Fusion - the Fusion 360 Bridge
Public Made by Adomby adom
Drive Autodesk Fusion 360 from the cloud via Adom Bridge: component libraries, IPC package generation, board layout, exports (STEP/Gerbers/BOM/CPL), fast APS cloud search, and parametric modeling.
master
a06dfd7
26d ago
R21 author review — source scope only
The author reviewed the new policy, typed ABI, native adapter, route/header and journal paths against the archived R20-fix3 baseline. This is not independent review. Parent root remains the independent reviewer and sole host operator.
Verified fixed reference capacity and adoption before fallible snapshot work; exact current root occurrence/component/snapshot membership; public retained snapshot identity; no borrowed SDK handle crossing a worker; no implicit release on ambiguity; child-before-parent release ordering; finite position/unit limits; identity orientation; both grounding flags refused; complete before/intermediate/ after projection comparison; no mutation retry or rollback; stale revalidation before start and before SDK execution; separate acknowledged transform/capture stages; durable-start and retained terminal-loss behavior; old operation headers and primary contract preserved; truthful narrow capability/schema.
One ordering gap was found and corrected before freeze: the second position getter set originally followed the final root capture. It now lies between the two full captures. A deterministic regression changes root revision in that getter and confirms unknown refusal before any snapshot add. No extra SDK reads or settling allowance were introduced. Other initial development failures (two ABI prototype mismatches, test access spelling and two inherited operation count expectations) remain in working logs; final gates all pass.
The maximum snapshot/result test uses 32 components, 64 occurrences, 63 prior snapshots, near-128KiB projection metadata and a 64th new snapshot; the result fits the reserved 256KiB observation limit. Failure tests verify exact effect call counts and book faulting, including capture/command panics. Actual SDK matrix/snapshot health, root/component revisions, timeline behavior and position persistence remain unqualified by these tests. No SDK health code is silently treated as healthy. No successful prior unknown operation is inferred.
The SDK-linked Windows build, C21 live campaign and independent public SDK observer are not included in this source checkpoint. No root verification source is part of the product manifest. No John review rules or UI were added.
Freeze2 classification correction
Post-freeze author review identified that the newly named orientation refusal and translation bound were absent from the read route's known-refusal list. Freeze2 adds precisely those two pure projection-validation errors. A public Book::inspect test verifies both invalid projections invoke no effects and are classified as known refusals; SDK errors, changing observations and uncertain setter acknowledgment remain unclassified as known refusals. Exact freeze1 bytes/manifest are archived in evidence/freeze1-source. This is a narrow error classification correction, not new placement support or live qualification.
# R21 author review — source scope only
The author reviewed the new policy, typed ABI, native adapter, route/header and
journal paths against the archived R20-fix3 baseline. This is not independent
review. Parent root remains the independent reviewer and sole host operator.
Verified fixed reference capacity and adoption before fallible snapshot work;
exact current root occurrence/component/snapshot membership; public retained
snapshot identity; no borrowed SDK handle crossing a worker; no implicit release
on ambiguity; child-before-parent release ordering; finite position/unit limits;
identity orientation; both grounding flags refused; complete before/intermediate/
after projection comparison; no mutation retry or rollback; stale revalidation
before start and before SDK execution; separate acknowledged transform/capture
stages; durable-start and retained terminal-loss behavior; old operation headers
and primary contract preserved; truthful narrow capability/schema.
One ordering gap was found and corrected before freeze: the second position
getter set originally followed the final root capture. It now lies between the
two full captures. A deterministic regression changes root revision in that
getter and confirms unknown refusal before any snapshot add. No extra SDK reads
or settling allowance were introduced. Other initial development failures
(two ABI prototype mismatches, test access spelling and two inherited operation
count expectations) remain in working logs; final gates all pass.
The maximum snapshot/result test uses 32 components, 64 occurrences, 63 prior
snapshots, near-128KiB projection metadata and a 64th new snapshot; the result
fits the reserved 256KiB observation limit. Failure tests verify exact effect
call counts and book faulting, including capture/command panics. Actual SDK
matrix/snapshot health, root/component revisions, timeline behavior and
position persistence remain unqualified by these tests. No SDK health code is
silently treated as healthy. No successful prior unknown operation is inferred.
The SDK-linked Windows build, C21 live campaign and independent public SDK
observer are not included in this source checkpoint. No root verification
source is part of the product manifest. No John review rules or UI were added.
## Freeze2 classification correction
Post-freeze author review identified that the newly named orientation refusal
and translation bound were absent from the read route's known-refusal list.
Freeze2 adds precisely those two pure projection-validation errors. A public
Book::inspect test verifies both invalid projections invoke no effects and are
classified as known refusals; SDK errors, changing observations and uncertain
setter acknowledgment remain unclassified as known refusals. Exact freeze1
bytes/manifest are archived in evidence/freeze1-source. This is a narrow error
classification correction, not new placement support or live qualification.