# September 23 combined source candidate

This is an unverified integration candidate, not an accepted release. Gitea
issue27 remains the live integration/release record; Codex-956atf owns it.
User-owned requirements and current root instructions are unchanged.

## Source provenance and decisions

The base is the saved Scenario/Text composition rooted at
`5960563b23b3f28f83343c283b7c156bf248994a`, then Scenario integration
`6f4dfc67b3443b2b1725e2b576eb37224977b9b2` and early stale-event correction
`0b61771d079cae4a513958fad04e6c3e785d458f`. These retain the root's ordinary
defaults, Text/Tab, always-active canvas, rename-scroll and allocation authority.

The next two source commits compose exact delivered code/documentation deltas:

- `a9b9c81`: ordinary/Text `3f9e6a18..b8685416` under `crates/` and `docs/`,
  including final attachment Escape and State read-only pan corrections. The
  Scenario invalid-route fixture additionally supplies absent attachment fields.
- `ffa7b16`: `3f9e6a18..55198ee5` under those paths, retaining the final persistent
  transition decision footer. The already-identical `diagram/focus.rs` formatting
  is retained once. The bidirectional lifecycle fixture now authors nonempty
  source/target attachments, bends and labels; its existing whole-Project,
  deletion/history/recreation/relink/round-trip oracles must conserve them.

Both net patches applied cleanly with three-way reconciliation. That is not
evidence of behavioral compatibility. In particular, preserve Scenario's
architecture-destination refusal, exact view/visit authority, State edit-only
cancellation versus camera cancellation, and semantic ports' authority over
dormant attachment preferences.

The first Scenario-only run at `0b61771` compiled and listed 129 core-library
tests on 742 unchanged inputs. Its `scenario_layout` run passed 20 and failed
seven source-lifecycle cases: the ordinary parser's closed field inventory
rejected `layout`. Correction `2a0579daafb89994730284c108f460782e5e6650`,
incorporated here as `846cae8`, adds `layout` to parser admission only. It does
not add a synthetic null to CompleteForEdit or change omission defaults; real
payload clearing remains distinct from unrelated edits beside explicit none.
The failed receipts remain retained. At corrected `2a0579d`, all 27 Scenario
layout and 22 source-default library cases pass against the actual 129-name
inventory on 742 unchanged inputs. These are local Scenario-only receipts;
they do not establish the combined attachment/State/transition result.

## Evidence scope

Existing peer evidence remains tied to its original code, not this candidate:

| Contribution | Final source | Published evidence |
| --- | --- | --- |
| Saved Scenario / inspector actions | `14e7e6aa46f59346c7a9a478fc9498fd0b496aaf` | `3a5fd8e648797837bbfcd8ede1edab96216d7ab0` |
| Attachments | `53d01a3d4451b94792986f2dfac56f6f04e111a4` | `bb7edfbf01b43577b01d4dc7f6b8dd9240145759` |
| State bends / read-only pan | `b8685416f56e53e30bd11ecf93a54cea10ac8458` | `d6aabb5dec82900ebfac1183abb4036ca06cc0e5` |
| Transition lifecycle/footer | `55198ee5112aa67e063436e4c6a253368e420689` | Same commit |

The independent September 23 evidence audit verified five supplied portable
verifiers and supplemental actual outcome/source-binding checks. Historical
failed attachment comparisons and State dirty-source gaps remain disclosed.
The audit did not run their application binaries or repeat their native trials.

Retained reviews: [composition plan](reviews/peer-composition-plan-20260923.md),
[Scenario source and stale-event correction](reviews/scenario-text-union-source-20260923.md),
[combined source](reviews/peer-union-source-20260923.md), and
[peer evidence audit](reviews/peer-evidence-audit-20260923.md). Their historical
source pins and pre-execution status remain explicit; later passing gates do
not retroactively turn a source review into runtime evidence.

## Combined acceptance still required

First resolve any actual constructor/API/parser failures on frozen source.
Verify actual test inventories and outcomes rather than adding historical totals.
Core checks include Scenario source/reference lifecycle, ordinary defaults,
attachments/routing/ports and both-direction transition lifecycle. Desktop
checks include stale-input/navigation/spacing, all nine Text indentation cases,
inspector tab return, attachment Escape and all State bend/pan regressions.
HTTP checks cover conditional Scenario and attachment transactions and restart.
Run full workspace tests, strict all-target lint and matching GUI build after
targeted defects are corrected.

A combined native trial must exercise saved arrangement/Details/inspector
Apply/Revert and Connections return, spacing/cancel, Text indentation and
Apply/return, attachment versus semantic reconnect, State bend editing/cancel,
transition footer/relink, and exact saved-model reopen. Peer native evidence
does not replace observation of these integration seams.

Source7/schema7 versus the proposed coordinated source8 boundary is unresolved;
the intermediate Scenario contribution does not settle public format adoption.
Normative Scenario language/API/OpenAPI documentation is incomplete. Private
analysis source8 is not merged here. Resize policy, complete accessibility,
Linux and packaged-release acceptance remain separate requirements. Linux
checks stay deferred under the user's instruction.