLab commit: solanaqubits/simlab@47e03e9 — 35 protocol/content-pin/archive tests, 16 native scenarios and 16 receipt consistency checks.
  • Public sources and native executable were checked from the exported tree. No private history or machine-local runtime PIN is included.

DECODE_ERROR is not a mathematical verdict. IDENTITY_MISMATCH rejects this certificate, not the target implication. Compiled execution is not a new per-input kernel proof; archive consistency is not authenticated execution. Expected-passport authority, Python refinement, OS/runtime behavior and collision resistance remain separate obligations. No production service or authorization barrier was changed.

Hosted CI deliberately excludes the separately target-qualified SQLite live scope; the complete local gate enabled it in the pinned isolated environment. The attached asset records actual local validation; hosted CI status is reported separately.

v0.7.48 — combined stream synchronization

Choose a tag to compare

@solanaqubits solanaqubits released this 24 Oct 03:40

Combined synchronization of completed provenance, quantum and Prometheus streams.

  • 284 MasterSuite imports / 328 byte-identical formal Lean sources against the pinned private reference.
  • Full independent local release gate: 577 tests, zero skips, strict Lake and source checks, explicit qutrit Spec, 38986 audited declarations. Allowed axioms: propext, Classical.choice, Quot.sound. Build jobs and freshly built modules are recorded separately in the attached receipt.
  • Both hosted CI runs passed. Their portable test scope deliberately excludes target-qualified SQLite cases: 96 on main, 96 on the tag. The complete local gate explicitly enabled those cases using the existing pinned artifact in an isolated container; no system library replacement or production rollout.
  • Corrected two missing explicit provenance root imports; preserved existing historical replay semantics and all earlier publication material.

Completed formal checks were retained across native test setup retries only after exact source, runtime and log identity checks; the complete unittest suite was rerun.

This is full release verification, not an independent whole-system/security audit or universal specification proof. External Witness freshness, physical durability and implementation refinement remain separate obligations. Authorization remains blocked. Historical tags and the designated v0.7.1 beta basis are unchanged.

Main CI: https://github.com/solanaqubits/leanbook/actions/runs/40052897963
Tag CI: https://github.com/solanaqubits/leanbook/actions/runs/40052898215

The attachment contains public source hashes and a sanitized verification receipt; private raw logs and Digital Brain material are excluded.

v0.7.47 — provenance reports and replay recovery

Choose a tag to compare

@solanaqubits solanaqubits released this 23 Oct 18:08

v0.7.47: provenance reports, explicit native IO and replay recovery

This release exports the integrated development snapshot for epsilon258–259
and the native provenance IO layer; reserved or unintegrated modules are excluded.

  • Granular reports separate limits, key/role/revocation checks, signatures,
    artifact hashes, context, challenge and time-policy results. Data-only preflight
    suppresses backend calls on limit refusal. Report production is not acceptance.
  • The explicit IO/FFI adapter qualifies an exact libsodium binary and executes
    detached PureEd25519/SHA-256 requests. Fixed known-answer, malformed-input,
    failure-injection and observed-request tests accompany the native code.
    It is excluded from the pure Verification import graph.
  • Replay recovery uses prepare, completed persist, trusted-prefix commit,
    acknowledgement, crash and recovery events. Reachability proves retention of
    acknowledged challenges, including the protected base, and refusal of replay
    after accepted restoration. Exact completed-image recovery is separate from
    prefix-only validation, which can accept unauthenticated suffixes.

261 direct MasterSuite imports and 305 Lean sources including Verification.lean
are byte-identical to the pinned development snapshot. The new native toolchain
and CI configuration require full beta-level public validation: strict Lake,
explicit qutrit Spec, verify-all, registry and independent audits, all live tests
and catalog checks. Local checks run under an 8 GiB cgroup. Resource-only Lean
single-thread elaboration preserves kernel checks. Actual check results are recorded
in the release receipt; this document does not substitute for those results.

CI uses Ubuntu24.04 and installs the exact qualified libsodium23 package, then
checks the committed binary SHA-256 before running native tests. A changed binary
fails closed and requires explicit requalification. No automatic profile updates
or skipped native tests are permitted. The adapter itself does not install packages.

No physical durability, trusted-anchor implementation, concurrency, exactly-once
network delivery, universal malformed-input coverage, cryptographic strength,
side-channel safety, FFI refinement or hardware authorization is proved.
Authorization stays blocked. The pure recovery result depends on retained completed
data and the independently protected prefix; serialization alone cannot prevent rollback.

The designated full public beta baseline remains
2419489 (v0.7.1); The Lake package version remains0.7.1.
A new beta tag/baseline is not allocated. Private logs, archives and organizational
materials are excluded from this public export.

Verified publication

  • Main / peeled tag: 3a730d0bd4d437f033845447226a4da2b2f4c731.
  • Annotated tag object: 440cad109ac43f601419feef2dfa96c42b8a5a07.
  • Pinned development source: edca94a093a86bfe8892118ab329b41a41591bc9.
  • 261 direct imports / 305 byte-identical Verification Lean sources.
  • Independent full beta-level public checks: 3944 strict Lake jobs, explicit qutrit Spec, 367 live tests without skips, verify-all audit of 35961 declarations / 304 sources; independent audit of 35961 declarations. Only permitted axioms.
  • Local cgroup peak: 6.750 GiB under 8 GiB; no OOM.
  • main CI: success, 367 tests; full CLI audit 35961 declarations.
  • v0.7.47 CI: success, 367 tests; full CLI audit 35961 declarations.

Native IO conformance is empirical and separate from kernel-checked pure contracts. No new designated beta baseline or hardware authorization.

v0.7.46 — provenance, replay and conditional models

Choose a tag to compare

@solanaqubits solanaqubits released this 23 Oct 12:51

Release v0.7.46 — provenance, replay and conditional models

This cumulative synchronization includes nine new modules relative to v0.7.45:

Module Scope
DigitalProvenanceVerificationContract Exact acceptance relative to explicit crypto callbacks, independently supplied context and key/time policy
DigitalProvenanceReplayContract Raw-byte challenge identity, refusal preservation and inductive sequential replay rejection
PrometheusChannelAndInjectorAudit Conditional finite-channel and injector bookkeeping
PrometheusDecoderQueue Conditional single-server queue and serialization algebra
PrometheusFiniteBlockAudit Constructive finite-block conditional reliability bounds
PrometheusPipelinedQueue Conditional pipelined queue and finite-buffer obligations
PrometheusMetrologyJitter Conditional phase response and timestamp covariance
PrometheusNQKDAudit Explicit ideal-qubit interception/noise model and counterexamples to universal disturbance
PrometheusTAKLAudit Phase invariance and conditional density/ring-model bounds

The registry contains 259 direct MasterSuite imports and 303 Lean sources including
Verification.lean. Test/tool Lean files are separate from this source count.
Proof sources are exported byte-for-byte from one pinned development snapshot.
The publication report records that pin, final commit/tag identities and actual
public validation results. English explanatory cards and regression tests accompany
the new modules; private numerical studies and knowledge archives are not exported.

Verification includes strict Lake builds, the explicit qutrit Spec target, all public
unittests with live checks and a transitive axiom audit. Only propext,
Classical.choice and Quot.sound are permitted. Validation scope is selected from
the actual change impact; historical development checks in cards are not substitutes
for this independent public-tree control.

Provenance acceptance is relative to supplied functions, not an implementation or
security proof of SHA-256/Ed25519 or trusted key/time infrastructure. Sequential
replay rejection excludes rollback, concurrent stale-state readers and storage loss.
Authorization remains blocked in the modeled result. Recovery and persistence
require separate contracts. The conditional physics/queue results do not establish
physical apparatus qualification, universal deadlines, composable QKD security or
a new weak-interaction mechanism.

Lake remains 0.7.1. The designated public beta baseline remains v0.7.1 at
2419489; this release assigns no new beta baseline tag.

Verified publication

  • Main / peeled tag: aa5f03f2e1cf8777a819946bfd6328c3f66c3f7b.
  • Annotated tag object: 6fe79ae6dbc6ebf238347077df629e4d41260dac.
  • Pinned development source: 267593bf689d51fb9c011c9aa3e13314cc0d3d3a.
  • 259 imports / 303 byte-identical Lean sources.
  • Independent gamma/scoped checks: 3942 strict Lake jobs, explicit qutrit Spec, 352 live unittests, one audit of 35085 declarations / 300 sources.
  • Local cgroup peak: 7.90 GiB under 8 GiB; no OOM.
  • v0.7.46 CI: success, 352 tests; full CLI audit 35109 declarations, only permitted axioms.
  • main CI: success, 352 tests; full CLI audit 35109 declarations, only permitted axioms.

No new designated beta baseline. CI full-module verification and local gamma/scoped evidence are reported separately.

v0.7.45 — profile bytes, manifests and conditional models

Choose a tag to compare

@solanaqubits solanaqubits released this 23 Oct 00:53

v0.7.45 — profile bytes, manifests and conditional models

This cumulative release extends v0.7.44 with epsilon243–248: QCP/01 canonical
finite-profile bytes and a bounded reader/CLI; the QPM/01 canonical provenance
manifest; and Prometheus-Pi, noise, UCN and p-B11 conditional scalar contracts.
There are 250 direct MasterSuite imports and 294 Lean source files including
Verification.lean. Previously published tags remain unchanged.

QCP/01 has strict decoding, size/count limits, round-trip and canonicality proofs,
and a restoration equivalence under an independent expected context. Its
reader decodes accumulated bytes only after EOF.
QPM/01 proves round-trip, canonicality and injectivity of encoding and the fixed
prefix signing message. It implements neither hashing nor Ed25519 verification,
key authorization, revocation, trusted time or atomic replay consumption.

The physics modules prove conditional scalar identities and inequalities:
weak-charge-square comparisons and spherical-route propagation; constant-medium
stopping and receiver error accounting; conditional UCN kinetics and mass/energy
bookkeeping; and p-B11 shot energy conservation and extraction bounds. Model inputs
and their physical interpretations remain explicit premises. These proofs do not
qualify a detector, communication system, reactor or energy-producing device.

Public acceptance checks include strict Lake builds with an explicit qutrit Spec,
all live-enabled public unittest, catalog validation, the transitive union axiom
audit and a pinned independent axiom-dependency audit. Only propext,
Classical.choice and Quot.sound are allowed; native_decide is not admitted.
Exact executed results and main/tag CI outcomes are attached to the GitHub release.
Local checks use a memory ceiling of 8 GiB and shared Mathlib artifacts.

The full public beta baseline remains
2419489 (v0.7.1); Lake remains 0.7.1.
No new beta baseline/tag is allocated. Test/audit counts are evidence for the
specified checks, not exhaustive semantic correctness or physical provenance.
Authorization remains blocked. Hardware qualification is outside this release.

Verified publication

Commit: 8de4ac8436dafbae67731e9974f1ead85feff1dd. Annotated tag object: 93faa6f82a807892683cd9654383481500071c56.

Independent public full validation gate: 3933 Lake jobs, 1 fresh compilations; 336 live-enabled public tests, no skips; 34134 audited declarations across 293 sources.

v0.7.44 — canonical snapshots and finite profiles

Choose a tag to compare

@solanaqubits solanaqubits released this 22 Oct 23:09

v0.7.44: canonical snapshots, bounded IO and finite profiles

This release publishes epsilon240–242 and the bounded snapshot reader/CLI:

  • QCS/01 byte frames: canonical ULEB128 and reduced signed rationals, exact EOF,
    inclusive size/bound/row limits, proved round trip and inverse canonicality.
  • Strict structured RatWire/QuotaSnapshotWire decoding: positive denominators and
    coprimality, rejection of reducible pairs and noncanonical zero, unconditional
    decode w = some s iff encode s = w, and certified cursor restoration.
  • Finite profile transfer for precisely p(0),...,p(n-1), with checks against an
    independently supplied expected profile and proved restoration equivalence.
  • Bounded read-only file IO and CLI: short-read handling, an EOF probe at the exact
    limit, deterministic reported IO failures and strict decoding only after EOF.

The registry contains 244 direct MasterSuite imports and 288 Lean sources,
including Verification.lean. Independent public verification includes
strict Lake compilation, an explicit qutrit Spec build, all public unittest tests
with live mode and a union transitive axiom audit. The conservative planner requires a full validation gate
for the new tools/ IO adapter: verify-all, registry and independent audits are
therefore included. Lake uses weakLeanArgs = ["-j1"] to bound per-file elaboration concurrency; this
resource-only option preserves kernel checks and build traces. Individual local
Lean checks also use one elaboration thread under
an 8 GiB process-group limit. The validation scope is beta-level; no new beta tag
or baseline checkpoint is allocated. Lake stays at 0.7.1 and the full public beta basis remains
2419489 (v0.7.1).

Byte format · Reader/CLI.

The structured codec, byte codec, semantic restoration and expected-context
checking are separate layers. Arbitrary rational coefficients are permitted;
probabilistic interpretations require additional premises. File IO is tested,
not a formal OS refinement. No provenance, authorization, universal performance
improvement or total runtime resource bound is established. Profile byte encoding
and profile file IO are not part of this release.

Verified publication

Commit: 22137c7cc005c35f12ce8de8ef7692df0829f3e5. Annotated tag object: a4f6153f7629c1b69b87c64014611fa0050d548c.

Independent public full validation gate: 3926 Lake jobs, 1 fresh compilations; 317 live-enabled public tests, no skips; 33359 audited declarations across 287 sources.