# PF Simulation r34 — ISAES implementation and acceptance

This dated record retains its original release identity and evidence scope. For the latest r34 package, see [current release validation](r34-validation.md), the [cumulative ledger](validation-cumulative.md) and the [hosted trial guide](hosted-trial.md). Earlier hashes and test counts below continue to identify the work originally recorded.

Reporting/ISAES **2.2.0**; extension SDK **1.0.0**. This is a documentation and Documentation-navigation maintenance release. Product and archive filename remain r34. Fresh scoped execution was recorded at `2026-09-28T05:22:15.473246+00:00`.

## Delivery decision

The documentation maintenance has passed its recorded local acceptance and package integrity checks. The complete End-User Manual now incorporates ISAES and its related user workflows. The release remains ready for controlled report-assistant evaluation, with that feature disabled by default. Full original-scope acceptance remains pending native Windows execution, representative researcher usability and applicable production/BYOS qualification. Live provider/model scientific quality and hardware qualification are not claimed.

## Documentation and navigation changes

The manual is **Edition 1.18.0, 28 September 2026, 236 pages**. Chapter 32, **ISAES: from workflow results to evidence**, provides an illustrated task walkthrough. It explains the automatic processing alongside the researcher’s choices: report setup, status interpretation, evidence inspection, linked figures, saved views, revision comparison, exports, literature management, optional assistant proposals and recovery. It also covers standalone and hosted use, credentials/storage boundaries and supported limits.

The Documentation tab in standalone and signed-in SaaS now exposes the ISAES, report-assistant, private-workflow, assistant, SaaS and GUI guides. File labels distinguish PDF from Markdown; Markdown is served as readable text. Existing document allowlists and access checks are retained. Six guides have matching navigation and manual cross-references.

The manual build record, document-source and image hashes, qualification routing, source manifest, package checksums and repackaging records have been refreshed. The original versions of changed inputs are preserved in the maintenance evidence. Prior scientific and software acceptance records retain their exact bytes and original scope.

## Fresh maintenance validation

**94 recorded Python cases passed**, with no failures, errors or skips, plus **1 separately recorded JavaScript command**. These current-source checks cover the documentation routes and real browser navigation, authentication/allowlist boundaries, documentation-release integrity, qualification readers and startup compatibility. This is a scoped maintenance gate; the numerical engines, SDK enforcement and assistant engine behavior were not reimplemented or rebenchmarked.

| Suite | Recorded Python cases | Failures | Errors | Skips |
|---|---:|---:|---:|---:|
| documentation_integrity | 26 | 0 | 0 | 0 |
| documentation_navigation | 5 | 0 | 0 | 0 |
| reader_and_startup | 63 | 0 | 0 | 0 |

JUnit cases can include subtests; they are not counts of unique test functions. The browser test executes actual Chromium interactions. It does not constitute representative researcher usability testing.

Current receipts are in `validation_r34_isaes_documentation`. They bind the frozen source, documentation assets, runtime inventory, command logs and JUnit records. The reconciliation enforces a closed documentation/navigation/verification change boundary; scientific, analysis, SDK and assistant engine inputs are unchanged from the incoming release.

## Retained implementation baseline

The incoming archive `870b7de102f37d83f0988fbbfac91186030b534c265caf934784f3be09b70eb0` contains the completed SDK correction and reporting-specific assistant work. Missing/null native SDK prerequisites reject invocation; dependent calculations are blocked. The assistant binds an authenticated owner to an exact report identity and can read bounded evidence and submit attributed proposals through the existing provider/BYOK controls. Researchers review a proposal, apply an approved change to a queued child revision, and explicitly run that revision. Approval alone does not execute it. Original simulation and report records remain retained.

Trusted invocation records preserve provider/model identity, supplied evidence, response or failure and proposal references. Cooperative pause and explicit continuation operate at eligible checkpoints. Structural reference checks establish membership in supplied evidence; they do not certify scientific interpretation or free-text citation correctness. Dense linked-view restoration batches rendering updates while retaining exact values and observation identities.

The baseline acceptance at `2026-09-28T04:05:47.049817+00:00` recorded **1436 Python cases and four JavaScript commands**, with no failure, error or skip. Those receipts remain under `validation_r34_isaes_ai` and are **historical baseline evidence**, not newly executed results for this documentation revision. Provider protocol fixtures did not use a live AI provider, and that feature release did not rerun the complete PF scientific benchmark catalog.

## Retained performance measurements

The following measurements belong to the preceding feature release and were **not rerun for this documentation update**. They characterize bounded Linux workloads rather than a production SLA.

| Workload | Baseline measured p95 | Recorded limit | Repetitions |
|---|---:|---:|---:|
| Analysis of 100,000 observations | 1.273 s | 20 s | 5 |
| Standard report/export workload | 2.272 s | 60 s | 3 |
| Workbench first useful view | 1616.22 ms | 10000 ms | 3 |
| Workbench linked selection | 80.63 ms | 1000 ms | 3 |
| Workbench saved-view restore | 386.57 ms | 2000 ms | 3 |
| Dense 100,000-point first useful view | 1675.53 ms | 10000 ms | 3 |
| Dense 100,000-point linked selection | 385.73 ms | 1000 ms | 3 |
| Dense 100,000-point saved-view restore | 438.17 ms | 2000 ms | 3 |

Baseline peak process RSS was 62.8 MiB for analysis and 138.5 MiB for export, against limits of 1024 and 1536 MiB. Dense restore left 1561.83 ms of measured margin. Nearest-rank p95 is the slowest observation at these repetition counts. Browser heap observations are not process peak RSS.

## Remaining acceptance gates

| Gate | Evidence still required |
|---|---|
| Native Windows | Supported Windows/Python 3.12 execution and retained current-source receipts. |
| Representative researchers | At least five representative researchers completing the five defined tasks; at least 80% unassisted completion, no critical scientific misinterpretations and independent review. |
| Hosted production isolation | Applicable deployed containment, resource, cancellation, publication and recovery checks. |
| Live BYOS | Applicable authorized storage, ownership, failure/recovery, retention and deletion checks. |
| Live provider/model evaluation | Actual-response assessment of interpretations, citations and proposals before broader assistant activation. |

Conditional deployment gates may be explicitly and justifiably out of scope for a deployment that does not use those capabilities. Unknown applicability remains pending. A corrected guide and browser navigation tests do not close human or deployment qualification.

## Package identity and verification

Final archive: `PF_Simulation_Standalone_and_SaaS_r34_private_workflows.zip.xz` (463,460,088 bytes).

SHA-256: `1987acf5712a7d2311744089bfb350885501bd925e18a2c31ee43447329d6a74`

The seal read back **31,058 members**, checking membership, CRCs, each member’s SHA-256 and the complete decompressed ZIP stream. Source inputs: **1208**. Source-map identity (canonical JSON SHA-256): `d6c78a90974fe982c280a870e471058c547bc4cf84b8a94cb113be2246ab7a19`.

After extraction, run from the `privileged-frame-simulation` folder:

```text
python release_tools/validate_r34_isaes_documentation.py verify --package
```

Use the accompanying SHA256SUMS file to check the downloaded archive, manual and guides before extraction. The package contains its own `PACKAGE_CHECKSUMS.json` and preserved prior evidence. The main manual and guides are available from the Documentation tab. For controlled optional assistant evaluation, see `docs/ISAES_AI.md`; enabling `PF_ISAES_AI_ENABLED=1` grants neither credentials nor additional permissions.