A Pitch Reel Lost Its Media. Relink the Sources—or Rebuild the Edit?

The media comes back online and everyone relaxes. The offline badges clear, the sequence plays, the timeline looks like a timeline again. That moment of relief is exactly where pitch reels get quietly ruined.

Relink or rebuild is the wrong first question. The first question is whether the file you are about to attach is the file the edit was cut against. A relink is a successful recovery only when the source identity is established on more than a plausible name, and when the interval the sequence actually uses is present in the file you attached, behaving the way it behaved before.

Where identity can be established, relink — and verify the result in the timeline, not in the project panel. Where it cannot, keep whatever is genuinely recovered and declare the rest as reconstruction. Making an offline warning disappear is not restoring an edit.

Preserve the interrupted state before you fix anything

The project as it stands is evidence. Copy the project file and work in the copy; leave the original where it is, opening offline, exactly as it arrived. That original is your only record of what the edit referred to before anybody started making well-intentioned repairs.

The same applies to the media. During diagnosis, do not reorganize the folders, do not rename candidate files, and do not delete the ones that look like duplicates. Two files with the same basename in two folders are not a mess to be cleaned up; they are the two candidates you have to tell apart. Every tidy-up destroys the property you were about to compare.

Keep any reference export. If a review file or a delivered cut exists, it is the most valuable object in the whole job, and I will come back to why — and to what it cannot do.

Then inventory. For each offline item: the last known path the project recorded, the basename, the duration and frame rate the project recorded, and every place the item is used — which sequence, which timeline position, which source in and out. That last column is the one most people skip and the one that decides the outcome. A clip used in three sequences has three chances to be wrong.

Relinking is a claim about identity, not about availability

Adobe's documentation for the Link Media workflow, checked in September 2026, describes the dialog as exposing the original path and names and letting you match on a combination of properties — and states that the combination you choose should uniquely identify each file you are relinking. That is the whole discipline in one sentence: the property set is supposed to be unique, and metadata agreement is a candidate match, not proof of footage identity.

Three questions hide inside that:

Which take? Same camera, same setup, second pass. The filename cannot tell you, and neither can file size.

Which flavor? The camera original or a transcode of it. Same frames, potentially different timecode base, different extent, sometimes different audio structure.

Which extent? The full card clip or a partial copy that stops early. Extent is where a relink can be correct about identity and still useless.

A stale path, a name, and a duration is a decent start. It is not a verification.

A constructed fixture: two takes both called MVI_0142.MP4

What follows is a constructed example, not a recovery anyone ran. The numbers are chosen so the checks are easy to follow, and nothing here was tested in an application.

A fictional 90-second pitch reel, Harborline_Pitch_v4, lives in three sequences: Reel_90s, Cutdown_15s, and CaseStudy_45s. Nine media files went offline after an assistant consolidated footage across two drives and the drive letters changed.

One of them is MVI_0142.MP4. It was last at Card_A/PRIVATE/M4ROOT/CLIP/MVI_0142.MP4, and the project recorded its duration as 00:00:12:04 at 24 fps — 292 frames.

Three sequences use it:

There is one reference export: Harborline_Pitch_v4_Review.mp4, a 90-second H.264 of Reel_90s made two days before the drive move. It covers one sequence out of three. It is graded, compressed, and flattened. It is a reference, not a source, and it can never be relinked to anything.

A search for the basename turns up three candidates:

The relink that works and is wrong

Accept (b). It has the right name, it is the same storefront at the same hour, and both takes fit the used source ranges. The duration disagrees — (b) reports 297 frames against the 292 the project recorded — and five frames is exactly the size of discrepancy that gets explained away rather than investigated: a rewrap, a camera that rounded its last frame, a number nobody read twice. The offline badge clears in all three sequences at once, because all three reference the same project item.

Now look at what the reel actually does.

In Reel_90s, the export shows a cyclist crossing frame and exiting right. On take B, that interval is a static storefront — the outgoing movement that the next cut lands on is gone, and a cut that was built on motion now reads as a cut away from nothing. This is the three-second interval where the two takes differ most, so this is the failure you are most likely to catch.

In Cutdown_15s and CaseStudy_45s, the used intervals sit before frame 200, where the two takes differ only by the shadow line and a small exposure shift. Those two uses will pass a glance. They will pass several glances. They are still wrong, and they will be wrong in a way that is much harder to argue about once the deck has been sent.

That asymmetry is the argument for a method rather than a spot check. You cannot verify the uses you happen to look at first; you have to verify every use, including the ones on sequences nobody has opened in a week.

The relink that holds up

On the copy, build the candidate table first — path, basename, recorded duration, extent, and a content anchor inside or near the used interval. Then:

Candidate (a) matches the recorded duration exactly, 292 frames, and lives in the consolidated folder where the card was copied. Candidate (b) does not match duration — the same five frames that were talked away in the first attempt — which is not proof on its own but is a legitimate discriminator against this pool. Candidate (c) matches duration but is a derivative of the same take with the audio folded into a single mix — the frames would come back, the channel structure the mix was built on would not.

Then compare frames. Pull Reel_90s to the offending interval and step through it against the export. At 192–264 the export has the cyclist. Candidate (a) has the cyclist. Candidate (b) does not. That comparison — export against candidate, at the interval the edit uses — is what turns a plausible match into a verified one.

Relink (a) into the copy. Check the clip's used interval, its first and last used frames, its sync, its audio channels. Then check the other two sequences, and be honest in your notes about what you could and could not confirm there: the deck stills pulled from the reel can tell you it is the same location, same framing, same time of day. They cannot tell you the awning shadow line is the one the cut was made with. Two sequences out of three have no reference that shows their intervals. Say so.

One wrong relink becomes a method

The real danger is not the single bad link. It is the precedent.

Nine files went offline in the same reorganization, which means nine files probably share the same stale path. Once (b) "works," you will run the same search on the other eight, because the approach just produced a clean timeline. An unverified match that landed is indistinguishable from a recovery until somebody checks, and it may never be checked.

So when the evidence disagrees — a duration that will not line up, a frame that is not where it should be — stop that branch. Do not let a convenient match license the same shortcut across the folder.

Two more failures are worth naming, because both leave the timeline looking healthy.

Extent. A different offline item, storefront_Bcam_pass.mov, recorded at 00:00:19:02 — 458 frames — with Reel_90s using source 00:00:09:00–00:00:13:00, frames 216–312. A partial copy survives on a shuttle drive, truncated at 00:00:06:18, 162 frames. That copy is the right file; if you relink it, the badge clears and the clip comes online. The sequence's interval starts 54 frames past the last frame in the file. There is no arrangement of that relink that puts the cut's frames back. Identity and extent are separate questions, and both have to be answered.

Interpretation and flavor. A correctly identified source can still behave differently from how the project last read it — a different frame rate interpretation changes how much real-world time a sequence interval contains, and a transcode with a folded mix changes the audio a cut was built against. The clip is the right take and the edit is not the edit.

When the honest answer is a declared rebuild

Some material cannot be recovered. In the fixture, the second-unit drive is dead and there is no copy of the rest of that folder. Nothing you do with Link Media will bring it back, and the temptation is to reach for the export — lay the flattened 90 seconds on V1 and call the reel restored.

Do not. That move recovers the picture and destroys the edit: one clip, no per-shot grades, no per-shot versions, no ability to re-time a cut or swap a take later. It also produces the most convincing-looking "successful" relink of all, because everything lines up perfectly against itself.

Rebuild when the candidate pool cannot discriminate, when no surviving copy contains the used interval, or when the only thing left is a flattened render. A rebuild is a legitimate answer. An undeclared rebuild is not.

What a declared reconstruction contains: the intervals you could not restore, the substitute material you used, and the creative differences it introduces. A replacement take that fits is still a different take, and the person approving the cut should be told that in plain language, not discover it in a side-by-side.

The handoff is a list of intervals, not a badge

Separate what you verified from what you did not, interval by interval. Three states are enough:

Confirmed. A reference that shows that specific interval — an export, a delivered file, a review note with a timecode, a still pulled from that moment — agrees with the recovered source.

Consistent. Nothing contradicts the recovery, but no reference shows that interval. The candidate is unique against the pool you have, the interval sits inside a file whose identity is established elsewhere, and you are reasoning rather than comparing. That is worth saying out loud.

Reconstructed. No identified source covers the interval. The material there is a substitution, and it is labeled as one.

In the fixture, that produces something like: MVI_0142.MP4 recovered from candidate (a), confirmed at Reel_90s 00:00:18:00 against the review export, consistent at the other two uses, no reference available. storefront_Bcam_pass.mov extent incomplete — right file, 54 frames short of the used interval; rebuild required for that range. Second-unit material unavailable; two shots reconstructed from a later take, differences noted for approval.

That list is the deliverable. It is also the only thing standing between a plausible filename and a client who finds out in the room.

Frequently asked questions

What should be preserved before trying to repair an offline project?

Copy the project file and work in the copy; leave the original opening offline exactly as it arrived. Do not reorganize folders, rename candidate files, or delete apparent duplicates during diagnosis. Keep any reference export. Then inventory each offline item by last known path, basename, recorded duration and frame rate, and every place it is used—sequence, timeline position, and source in and out.

What does a relink actually claim?

A relink is a claim about identity, not availability. The Link Media dialog may expose original paths and names and let you match on properties, and the chosen combination should uniquely identify each file, but metadata agreement is only a candidate match, not proof of footage identity. Verification belongs in the timeline, not the project panel. Three hidden questions are which take, which flavor, and which extent.

In the constructed example, why is candidate (b) a wrong relink even though the badge clears?

Candidate (b) has the right basename and same storefront, but it is take B and reports 297 frames against the recorded 292. In Reel_90s, the used interval 192–264 should contain a cyclist crossing and exiting right; on take B that interval is a static storefront. The other two sequences use earlier frames where the takes differ less, so they may pass a glance—but they are still wrong. A badge clearing in all three sequences does not restore the edit.

Why does candidate (a) hold up, and what remains unconfirmed?

Candidate (a) matches the recorded 292 frames and lives in the consolidated card folder. Pulling Reel_90s to the offending interval and comparing against the review export shows the cyclist in both. That turns a plausible match into a verified one. But only Reel_90s has a reference showing its used interval; the other two uses are merely consistent—nothing contradicts them, but no reference confirms them.

When is a declared rebuild the honest answer instead of a relink?

Rebuild when the candidate pool cannot discriminate, when no surviving copy contains the used interval, or when the only remaining material is a flattened render. A rebuild should declare the intervals that could not be restored, the substitute material used, and the creative differences it introduces. Laying a flattened export on V1 recovers the picture but destroys the edit: one clip, no per-shot grades, no versions, no ability to re-time or swap takes later. The handoff should separate Confirmed, Consistent, and Reconstructed intervals.

Further reading

• Move a Pitch-Reel Edit to Another Editor Without Losing Its Timing — Read before handing the recovered reel to another editor, so the native project, reference movie and translated edit are verified as three separate deliverables.
• Mixed Frame Rates in a Pitch Reel: Preserve the Event's Duration—or Deliberately Change It? — Read when a recovered clip plays at the wrong speed, to decide whether to change source interpretation or only one timeline instance.
• Read Video Timecode Before Giving Notes on a Commercial — Read before comparing source in/out positions, so source, sequence and player clocks are named and an interval note cannot send people to different frames.