The queue page loaded up to 200 exports in one table with the Blender
renders appended underneath, so a Blender job sat below the whole export
history. Exports and Blender renders are now separate tabs; the Blender tab
label shows how many are active or failed so a render stays visible from
the exports tab.
Exports page 25 at a time. The summary cards now come from a server-side
count per status (listExports.statusCounts) instead of counting the loaded
rows, which was already wrong past 200, and clicking a card filters to it.
The project filter moves to the header since it applies to both tabs, and
the duplicated Deliveries button is gone.
Adds .claude/launch.json for running the dev server from the preview pane.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Blender 5 logs through CLOG, prefixing every line with elapsed time and a
category ("00:02.954 render | Saved: '...'"), and reports errors as
"ERROR ..." instead of "Error: ...". The parser anchored on "Saved:" at the
start of the line, so it counted no frames from a render that wrote all of
them and the job failed as "saved no frames". Progress (Fra:) and error
detection were dead on 5.x for the same reason.
The prefix is now stripped before matching; 4.x output parses as before.
Warnings no longer count as errors: add-ons print them on every launch and
one would otherwise be reported as the reason a render failed. A run that
saves nothing now reports Blender's own error when there is one.
Verified on the real DH2 test.blend under Blender 5.2: the 1.6.2 parser
counts 0 of 3 saved frames, this one counts 3.
Bump worker to 1.6.3.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
LogTail was read by the progress loop while the output pumps appended to
it. Blender rewrites its status line many times a second, so the first
progress report of every job threw; aerender hit the same race only rarely.
LogTail now locks internally.
The exception also escaped BlenderRunner without ending the process, leaving
blender.exe rendering unsupervised on the node on every failed attempt. Any
unexpected exception in the monitor loop now kills the process tree.
Bump worker to 1.6.2.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
When the timing plate disagrees with the render, the worker now checks the
manifest's other plates[] that sit under a plate search root (or beside the
timing plate's footage folder). A match is a PASS naming that element; no
match still FAILs against the timing plate, listing the alternates tried.
Pre-renders are excluded because they agree with the render by construction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds a third section to the Batch Upload page that attaches plate movies to
shots derived from the filename: DH2_VFX_058_120_REF.mov becomes the plate
"REF" on shot DH2_VFX_058_120. Shot matching reuses findMatchingShotCode, so
it behaves like the rest of the page (case-insensitive, longest code wins,
underscore-boundary only) and the filename remainder becomes the label.
Bytes go straight to Hetzner via the existing presign endpoint since plates
are multi-GB, with XHR for per-file progress. Files whose shot already has a
plate of the same name are skipped so re-running a batch does not stack
duplicates.
Also resets the thumbnail and plate previews when the project changes — they
are resolved against a specific project, but only the version-upload flow was
being cleared.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A grid of twenty projects was twenty near-identical blocks of text. The cards
now lead with a still, with the code, client and title set over it.
There is no poster column on Project and adding one would mean somebody
maintaining it, so the image is chosen from the shots the project already has:
a key shot with a still first, then the most recently touched shot with one.
Within each, thumbnailUrl beats generatedThumbnailUrl - the same precedence
the rest of the app displays shots by, since the reference frame from the
brief is a deliberate choice and a promoted render frame is whatever the
pipeline landed on. The pick is deterministic rather than random: a card whose
image changes between two loads reads as a glitch, and the eye stops using the
image to find the project.
One DISTINCT ON query covers every project on the page, instead of a query
each or dragging back every shot that has a still.
Equal heights come from three things: auto-rows-fr on the grid, h-full on the
card, and fixed room for the description, which was the only part that varied.
A project with no stills gets a placeholder of the same footprint rather than
a hole in the row. The whole card is now the link, so the Open button is gone.
Also fixes a latent bug: the card read project.deadline, but the column is
dueDate, so the due date never rendered. Both are accepted now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Completes the operations dashboard.
QC queue: what is waiting on a person to watch the frames, oldest first.
Every stage before QC is worked by the farm, so this is the only one that
cannot drain itself - depth here is the best early warning that a delivery
date is going to move, and it had no home outside a filter on the queue page.
Deliveries: the last few packages, flagging FAILED and builds stuck in
BUILDING. A failed build is otherwise silent - the exports stay
READY_FOR_DELIVERY, nothing reaches the delivery volume, and nobody finds out
until the client asks where the shots are.
Whose court: versions sent to the client and unanswered, against shots with
unresolved notes. Two numbers rather than one because they call for opposite
responses - a fortnight-old review is a phone call, unactioned notes are a
scheduling problem.
The artist view now scopes to the artist throughout: the blocked list, and
the counters above it, which otherwise quoted a studio-wide figure next to a
personal list.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Turns "23 shots to go" into "you are not going to make it". One card per
episode from EpisodeDueDate (or the project's own date on a standard show):
days left, the episode's shots by where they actually are, and the only
comparison that matters - shots per day still needed against shots per day
the studio managed over the last week.
A shot counts as delivered when it appears in a DeliveryItem inside a
DELIVERED EXR_DELIVERY package. Shot.status is deliberately not the measure:
it says where the work is, not whether the client has the frames, and a
MOV_REVIEW package ships nothing.
The pace arithmetic is a pure function so the sentence a producer acts on is
testable: a rate that only just lands on the due date reports as "no slack"
rather than fine, and a week with no deliveries reports as stalled instead of
projecting infinity.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The dashboard was a generic task-manager view of a studio that does not work
that way. Two of its five headline numbers counted Task.dueDate, which no
pipeline code writes - the only writer is a manual field on the task detail
page, so those cards read zero by construction. The rest were unscoped and
cumulative ("approved" across every project ever finished), the activity feed
was the notification table, and the shot queue sampled 15 rows by updatedAt,
a column the farm touches constantly. Nothing about failed exports, QC depth
or the render farm appeared at all, so a RENDER_FAILED was only visible if
someone thought to open the queue and filter for it.
Replaced with pipeline state, which the system maintains on its own:
- Blocked queue: the four failed statuses plus exports that stopped
progressing, each with the reason read off the actual cause - the failing
validation check, the QC note, the aerender error, or the machine that
claimed it and went quiet.
- Farm strip: machines, queue depth by job type, oldest unclaimed job, and
worker-version drift, since a worker's version decides which fixes it has.
- Five counters that can go down, scoped to live productions, each linking to
the rows it counted.
The queue page ignored ?status=, so the counters had nowhere to link; it now
takes an initial filter from the URL. Stat cards are hidden from CLIENT users
- the old ones were generic version counts, these are internal pipeline state.
Thresholds live in SystemConfig (dashboard.qcStaleHours, .stalledJobHours) so
they can be tuned without a redeploy.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The version is what a machine reports on startup, so it has to move for the
Machines page to show which nodes are still clearing the preview MOV out of
the version folder.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
PrepareOutputDir cleared every file in the output dir whose name started
with the output pattern's prefix. The preview stage writes its MOV and MP4
into that same folder from the same basename ({outputDir}/{base}.mov), so
they matched the prefix and were deleted before aerender started.
The dir is not private to one export: outputDir is keyed on the shot's
manually-set shotVersion, not on the export, so re-queuing without a
Version Up renders into the folder again. That is exactly the "No preview"
workflow - export scaled, reverse the scale, re-render only the EXRs and
inherit the earlier MOV - and inheritPreviewMedia then handed the new
export a deliveryMovPath pointing at the file the render had just removed.
Nothing surfaced it: the review MP4 is an uploaded blob and still played,
and DeliveryStage treats a missing MOV as non-fatal, so the package
silently shipped EXRs with no viewing copy.
Match on prefix and extension instead, via a new
SequencePattern.MatchesOutputFile - the predicate ValidationStage and
DeliveryStage already applied inline. Crashed-attempt cleanup is unchanged;
only files with a different extension are now spared.
JobExecutor's post-render stats loop matched on prefix alone too. With the
media no longer deleted it would have counted the MOV/MP4 as frames and
failed validation's file_count check, so it uses the same matcher.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Numbering was scoped to the project, so the day's first delivery of
episode 102 was named _02 purely because episode 101 had already gone out
that morning. The count is meant to distinguish repeat deliveries of the
same content, and packageDir already separates episodes — three episodes
delivering today each land in their own folder and each are that folder's
first delivery. Adding `episode` to the lookup makes the suffix mean what
the folder layout already implies: only a second delivery of the *same*
episode becomes _02. The [projectId, episode] index already exists, so
the extra filter costs nothing.
packageNameTemplate also moves from {yymmdd} to {yyyymmdd}. The AE panel
has always built its local delivery folders with a four-digit year, so
this is the server matching the connector rather than a new convention.
The {yymmdd} token stays in TokenValues: a project carrying its own
template in deliveryConfig is unaffected either way.
Existing packages are untouched — buildDeliveryJobManifest reads rootPath
off the record. On the day of the changeover an episode that already
delivered as 260903_Delivery starts again at the bare 20260903_Delivery,
which is a different name and a different folder, so nothing merges.
Producer reviews keep their per-project numbering: a review can span
several episodes and stores episode as "", so scoping it the same way
would put every such review in one bucket.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds an "Assign Artist" entry to the bulk actions menu, backed by a
bulkAssignArtist server action. The action re-checks the target user
server-side — a deactivated account or a CLIENT is rejected rather than
trusted from the picker — and passing null clears the assignment, so
reassignment and hand-back are the same flow.
Like single-shot assignment in updateShot this writes the column and
sends no notification, so reshuffling a batch during planning stays quiet.
Also renders the Artist column the page was already querying but never
showing; without it there was no way to see the result of an assignment.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A shot is reviewed at the framing it was rendered at but usually delivered
cropped, so add a dropdown beside the playback-speed control that lays
delivery blanking over the video: 2.39 through 9:16, widest to tallest, with
"No blanking" as the default. Purely a viewing aid — nothing is written to
the review, and the video element is untouched.
The geometry is the awkward part. The video is object-contain, so the picture
is letterboxed inside the element and bars drawn against the element's edges
would be wrong on any panel whose shape does not match the source.
letterboxRect works out where the picture is actually painted and the bars are
positioned against that, kept current by a ResizeObserver and a loadedmetadata
listener so resizes and fullscreen stay correct. Ratios within half a pixel of
the source draw nothing rather than a hairline of rounding error.
Two modes, since blanking and crop bars are different jobs: solid black hides
the crop the way the master will, guide dims it so the excluded area is still
readable. The overlay sits under the annotation canvas — deliberately no
z-index, since the canvas sets none either and DOM order alone gives marks
drawn over a blanked area the right precedence.
Verified against real browser layout, not just unit tests: the computed
picture rect lands exactly on the image edges across three panel shapes, and
the visible area measures the requested ratio.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two faults in how a package's folder is named, both of which put unrelated
files in one directory on the delivery volume.
The default layout had no {showId}, so every show's episode 101 resolved to
{deliveryRoot}/101/{packageName}. The token existed and was populated — it
just never appeared in the path. With hardlinked packages that means two
shows delivering the same episode number on the same day write into the same
folder, and identically named shots overwrite each other. reviewDir had the
same hole, and is worse there: reviewNameFor scopes its lookup by projectId,
so a second show computes the bare base name and lands on the first show's
folder instead of suffixing past it.
createDeliveryPackage also never checked for an existing package of the same
name. packageNameTemplate is date-only and packageNumber is a per-project
counter that is not in the template, so a second delivery on the same day got
an identical name and rootPath and built on top of the first. MOV reviews
already did this correctly, so nextReviewName moves to deliveries.ts as
nextPackageName and both paths share it; the review-side name is kept as an
alias because it reads better at that call site.
Putting {showId} ahead of {episode} exposes an empty segment on non-episodic
shows, so paths go through tidyPath, which collapses it and leaves the UNC
root alone.
Historic packages are unaffected: buildDeliveryJobManifest reads rootPath and
configSnapshot off the record, so only new packages use the new layout. A
project with its own deliveryConfig.layout in the database still overrides
these defaults.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
loadQCQueue called /api/ext/qc/queue with no limit, so it always got the
endpoint's default page of 50. The query orders oldest-first, so a project
with more than 50 exports in READY_FOR_QC showed only the oldest 50 with
nothing to say the list was short.
Ask for 200 — the server's hard ceiling in getQCQueue — and say so in the
status line when the count comes back at the cap, so a truncated queue is
visible rather than silent.
The diagnostics call in checkServerPathReachable stays on the default: it
only reads entries[0].
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
app.scheduleTask outlives the panel instance that armed it. The repeating task
id was held in the pipelineState closure, so closing or reloading the panel took
the id with it and left nothing able to cancel the task: it kept firing every
15 s for the rest of the AE session, and every tick landing while a modal dialog
was open raised "Cannot run a script while a modal dialog is waiting for
response."
The id now lives in global scope beside the tick function that scheduleTask
already evaluates there, so the next panel instance can still reach it.
getPollTaskId and setPollTaskId wrap the access and treat anything that is not a
positive number as no task. Panel startup cancels whatever a previous instance
left armed before pointing the tick at itself, rather than letting the old task
fire into a fresh, empty state whose shot code went with the closure that armed
it, and panel.onClose stops the poll on the way out.
The export is unaffected either way: it runs on the farm, and the header picks
its status back up from the server on the next Refresh. Section 12.1 of
RenderPipeline2.md records that.
PANEL_VERSION goes to 2.6.1 per the documented rule, so Pipeline → Machines does
not show two workstations on one version behaving differently.
This change was authored in the working tree, not by Claude; it is committed
here alongside the status and client-portal work at the author's request, and
the message was written from reading the diff. It has not been exercised in a
live After Effects session as part of this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Notes left through a review link were write-once. A client who mistyped a
timecode or changed their mind had no way to correct it and had to add a second
note contradicting the first, which the studio then had to reconcile by hand.
PATCH and DELETE on comment/[commentId] cover both. Only the reviewer's own
notes are touchable: the studio's comments and those of every other review
session are read-only through a link, whatever the holder sends. Ownership is
re-checked server-side on every write, so the canEdit flag now returned by the
versions route only decides whether to offer the controls and grants nothing.
Refusals are 404 rather than 403 throughout, since a link holder has no business
learning which comment ids exist outside their own notes.
Only the text is editable. The frame a note is pinned to is what the studio
works from, so moving it would silently repoint feedback someone may already
have actioned; a note about a different frame is a new note. Deleting a drawing
note also removes its strokes, mirroring deleteComment in actions/comments.ts,
and the page reloads the player afterwards so the frame does not keep showing
marks that no longer exist. Delete goes through a confirmation that shows the
note and says plainly that it disappears for the studio too.
Worth knowing before this reaches a show: everyone holding the same review link
is the same author. A ReviewSession with no email resolves to a token-derived
guest user, so one link sent to three people gives all three one identity and
any of them can now edit or delete the others' notes. That model is not new,
which is why their comments already showed as one name, but edit and delete is
the first feature where it grants real power rather than muddling attribution.
Per-reviewer links with email set are the fix if that is not wanted, and that is
a change to how links are issued rather than to this code. The scope is
documented on reviewClientEmail so the next reader meets it before it surprises
them.
getOrCreateClientUser was copy-pasted into the comment and approve routes with
the fallback-email expression spelled out in three places. It is one set of
helpers in review-auth.ts now, including findReviewUser, a read-only variant so
that merely opening a review link does not conjure a User row just to answer
"is this theirs?", and ownsComment, the single rule shared by the route that
offers the controls and the route that enforces them.
Nine tests cover that rule directly: internal users' notes stay read-only, a
session that has never posted owns nothing, and an authorless comment — one
whose User was deleted, since Comment.authorId is SetNull — cannot be claimed by
anyone. The routes themselves are not covered, so exercise edit and delete
against a real review link before handing one to a client.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Requesting changes on a shot and then re-rendering it through the render queue
left every layer of the app still showing the old verdict. Three independent
causes, all on the pipeline path.
finalizePreview inherited the superseded version's approvalStatus
unconditionally. The carry-over exists so a re-export of a signed-off shot does
not reappear as awaiting review, but it applied equally to NEEDS_CHANGES and
REJECTED, so the render queued to address a note arrived already wearing that
note and nothing registered that the work had been redone. carryApproval now
carries only APPROVED.
The pipeline never moved the task. One left on CHANGES by a review or a QC fail
stayed there through every re-render, with no way out but a manual kanban drag.
A genuinely new version now moves it to INTERNAL_REVIEW, as actions/versions.ts
and batch-upload already did. A replaced version is excluded: regeneratePreview
re-delivers the same version after a slate fix and is not a new submission.
The pipeline never called recalcShotStatus at all, so shot boards went stale the
moment a render changed a task. finalizePreview and the QC fail path both
recalculate now; the latter already documented the REVISIONS derivation it was
relying on and never triggering.
Shot.status is derived from tasks and approval flags, but six call sites wrote
it directly with the status they expected: internallyApproveShot,
shareWithClient, unshareFromClient, clientApproveShot, unapproveShot and the
client approve route. Rule 1 of the derivation is that any CHANGES task outranks
an approval, so a shot could be pushed past open rework and then snap back to
REVISIONS on the next unrelated recalc, which is close to unreportable as a bug.
They now set the approval inputs and let recalcShotStatus derive.
unshareFromClient additionally hardcoded READY_FOR_CLIENT even for a
client-approved shot. updateShotStatus keeps its direct write as the deliberate
manual override, now documented as transient for the same reason.
markShotForRevisions replaces the duplicated change-request logic. Both copies
marked all non-DONE tasks CHANGES, which matches nothing on the shot this is
most often called for: one out for client review precisely because all its tasks
are finished. The request then derived to INTERNAL_REVIEW and had no visible
effect anywhere. When nothing is picked up it now reopens the task behind the
shot's latest version, which is the deliverable the reviewer was looking at.
A client's shot-level change request also marks the versions and writes Approval
rows now. The dashboard counters, the version badges and the portal's own
filters all read Version.approvalStatus, so a shot-level verdict was invisible
everywhere those look.
clearStaleShotApproval covers the reverse case: uploading a new version onto an
approved shot moved the task to INTERNAL_REVIEW and changed nothing visible,
because the derivation short-circuits on shotApprovalStatus. The shot kept
reading COMPLETE while carrying work nobody had looked at. This is the one
user-visible behaviour change here, so watch the boards on a live show: an
approved shot now drops back to INTERNAL_REVIEW on a new upload. sharedWithClient
is deliberately left alone, since revoking client access stays a separate act.
submitApproval and unapproveVersion are transactional now. The verdict, the
version, the task and the derived shot status are one decision, and a failure
partway through left the three layers disagreeing with no way to tell which one
was right.
KanbanCard compared approvalStatus against "CHANGES_REQUESTED", which is not a
member of the enum, so the changes-requested icon never rendered at all.
Thirteen tests cover carryApproval, markShotForRevisions including the all-DONE
regression, and clearStaleShotApproval. What is not covered is the pipeline
transaction itself, which needs a database: queue a render against a shot in
REVISIONS and confirm the task and shot both move before trusting this on a
show.
Version.reviewStatus is still written nowhere while being published through
ext/shots/lookup and EXT_API_REFERENCE.md as a permanent "PENDING". Left alone
deliberately: removing a field from that contract could break the connector.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Build Shot produced comps starting at 00:00:00:00 because the panel never
actually read the EXR timecode. getFootageStartFrames ranked a fragile "New
Comp from Selection" probe first: it toggles the Project panel via the Window
menu to get focus, and where that does not land the menu item is greyed out,
executeCommand silently does nothing, and the probe returns null. A shot with
no timecodeStart then fell through the API rank to zero.
The worker already reads EXR headers properly in ExrHeader.cs, and its
validation ranks the plate's own timeCode above the API value, which stores the
editorial in-point and sits a handle-length ahead of the plate by design. Shots
were therefore being built from rank 3 and validated against rank 1. That is a
timecode desync waiting to reach a delivery, not just a cosmetic start frame.
readExrTimecode ports the worker's reader to ExtendScript and becomes rank 1:
one short header read, walking `name\0 type\0 int32 size data` for timeCode and
falling back to timeCodeString where a writer omitted the packed attribute.
decodePackedTimecode mirrors Timecode.FromPacked, decoding the SMPTE 12M BCD a
byte at a time so the arithmetic never touches the sign bit. The menu probe is
kept as rank 2 and now runs only when the plate carries no timecode at all, so
the Project-panel toggle is off the normal path.
setCompStartFrame was the second failure. Assigning displayStartFrame on a
version of AE that does not have the property is ignored silently rather than
throwing, so the early return skipped the displayStartTime fallback entirely.
It now reads the value back, falls through when it did not take, and returns
whether anything stuck; buildShotFromCode logs when nothing does. A start frame
that cannot be set is now visible in the diagnostics report instead of a comp
quietly sitting at zero.
A plate whose timecode is present but zero still falls through to the API rank,
keeping the existing frames > 0 convention so the safety net survives. If a
genuine 00:00:00:00 plate should win outright that is a deliberate follow-up.
The parser is covered against synthetic headers, including the 0x18585712
example documented in ExrHeader.cs, drop-frame, truncated and non-EXR input.
What is not yet exercised in a live After Effects session is mainSource.file
resolving to the sequence's first frame and displayStartFrame taking on a given
AE build, so build one shot and check its start against the plate before
relying on this. Picture-lock automation was reverted for a timecode desync of
the same shape.
PANEL_VERSION goes to 2.6.0 per the documented rule, so Pipeline → Machines
does not show two workstations on one version behaving differently. The
startSource comment in the manifest schema enumerated only the two older
values; the queue-time provenance check keys on /\bAPI\b/ and so already reads
"EXR timecode …" as trustworthy rather than suspect.
The same silent-failure setCompStartFrame still exists in the farm-side
vfxr_build_preview.jsx, which is fed the right value from the manifest but can
drop it the same way. Left for its own change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A shot with no COMP task rendered and previewed fine and then showed
nothing in the app. createExport set Export.taskId from the shot's first
COMP task or null, and finalizePreview registered the review Version with
that same null taskId — so the MP4 was uploaded and stored, but the
Version hung off no task and none of the review flows had anywhere to
draw it. From the artist's side the export completed and produced
nothing.
E1 now resolves the shot's comp task inside the version-allocation
transaction and creates one when there isn't any, titled after the
preview file the export will produce: {shotCode}_cmp_TT_{version}, the
same convention the manual batch upload's create-task path already uses.
renderOutputBase() is exported and buildPreviewManifest takes its `base`
from it, so the task title and the MOV/MP4 name come from one function
rather than two copies of the same template. The title therefore already
carries the version, which makes finalizePreview's rename-to-the-
delivered-version a no-op on that first export rather than an immediate
rename of a task created moments earlier.
The lookup and the create sit inside the transaction so the P2002
version-collision retry rolls a losing attempt's task back with
everything else, instead of leaving a duplicate behind.
recalcShotStatus runs only when a task was actually created: a shot with
no tasks derives WAITING and now has one. Approval-driven statuses still
take precedence, so an already-approved shot stays COMPLETE.
Both the queue-time ExportEvent note and the E1 response record what
happened, so a task appearing under a shot on its own is traceable to the
export that made it.
Exports already in flight with a null taskId keep their orphaned
versions; backfilling those is a separate migration.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two small board additions.
The shot version now shows next to the shot code. It is Shot.shotVersion —
the value the AE panel burns into overlays and queueExport renders under,
and the one the shot detail page edits — not a count of uploaded review
Versions, so the board agrees with what the artist has in the panel.
Shot.isScaled records that a shot is scaled in the edit: the plate is not
used at its native size, which changes how the shot is comped and
reviewed. That makes it a standing property of the shot rather than
something buried in the notes, so it is flagged the way isKeyShot is — a
cyan pill next to the shot name, and a Scaled toggle column beside Key.
toggleScaledShot() (actions/shots.ts) mirrors toggleKeyShot, same roles.
KeyShotToggle's switch markup would have been duplicated verbatim for the
second flag, so both now sit on a shared FlagToggle taking a colour tone
and a setter. The key-shot switch behaves exactly as before, optimistic
update and rollback included.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A formal review is shared as a flat, dated folder of the preview MOVs —
{deliveryRoot}/PRODUCER_REVIEWS/20260828_Producer_review, then _02, _03
for further reviews the same day. It reuses the delivery package's
machinery end to end: DeliveryPackage/DeliveryItem, the DELIVERY_BUILD
job, the same claim path and the same hardlink-or-copy builder, with
DeliveryPackage.kind telling the two apart.
Three things separate a review from a delivery:
A review never advances an Export. Reviews happen throughout a shot's
life, long before QC passes it, and the same version is routinely shown
twice — so nothing is marked PACKAGED, and markDelivered() refuses a
review package outright (the button is hidden too).
Selection is by shot, not by export. Shot Status → Bulk Actions → Build
MOV Review Package; the server resolves each shot to its newest export
that actually has a deliveryMovPath, which is not always its newest
export — a shot whose latest version is still rendering belongs in the
review as the version that can be watched. Shots with no MOV at all are
listed in the confirm dialog with the reason rather than silently
shrinking the folder.
The folder is exactly the MOVs. No CSV, no manifest.json written beside
them; the manifest still returns to the server, so the record of what was
shared is kept without cluttering the share. Names follow the same
client-naming rule as a delivery, so a shot reads identically in a review
folder and in the delivery that follows it.
Verification (DeliveryStage.BuildReviewItem) is proportional as it is for
EXRs, adapted to a single file with no validation-time digest: a hard
link is the source and needs nothing, a copy is checked against its
source by length always and by full hash under "always-hash". A missing
MOV fails the item here — in an EXR delivery the MOV is a viewing copy
beside the real deliverable, but in a review it is the deliverable, and a
review folder quietly one shot short is worse than one that says so.
Worker bumped to 1.6.0 for the new manifest field (DeliveryManifest.Kind).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Shot.clientCode already drove slates, overlay burn-ins and delivered file
names, but the client portal — the one place the client actually reads — was
still showing our internal codes.
Task titles were the larger leak. They follow the house convention
"UNG_106_010_020_cmp_TT_v004", so the internal code sat embedded in the
string the client reads and copies, in three places on the shot list and
again in the player header. Swapping only the heading would have left it on
display directly underneath.
lib/shot-code.ts resolves both: clientShotCode picks the alias when set, and
clientTaskTitle swaps the code out of a title by substring, leaving a
hand-edited title that does not carry the code untouched rather than guessing
at it.
Applied at the API boundary rather than at each render site, so the shot
list, search, playlist labels, alt text, player header and every
copy-to-clipboard button inherit it and nothing downstream has to remember
which code it is holding. The internal code is no longer sent to the browser
on these routes at all.
Slack notifications deliberately keep the internal code: those go to our
artists, who need the code that matches the comp and the SAN path.
The two existing "clientCode || shotCode" expressions in preview.ts and
deliveries.ts now use the same helper so slates, deliveries and the portal
cannot drift apart. Shots with no clientCode are unchanged throughout.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ffmpeg was started with stdout and stderr redirected and neither pipe ever
read. Windows gives an anonymous pipe a ~4 KB buffer, which ffmpeg's banner
and stream summary exceed on their own, so it blocked forever on the write,
never exited, and the two-minute cap fired:
TaskCanceledException at Process.WaitForExitAsync
Preview complete for job ... in 221s (mov + mp4)
The exception was swallowed as "non-fatal", so every preview silently
produced no thumbnail while paying 120s for the privilege. Nothing
downstream was wrong — there was simply never a thumbnail to promote.
Both pipes are now drained concurrently before the wait. A timeout kills the
process tree rather than leaking an ffmpeg per job, and attempts report a
three-way outcome so a hung or unstartable ffmpeg is not retried — without
that, the slate fallback would have doubled the wasted time to four minutes.
Failures are no longer silent: an unset or missing ffmpegPath, the seek that
was chosen, and ffmpeg's own exit code, command and stderr are all logged.
finalizePreview likewise records every promotion outcome on the export
timeline, not just success. Silence there was indistinguishable from a broken
worker, a project whose flag never saved, and the feature working as
configured — which is what made this take so long to find.
Adds scripts/diagnose-thumbnails.mjs, which walks the chain (ffmpeg →
Export.thumbnailKey → project flag → Shot.generatedThumbnailUrl → what the UI
resolves) and names the first broken link.
Worker 1.5.1 -> 1.5.2; republish required for the fix to take effect.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two independent features landing together:
Generated shot thumbnails — a shot can now carry a thumbnail promoted
from a render export (Shot.generatedThumbnailUrl/generatedThumbnailFrom)
alongside the existing brief-supplied thumbnailUrl, which always takes
display precedence (lib/shot-thumbnail.ts) so a promoted frame can only
ever fill a gap, never overwrite the brief. Shows with
pipelineConfig.autoShotThumbnail promote automatically on every
finished export (lib/render-pipeline/preview.ts); promoteExportThumbnail/
clearGeneratedThumbnail (actions/shots.ts) are the manual equivalent for
everyone else. Worker-side seek logic for picking the promoted frame is
covered by RenderWorker/VFXReviewWorker.Tests/ThumbnailSeekTests.cs.
Bulk shot delete — the shot-status page's selection bar gets a "Bulk
Actions" dropdown (alongside the existing Change Due Date) with Delete
Shots. bulkDeleteShots() (actions/shots.ts) deletes one at a time rather
than in a single statement: exports_shotId_fkey is ON DELETE RESTRICT
by design, so a shot with renders or deliveries on record is protected
and gets skipped and reported by name rather than aborting the whole
selection or being silently destroyed along with its history.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Clients sometimes ship footage under a naming convention that doesn't
match our own shot codes. Adds an optional Shot.clientCode, editable on
the Slate tab: when set, it replaces shotCode only in what the client
actually sees — slate identity, overlay burn-in, and delivered EXR/MOV
file and folder names in Netflix packages — with our own version string
flatly appended (no "_cmp_TT_" studio jargon mixed in). Internal render
paths, EXR filenames on the SAN, comp names, and DB lookups are
untouched. Shots without a client code render identically to before.
Delivery packages denormalize clientCode at build time, same as the
existing shotCode/versionString snapshot, so a later shot edit can't
retroactively change already-built delivery history. The worker now
renames delivered EXR frames and the delivery MOV using the existing
frame-number parser (SequencePattern) so only the base name changes —
frame numbers and padding survive exactly. This is the one piece of the
change that could plausibly ship the wrong bytes, so it's covered by
new unit tests (renaming, padding, MOV rename, checksum-survives-rename,
and a graceful fallback for an unparseable frame number) rather than
resting on manual verification alone.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Build Shot adds the show LUT and overlay burn-in into the artist's main
comp, enabled by default, for reference while compositing. The legacy
manual "Queue EXR" button has always disabled both before rendering;
Queue Export (the pipeline's one-click path) never got that same step,
so EXRs queued through it could ship with the LUT graded in and the
burn-in baked on. Queue Export now disables them the same way, right
before the project save the render worker reads from.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Production fallout after enabling it: the trimmed EXR footage started
after the head handle as intended, but the overlay's burn-in timecode
(hardcoded to start at frame 1001) did not shift with it, so the
delivered burn-in no longer matched the footage. The picture-lock
reference render also never worked correctly in practice. Reverts the
AE panel checkbox, the preview-build script's trim/reference logic, and
the server/worker plumbing around it back to their pre-feature state
(82dfaee) — untrimmed previews, no picture-lock reference, overlay
timecode correct again.
The noPictureLock/piclockRefKey schema columns and their migration are
deliberately left in place: they're already applied in production, and
dropping an applied migration's file risks the next `prisma migrate
deploy`. They sit unused and harmless until this is revisited.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Manipulating startTime/outPoint on previewComp's SHOT layer touched a
comp duplicated from the artist-authored template, whose slate/shot
timing relationship the build script doesn't own -- and it dropped the
slate frame in production. The trim now happens one level down, on the
private single-layer footage precomp this script fully controls;
previewComp's own sizing logic is untouched from before this feature
existed, since it already generalizes correctly to any shotComp
duration.
Also threads the picture-lock skip reason through to the export's
timeline, so a shot missing sequence timecodes or a matching picture
lock file is visible without digging into worker-machine logs.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Previews used to ship the full padded (handle-included) render to
clients, with an artist manually pulling picture lock and hand-trimming
the comp to match. The automated preview stage now trims the client
comp to the picture-lock edit itself (a fixed handle count off each
end, per the studio's existing convention) and separately builds an
ungraded, full-duration picture-lock reference MP4 as a diagnostic
sidecar to check back against. A new "No Picture Lock" panel checkbox
is the single override for shots where picture lock is missing or
wrong, falling back to today's untrimmed behavior.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A FAILED delivery item never sets fileCount/totalBytes, and the worker's
Dictionary-based JSON payload serializes those as explicit nulls (the
WhenWritingNull option only suppresses nulls on POCO properties, not
dictionary values). The server's schema allowed the fields to be omitted
but not null, so the report was permanently rejected with 422 - and the
worker only ever treated 409/404 as terminal, so it retried the same
malformed payload forever across restarts.
Removes the three spots on the client review shot-list page that showed
a version's notes under the task title (flat task list, shot-grouped
list, asset list) — one of the three was already commented out by hand,
uncommitted, as a stopgap for the same issue.
The API route backing this page no longer selects notes at all, so it
stops being sent to the client rather than only going unrendered — the
same treatment as removing the pipeline's internal-only notes text in
53fc8dc.
The individual review player (app/client/[token]/review/[versionId]/)
has its own separate notes field and fetch and is untouched: this was
scoped to the shot-list overview only, per what was asked.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A shot is routinely re-exported for delivery after it has already been
client-approved — most often an unscaled/un-punched-in pass. Every
pipeline-created Version defaulted to approvalStatus PENDING_REVIEW,
which is exactly what the dashboard counts, the shot page badge and the
playlist read as "awaiting review" — so a re-export made an approved
shot look like new unreviewed work to the whole team, with nothing
actually different about its approval.
approvalStatus is now inherited from the task's previous latest version
rather than left at the create default. A task with no prior version
(its first export) still gets PENDING_REVIEW, which is correct.
Separately, every pipeline version was stamped with notes: "Pipeline
render vXXX (internal only)" — and notes is the artist's message to
reviewers, rendered verbatim in the client portal and the review player.
Nothing enforces that a pipeline version stays unshared once created (see
the gap noted in RenderPipeline2.md), so that internal-sounding string
could reach a client if the shot were re-shared after a re-export. notes
is left unset; what a version is and where it came from belongs on the
Export and its timeline, which is where a human would actually go
looking.
RenderPipeline2.md §9 documents both, plus the share-flow gap that let
notes reach a client in the first place: shareWithClient promotes
whichever Version is isLatest with no awareness that a pipeline version
is meant to stay internal permanently. Not fixed here — closing it needs
a way to identify a pipeline-origin version at the share call sites, and
Export.versionId is currently a one-way scalar with no inverse relation
on Version.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Blender renders failed with "produced no output for 600s" on every
attempt while rendering perfectly well.
Two causes, both mine. BeginOutputReadLine raises an event per newline,
and Blender writes its render status with a carriage return so it
overwrites one line in place — a frame taking longer than the stall
timeout emits no newline for its whole duration. stdout is now read as a
character stream with CR and LF both ending a line.
The watchdog was also stricter than its own message: it only accepted a
recognised Fra:/Saved: line as a sign of life, so any output we were not
parsing counted as silence. Any output now proves Blender is alive, which
is what "produced no output" is asking; finished frames still drive
progress, since a single frame emits hundreds of status updates.
Also passes -noaudio: a render node has no sound device, and initialising
audio under a session-less service account is a known way to hang.
Covered by tests over a realistic stdout chunk where progress is
CR-separated and only the saved lines end in LF — newline-only splitting
yields one unusable blob and no progress at all.
Worker 1.5.1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The project dropdown rendered garbled labels. EnumProperty does not copy
the strings a callback returns — Python has to keep a reference or
Blender reads freed memory, which shows as mojibake rather than as a
crash. The callback built fresh tuples on every redraw, so each set was
collected the moment it returned.
The items are now built once when the project list changes and held at
module level; the callback returns that list and allocates nothing, which
also makes a per-redraw callback free rather than merely correct.
Add-on 1.0.2.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Submitting failed with a 403 before the request reached the server:
Cloudflare error 1010, a browser-signature ban. Python's urllib
identifies itself as "Python-urllib/3.x", which is exactly the scripted
signature that rule targets. The AE panel was unaffected because curl
sends curl/8.x, and the worker because .NET sends its own.
Verified against the live zone, same request and token, only the agent
changed: Python-urllib/3.11 gets 403 while VFXReviewBlender/1.0.1 gets
200 — as do curl and a browser agent.
Naming the tool is also the honest thing: the studio's logs should say
what is calling them, and the version travels with it.
A Cloudflare block is also no longer reported to the artist as a wall of
JSON about browser signatures — it says the firewall stopped the request
before it reached the server, and to ask the pipeline TD.
Add-on 1.0.1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Blender work is not a shot submission: no Export, no version, no
validation, QC or delivery. It is "Ctrl+F12, but on the farm", so it is
modelled as what it is — a queue entry and nothing more.
That turned out to need very little. Claim, progress, fail and
complete-render already tolerated a job with no Export behind them, so a
Blender job reuses the whole lifecycle; RenderJob gains projectId, label
and submittedBy so a job with no shot code still has something to show.
Capability gating is new and was already missing: the claim now filters
job types by the machine's reported tools. Without it a workstation
without Blender would claim Blender work and fail it repeatedly, burning
the attempts while a machine that could have rendered it sat idle. The
worker asks for Blender jobs unconditionally and the server refuses them
— trusting the worker to self-select would have left the same hole.
The runner treats a clean exit with no saved frames as a failure:
Blender exits 0 when a frame range is empty or an output path is
unwritable, and a job that silently rendered nothing is the worst
outcome available.
Standalone jobs auto-requeue on a transient failure like any other work;
delivery builds are deliberately excluded, since a half-written package
wants a human first.
Adds the Blender add-on (BlenderPanel/) and its docs. The add-on refuses
to submit an unsaved file — the farm opens the .blend from disk, so
unsaved changes would render the previous state without warning. Its
payload is pinned by a contract test, since add-on and server only ever
meet over JSON.
Worker 1.5.0. Adds migration 20260815000000_blender_render_jobs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Initialise Workspace built UNG_101..UNG_117 from a hardcoded range and
imported one hardcoded AEP, and fourteen findComp calls named UNG's comps
literally. None of it worked for a second show.
Comp names now follow a convention derived from the show id —
{showId}_VFX_OVERLAY, {showId}_EXPORT_TEMPLATE, _SHOW LUT — which
reproduces UNG's existing names exactly, so nothing changes for the show
it was written for. A project whose template departs from the pattern can
override any of the three.
Paths cannot be derived: UNG's show id is UNG while its folder is
UNGO_VFX. Those live on Project.pipelineConfig and are edited under
Project → Settings → Pipeline. They are resolved server-side and handed
to the panel with the project list, and read by buildPreviewManifest for
the headless build, so the panel and the farm cannot disagree about which
template — or which show LUT — a shot was built with. That mattered more
than the folder names: a preview rendered through another show's LUT is
quietly the wrong colour rather than obviously broken.
Episode folders come from the project's episode list, falling back to the
folders under its footage root for a show whose episodes are not entered
yet.
Panel 2.5.0. Adds migration 20260814010000_project_pipeline_config.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Shot.fps existed, the lookup API returned it, and one button honoured it
— Set Start Timecode — while the shot builder ignored it entirely. Comps
were created at 24 and every plate conformed to 24.
DH2 is a 25 fps show. Its plates carry framesPerSecond 25/1, so comps
built at 24 conform every plate to the wrong speed and compute every
frame-from-timecode sum against the wrong rate.
The builder now takes the rate from the shot record, which is the show's
delivery rate and therefore the authority; a plate's own rate is whatever
happens to be on disk. Import Renders follows the comp it imports into,
and Import EXR follows the shot.
Validation gains plate_fps, comparing the render against the timing
plate. timecode_fps only compares the render to its manifest and both
come from the comp, so a shot built at the wrong rate agrees with itself
and passes; the plate is the independent witness. It is a WARN, never a
FAIL: sampling DH2 found one plate at 50 fps against a 25 fps show, a
deliberate over-crank, so a whole-number multiple is reported as exactly
that and anything else as a probable rate mistake.
Note this does not fix the data: all 175 DH2 shots still carry fps 24
from the create-time default and need correcting before their comps are
rebuilt.
Panel 2.4.0, worker 1.4.0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Overriding a failed validation queued the preview at priority 50
regardless. It took its rank from the export's last PREVIEW_ONLY job —
but validation runs before the preview stage, so a VALIDATION_FAILED
export has never had one, and the missing row fell through to a
hardcoded normal priority. An urgent shot lost its urgency at exactly
the moment someone was standing over it waiting.
Every preview now derives its rank from the export's AE_RENDER priority
rather than from a previous preview. That fixes the missing-row case and
also removes a slow drift: a preview's priority is already boosted, so
re-deriving from it moved an export a step up the queue on every retry.
createPreviewJob takes the render priority and the attempt number, so the
first build, a retry, a regenerate and an override all rank identically.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Urgent already propagated — a preview inherits its render's priority, so
an urgent export stays ahead all the way to QC. The complaint was the
other half: at equal priority previews always ran last.
The queue orders by priority then age, and a preview job is only created
when its render finishes, so it is always newer than the renders queued
beside it. Ten shots ran as ten renders and then ten previews, and
nothing was reviewable until the end of the batch.
render.previewPriorityBoost (default 1) subtracts from the render's
priority when the preview is created, so a shot finishes end to end
before the next render starts. The boost never promotes a job across the
urgent threshold: a normal preview must not inherit permission to run
outside the render window that its render never had. Set it to 0 for the
old behaviour.
Worth knowing separately: a worker will not claim preview jobs at all
while an artist has AE open on that machine, which during the day is the
larger reason previews appear to wait.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Some exports are deliberately framed differently from their review —
unscaled or un-punched-in EXRs — and a preview rebuilt from those frames
would be the wrong picture. The previous export's MOV is the correct one.
skipPreview renders the EXRs and goes straight to READY_FOR_QC. Available
at Queue Export (a "No preview" checkbox in the panel, E1 field) and as an
option when overriding a validation failure.
The inheritance is as important as the skip: deliveryMovPath, the review
MP4, the thumbnail and the Version are carried over from the newest
earlier export of that shot which has them. Without it the delivery
builder — which takes its MOV from deliveryMovPath — would package EXRs
with no MOV at all. Where there is nothing to inherit, the export still
reaches QC and the timeline says it has no review media.
QC is untouched by any of this: a skipped preview still needs a human
pass before the export can be delivered.
Panel 2.3.0. Adds migration 20260814000000_export_skip_preview.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Checks fail on renders that are genuinely fine — a plate the engine
cannot reach, an expectation that does not apply to this shot — and a
VALIDATION_FAILED export was a dead end, blocking delivery on a shot a
human can see is correct.
Override validation on the Export detail page resumes the pipeline where
a pass would have handed off: preview, then READY_FOR_QC. It never skips
QC and never delivers directly, so the frames are still looked at — the
override waives the automated check, not the human one.
The failed ValidationResult rows are left untouched rather than rewritten,
so the QC queue keeps listing them as warnings and whoever QCs the shot
sees what was waved through. The reason is required and recorded with the
checks it covered on the export's timeline, and the validation panel
keeps an "overridden" badge long after the event scrolls away.
Restricted to admin/producer/supervisor: waving a failed check through is
a supervisor's call.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A 25-shot episode took about an hour to package when the linking itself
is instant. All of it was verification: every delivered file was read
back and hashed, and at ~100 MB/s over SMB one 2.8 GB shot costs 30 s.
Re-hashing a hard link cannot disagree with validation. The entry is a
second name for the same data, so it re-reads the exact bytes that were
already hashed — half an hour per episode to prove something that cannot
be false. Verification is now proportional to what was actually done:
copies are read back in full, links carry over the validation-time digest.
Linking has one gap hashing covered incidentally: a source changed after
it passed QC, which a link would carry faithfully. Source timestamps are
now checked against the export's validation time, which costs a stat per
file rather than a full read.
"verifyMode": "always-hash" restores unconditional read-back for a
production that requires it.
Worker 1.3.0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
UNG_110_030_010 v002 failed with a 461614-frame offset on a render whose
timecode was correct. Two causes, both in how the plate reference was
recorded.
The comp held three same-length elements and the timing plate was picked
by duration alone, so a PRERENDER won on layer order. Using a pre-render
as the timecode reference is circular anyway: it was rendered from this
comp, so it can only ever agree with it — including when both are wrong.
Only footage under the Footage Root is now eligible; among those, longest
still wins.
The offset itself came from the plate's frameStart. AE writes renders as
"..._PRERENDER_01001.exr" while delivered plates use "....01001.exr", and
the parser only handled the dot form, so the number fell back to the
comp's frame range — 462615 against a render numbered from 1001. Both
separators now parse, and an unparseable name omits the field rather than
filling it with a number from the wrong coordinate system.
Verified against the real manifest: the BG01 plate is now chosen over
both pre-renders, and all three filenames parse to 1001.
Panel 2.2.1, worker 1.2.1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The preview manifest now carries the show name, taken from the project's
name rather than its code — a slate is read by a human, so "Ungovernable"
beats "UNG_S1".
Both the worker's headless build and the panel's own preview set it, so a
farm-generated preview and an artist-built one still match. The panel
writes it whenever it touches an overlay, not only when building a
preview, so existing comps pick it up on the next Add Overlay.
The slate's other properties are title case ("Date", "Scene") while the
overlay's are upper ("DATE", "SHOT NAME"), so both spellings of SHOW are
attempted; a template exposing neither reports its actual property names
as a build warning rather than failing silently, matching how the VFX
Scope and Submission Note fields already behave.
Panel 2.2.0, worker 1.2.0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A render carries frame numbers in two unrelated systems. The comp is
anchored to its plate's timecode, so its timeline runs 1247792-1247896;
the output module names files from 1001. Validation compared the file
numbers against the comp range and failed every correctly-rendered shot
by ~1.2 million frames — first_frame, last_frame and missing_frames all
reporting a defect that was not there.
The manifest now carries outputFrameStart (the output module's starting
number, house convention 1001) alongside the comp range. Count and
contiguity are measured on the output numbering; first/last compare
against outputFrameStart and are SKIPPED on older manifests, where the
count and gap checks still hold on their own.
The same conflation broke the API timecode fallback: it anchored on the
comp's frameStart while the render's first frame is 1001, producing a
+1246791 frame "offset". Shot.timecodeStart is the edit in-point and the
plate carries handleFrames of head handle, so the reference is the API
value less handles, anchored on the render's own first frame.
Verified on the shot that reported it: UNG_107_051_010 v004 now passes
count, contiguity and a plate-anchored timecode match. Regression tests
cover both coordinate systems, including that a genuine +8 handle offset
still fails.
Panel 2.1.2.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
twotalesanimation