PF Simulation r34 — delivered package and validation

The current r34 package consolidates the confirmed Assistant Recovery, Loading and Opening corrections and the private-workflow reliability maintenance. It retains the Astropy reference-cache correction, shared configuration-consistency controls, tracking preparation and controller diagnostics, and execution/log/Orbit Viewer recovery. The proposed Orbit Viewer Header patch is excluded. It also retains the earlier public-service, Micius preset and Refresh corrections. It retains autonomous ISAES, private workflows, source-layer overrides, project and disclosure controls, regular storage and optional BYOS. Reporting framework 2.3.0, SDK 1.0.0, the 239-page end-user manual, Edition 1.19.0, and the 50-page private-workflow guide remain the retained scientific references. Read the current workbench guide and dated maintenance guides alongside them for the later configuration and HF changes.

The hosted PF Simulation trial/demo is available for prospective users to register a PF Simulation account, sign in and try the workbench. Use the Getting Started launch button to open it in a new browser window or tab. Read the hosted trial guide, HF deployment guide and public-service guide for their respective user and operator workflows.

Cumulative validation across retained releases

The cumulative validation ledger reconciles 8,657 distinct passing parent test identities across 64,845 passing parent executions, 312 included JUnit records and 63 validation scopes. It inventories 720 JUnit files and selects records under the documented receipt and deduplication rules. Repeated identities count once. The 6,541 successful subtest executions and 6,823 numerical checks across 46 studies remain separate units, not parts of a grand total.

The scope begins with the earliest retained identifiable release evidence. Missing older records are not replaced by estimates. A cumulative passed identity records an execution against its recorded source; it does not assert that every test was rerun on the current source or that an arbitrary research program is scientifically valid.

Download the machine-readable identity reconciliation and release-group breakdown. The website source includes the accounting scripts and retained supporting records. This 17 October refresh adds 138 distinct identities and 441 passing parent executions to the 14 October Fix_23 website ledger (8,519 identities / 64,404 executions). The earlier Fix22 and Fix_23 identities are all preserved. The numerical ledger remains unchanged; subsequent executions of already retained numerical identities do not create new benchmark checks.

Current Assistant consolidation and preceding workflow correction

The latest receipt records 14 successful commands: two Python groups and 12 JavaScript UI groups. The Python groups passed 305 parent tests, with 70 successful subtests reported separately. The UI groups passed 232 cases and 49 separately reported assertion checks. Seven PostgreSQL-dependent cases were skipped, and one scientific-worker case was deliberately deselected. Skips and deselections are not passes.

Receipt: validation_r34_assistant_consolidation/test_execution.json. Immutable run: 20261016T175510460426Z. Receipt SHA-256: 75bbd938e8af2327550dd1f1c78a2a4836b5166e59282179eb9068e2c2d35e3b.

The Assistant consolidation publishes immutable source before owner-session transactions, avoids redundant unchanged index writes, reads saved workspace metadata together, and makes Open selected run an explicit evidence-capture action. It retains saved-key privacy, authorization, integrity checks and controlled mutation behavior. The shared service and UI changes apply to standalone and SaaS; the HF adapter retains its own source-publication and storage-recovery changes. Request-local evidence reuse does not cache simulation executions or reuse earlier results for a new run.

The preceding workflow reliability correction recorded 136 passing Python parent tests, 152 UI cases and 39 separate assertion checks across 11 commands, with no failures, errors or skips. It coalesces workflow refreshes, releases loading controls on errors, reads owner storage once per list request, records sanitized database categories and corrects native HF staged workflow paths. Its receipt is validation_r34_workflow_reliability/test_execution.json, run 20261015T160625498508Z.

Workflow maintenance adds 53 distinct identities and Assistant consolidation adds 85. Their 441 combined Python executions are added to the execution ledger; overlapping historical patch reruns are not counted again. UI cases and assertions remain separate from the parent-test identity total. No new numerical study or numerical-check identity is added.

The Astropy cache correction remains included with its original 106-test receipt. It creates an empty per-attempt reference-cache directory before restricted execution. Each simulation still calculates fresh results; dependencies, reference-data freshness checks, scientific equations and isolation restrictions remain unchanged. The configuration correction, tracking recovery and execution recovery likewise retain their original evidence scopes.

These targeted checks did not rerun the full scientific regression, make live AI-provider requests, measure browser performance or qualify production capacity. Frozen package receipts retain their original pending gates. Later operator-supplied deployment logs are reported separately below.

Retained release acceptance and counting units

The cumulative ledger now includes hybrid and native HF integration, execution compatibility, deployment documentation, user-interface controls, execution and publication lifecycle, runtime consolidation, scalable runtime, shared deployment fixes and public-service acceptance. The complete release-group ledger records each revision, receipt identity, Python executions, added distinct identities, subtests and JavaScript commands separately.

Selected retained receipts illustrate the scope:

These executions overlap; the rows must not be added to obtain a distinct-test total. Skips are not passes. The separately retained 144-pass isolated PostgreSQL qualification is an aggregate receipt without the individual testcase identities needed for this identity ledger and is not added to its headline.

The numerical ledger retains model-specific checks of quantum states, QKD distributions, SatQuMA quantities, gauge-wave links, channel tomography, phase correlation, orbits and Floquet dynamics. They remain scoped calculations with declared references and assumptions. Provider replies in local protocol tests are explicit test doubles; their success does not establish the scientific accuracy of a live model. Read the implementation and acceptance report, readiness assessment and external qualification guide for those boundaries.

Delivered package identity

The release contains 34,028 verified members and a selected source inventory of 1,641 inputs. Package readback verified all members, CRCs, SHA-256 values and the complete ZIP stream. Decompress the XZ wrapper, then extract the resulting ZIP. These identities refer to the research application, separately supplied from the website upload. Application source access remains through the contact route.

Read the current release metadata and SHA-256 list. Hashes establish byte integrity; they do not by themselves authenticate a publisher or certify a scientific conclusion.

From an untouched extraction:

python release_tools/package_r34_assistant_consolidation.py verify --package

Earlier release-specific verification commands apply to their recorded source identities. The current verifier preserves the earlier evidence chain; earlier hashes are not rewritten to present old executions as new ones.

Hosted service and qualification boundaries

The operator reported the hosted trial/demo live on 9 October 2026, with public registration and sign-in through the hosted workbench. On 16 October 2026, the operator confirmed the excessive Assistant loading delay was resolved with the Recovery → Loading → Opening patch chain. Subsequently supplied HF status and startup logs showed the consolidated full release installed and running at controller revision a9a912355f05ffc8085ea3474e57609fa2af59da, with source bytes and the running revision verified, preserved configuration and only the intended three code pins changed. Startup reached Assistant source readiness and HTTP serving. These are operator-supplied observations; they are not independently timed browser measurements or new simulation-run evidence. The deployment helper explicitly left HTTP readiness and performance unverified. It does not convert the frozen local package tests into a live simulation, production load, security or scientific qualification.

Admission settings and available resources determine concurrent execution and queue capacity. A configured job limit is not a measured concurrency guarantee. Mixed-workload capacity and soak testing, actual fleet failover, applicable production/BYOS deployment, external database restore, live AI/provider scientific quality, representative researcher evaluation and external scientific replication keep their own qualification requirements. Consult the public-service operating guide before deploying a separate service.

ISAES analyses retained scientific results; report regeneration does not rerun the original simulation. Shared analyses, private extensions and rendered outputs keep their declared origin and evidence. Reports work without AI-provider credentials. Optional report discussion and autonomous investigation follow their configured capabilities and consent. Review, apply and run are separate actions; interpretation remains separately attributed commentary.

Archiving literature is owner-wide and preserves existing evidence. Association release removes selected administrative links while protected scientific records remain. Permanent deletion requires an authoritative dependency check and a fresh coordinated check. Workflow and conversation lifecycle controls remain user actions, with explicit recovery for interrupted cleanup. See ISAES lifecycle and workflow lifecycle.

The cumulative record retains A2 source/input provenance, B1 Linux hosting procedures, B2 project authorization and B3 AI disclosure enforcement. These controls remain described in the adoption guide, CLI provenance guide, hosting reference, project confidentiality guide and research policy guide.