PF research library · 5 min read
ISAES original-scope implementation and acceptance map
r34 documentation · isaes-scope-completion.md
On this page
This original-scope acceptance map retains its recorded framework version and evidence. The current r34 package uses reporting framework 2.3.0; see current release and validation for its later changes and cumulative record.
Product r34; reporting framework 2.2.0. The preserved scope is docs/ISAES_ORIGINAL_SCOPE.md. This matrix describes implementation and its acceptance path; generated receipts, not this document, determine whether a gate passed. Full original-scope acceptance also requires genuine external evidence identified below. The optional report-assistant implementation and its controlled-evaluation boundaries are described in docs/ISAES_AI.md; its new release evidence is separate from the preserved 2.1.0 completion receipts.
| Original scope / readiness finding | Implemented path | Current acceptance source |
|---|---|---|
| Typed scientific values, dimensions, coordinates, units and missingness | Versioned contracts, xarray/Pint adapters, preserved masked numeric encoding, explicit eligible-row selection and inferential limitations | Core, scientific, methods and scientific-completion suites |
| Dependency planner and strong shared interfaces | Bounded DAG, explicit prerequisites and outputs, chained findings, preprocessing, capability snapshots, native SDK and declared legacy compatibility | Planner/SDK suite; documented extension examples |
| Confirmatory versus exploratory intent | Owner-locked policy receipt binds source, workflow, bundle and definition before scientific execution; changed/post-run plans cannot acquire prespecified status | Operations and core suites; not a scientific-design certification |
| Uncertainty, numerical analysis and exploratory ML | Covariance/Monte Carlo, independent and paired estimates, convergence/reference residuals, sampling-aware dynamics, PCA, clustering/density/anomaly, held-out regression | Methods/scientific-completion suites and independent mathematical fixtures |
| Quantum evidence and selected specialist protocols | Explicit acquisition/derived origins; local/joint state evidence, shadows, complete-Pauli tomography, declared Fock/spin Husimi, supplied ZNE/PEC/calibration/chemistry/optimizer records | Domain, quantum-protocol and scientific-completion suites; hardware collection stays optional |
| Actual Floquet cycle exploration | Separately versioned workflow records genuine reduced-q0 cycle checkpoints and links them to observables/table; original scientific outputs preserved | Real 6-qubit/8-cycle invariance test; live browser path; bounded 8-qubit cases |
| Nonquantum/PF domain path | Production radial-ray SR/GR workflow, explicit metric/coordinate/clock semantics, shared timing adapter | Domain-adapter suite including actual private executor and integration runner |
| Existing analyzer and legacy migration | Legacy receipt adapter retains imported verdict scope; integration runner adds common exports; verified non-destructive migration CLI; private SDK conformance | Domain-adapter and compatibility suites; old report bytes preserved |
| Findings, claims and evidence graph | Source slices, transforms, execution hashes, findings, diagnostics, claims, citations and views resolve through shared contracts; separate status dimensions | Core and evidence suites; imported/private authority stays explicit |
| Literature usability and lifecycle | Visible delete/review actions, owner-wide archive explanation, named blockers, document-level release of selected administrative history; scientific dependencies stay protected | Entire current literature-management suite and live browser regression |
| Rendering correctness | Frames cannot override kind/encoding/identity/scale; all frames validated; exact figure JSON retained; log/nonfinite/private cases fail explicitly | Presentation-completion suite and prior presentation tests |
| Guided planning and composition | Effective inherited configuration can be customized explicitly; guided datasets/methods/references/comparisons; compatible choices, audience/detail/notation/order/templates | Workbench DOM and real-server completion browser suites |
| Embedded linked exploration | Authenticated artifact loading into sandboxed iframe, bounded active Plotly charts, table filtering, linked observation/checkpoint IDs and saved state | Actual Chromium workbench path plus focused presentation browser tests |
| Static/current-view fidelity | Shared document and visualization plans; PDF/PPTX/HTML/JSON/figures; new render-only revisions replay validated saved views over original checkpoint, without recomputation | Service operations, presentation and actual browser export checks |
| Private extension authority | All calculations in a private extension/layer environment are conservatively attributed to that environment; SDK cannot use public registration arguments to claim core authority | SDK mutation/conformance tests and trusted-parent rejection of forged analysis/render output; existing isolated-worker security remains required |
| Execution and recovery | Immutable snapshots, separate analysis/rendering, checkpoint hashes, bounded retries/cancellation, owner locks, exact source/runtime invalidation | Existing and current service/lifecycle tests; no unattended distributed scheduler claimed |
| Controlled activation/rollback | Owner-scoped pause and workflow-version allowlist; compatible readers and deletion guards retained; bounded emergency stop at normal quota/history limits | Service/HTTP controls tests and live pause/resume path |
| No-AI independence | Reporting and export paths run with provider disabled and credentials removed; evidence/proposal API exists for the later AI phase | Release runner and no-provider HTTP/browser paths |
| Performance and interaction budgets | Frozen workload contract; subprocess analysis/export measurements; source-bound live-workbench and 100,000-point HTML navigation latency distributions, plus bounded resource cases | performance.json, completion-browser receipt and acceptance_status.json |
| Accessibility and visual quality | Keyboard/mobile automated interaction plus inspected representative screen/export fixtures | Presentation and completion-browser receipts; representative-user understanding remains a separate gate |
| Researcher usability | Predeclared five-task protocol, at least five representative researchers, completion/assistance/error/time records and review manifest | Requires actual participants; scripted browser tasks are not a substitute |
| Native Windows and intended deployments | Native Python 3.12 qualification command; source-bound deployment evidence and review procedures | Requires real Windows; production isolation and live BYOS when those modes are deployed |
| Final archive integrity | Pinned preceding archive, explicit source reconciliation, fresh tests, unchanged old evidence, package/member hashes and readback | validate_r34_isaes_completion.py verify --package and delivered archive checksum |
The supported catalog is bounded. This release does not claim automatic understanding of arbitrary scientific programs, unrestricted full-state scaling, universal credibility scores, arbitrary Bayesian/causal discovery, live device access or experimental validation from simulations. Optional OCR, embeddings, remote hardware and later AI activation remain optional or later-stage work as stated in the original scope.
Operational pause affects new reporting execution, including rendering revisions; existing reports and downloads stay readable. Pre-registration is an auditable prespecification record, not proof that a decision rule, power calculation or physical model is scientifically adequate. A view revision may change displayed filters, selected frame and permitted presentation options while original analysis and exact retained values remain available.
A machine-generated acceptance status can establish receipt integrity, numerical thresholds and source binding. It cannot certify that a person is a representative researcher or that a supplied declaration describes an actual deployment. External evidence therefore requires explicit accountable review, and missing external evidence must remain pending.