[AronAxe]

Implemented

Operational System One/JEV control across GraphPaper, with a real sidebar decision studio:

• Focused Noul predicates, descriptive Score levels and probability-aware Choice distributions. Authorial-fidelity failures cannot be compensated by fluent prose; mixed scored/unscored graph priorities use explicit tiers.
• Stage-level fan-out, exact-context cache, provider/model/policy receipts, validated distributions and bounded request sizing. Batch preflight reserves writing calls; unavailable/uncertain decisions never become automatic approvals.
• Semantic source selection, task-relevant saved voice habits, bounded graph exploration and explicit, undoable entity-identity proposals.
• One-pass angle-intent repair with originals preserved, independent passage checks, changed-prose checks and confidence-gated automatic refinement. Science triage/full-text priorities/appraisal gaps and fiction continuity are integrated. No new politeness or vocabulary filter.
• Reusable author corrections and prefer/reject feedback. Local regularized preference learning uses whole-project holdouts and activates only against a measured baseline; it never overrides fidelity constraints.
• Optional routing only among author-approved model alternatives/efforts on the same provider. Default off.
• Decision desk exposes question wording, all distributions, Score legends, context/model identifiers, resulting policies, originals, feedback, training, exports and configuration.

Verified before merge

Successful release candidate: https://github.com/AronAxe/GraphPaper/actions/runs/38521929515 . Exact checked/built source: 45a13c5edc762ed9ed0c9131c15edcdc73f79ef6.

• 343 automated Windows tests passed, no skips/failures. Linux: 342 passed, one Windows-only skip. Six geometry tests passed.
• All eight real-HTTP browser suites passed, 75 checks, including System One feedback persistence, truthful insufficient-data training, exported decisions, opt-in routing and narrow-window rendering.
• 33 checks passed against the actual compiled Windows executable: native mouse input, public Graphify worker, graph editing/undo, writing modes, reusable voices, typed decision desk, saved feedback, local configuration, export, native Save As and pending-edit persistence on close. No page errors, recursion errors or abnormal exit.
• Downloaded and inspected both hosted reports and actual application screenshots independently. The Windows package—not merely the browser interface—was tested.
• The standard quality matrix (Windows/Linux, Python 3.12/3.13) and documentation PR checks passed. Review regressions include corrected writer-call reservation, author corrections in judgments, source evidence in standalone audits, and no false successful scene repair when checks are unavailable/uncertain.

Synthetic provider fixtures validate transport, policy and UI behavior; no live JEV editorial-quality or cost-savings claim is made. Initial thresholds are engineering policies, not domain calibration.

Preservation and delivery

No user-machine access, private manuscripts, credentials or live model accounts. Existing graph-paper UI, approved header/schematic, project data, voice library, Graphify/Codex remain. Documentation updated for v0.7 and System One wiki guide added. Temporary integration files are absent from the final tree; release workflow contains no source-patch application step.

Merge triggers the main-branch release gates. Only the exact Windows archive that passes native tests and uploaded-checksum verification is publicly published as v0.7.0.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-14T06:38:04.335397Z	45a13c5	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 45a13c5edc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Retain passages outside the semantic scoring window

When a source pack contains more than 18 passages and JEV returns scores, this slice excludes every later passage from ordered and therefore from the returned pack, even when the context budget has room. This can also omit an explicitly requested required_id whose first passage falls after position 18, so outlining or drafting silently loses selected evidence; include required passages in the scoring window or carry unscored passages into the budgeted sequence.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Remove internal edges when merging connected nodes

When the two merged nodes are directly connected, redirecting both endpoints converts that relationship into a self-referencing keep -> keep edge. Graph extraction and import explicitly reject such edges, while motif mining will treat this malformed edge as a real relationship and can generate single-node motifs from it; discard edges whose endpoints collapse to the same surviving node during the merge.

Useful? React with 👍 / 👎.