Dailies workflow for small crews: from card to editor-ready in one evening

What “dailies” means when there is no dailies department

Strip away the department and the job description survives intact. Dailies means that by the next morning, the day’s material is safe (verified copies on more than one device), legible (organised so anyone can tell A‑cam from drone from sound roll), married (external audio synced to picture while the shoot is fresh), playable (proxies for anything the edit machine chokes on) and searchable (transcripts for anything spoken). The output of all five is the editor package — even when the editor is you in three weeks, with no memory of which take was the good one.

The order is not negotiable. Safety comes before everything, because every other step can be redone tomorrow — a failed offload cannot.

Stage 1: Offload — a copy is not an offload

Dragging a card’s contents onto an SSD in Finder produces a copy. It does not produce an offload, because nothing has confirmed that the bytes that arrived are the bytes that left. Finder will happily report success on a copy that a flaky reader, a failing cable or a dying destination has silently corrupted — and the corruption surfaces weeks later, in the edit, when the card is long since formatted.

An offload is a copy plus a checksum verification: every source file is hashed while it is read, the destination file is read back and hashed again, and the two values must match. The standard hashes are xxHash‑64 (computed faster than any disk can deliver data, so it costs nothing) and MD5 (slower, single-threaded, but the value some post houses still ask for in the report). Either way the point is the same: a checksummed offload turns “I think it copied” into a per-file, documented yes.

Copy count follows the 3‑2‑1 rule, scaled to small-crew reality: three copies of the material, on at least two different devices, one of them stored somewhere else. In practice: the working SSD that goes to the edit, a second drive that stays on a shelf, and the card itself as the third copy until both drives are verified. Which leads to the two card rules that outrank everything:

• Never format a card until two verified copies exist. Not one copy, not two unverified copies — two copies whose checksums have matched.
• Never work on the card. No renaming, no deleting false starts, no “quick look” in an NLE that writes sidecar files. Cards are read-only from the moment recording stops.

Stage 2: Organise — boring and identical every time

Organisation is the cheapest stage and the one most often skipped at 11 p.m. The structure itself matters less than its constancy: date-stamped shoot folder, one folder per camera, sound rolls as recorded, proxies mirroring originals one-to-one, projects and reports alongside — the exact layout, folder by folder, is specced in the editor package guide.

Two habits carry the stage. First, preserve card structure where the format demands it: BRAW, R3D, XAVC and AVCHD all have relatives that expect their directory scaffolding intact, and a “cleaned up” card can be an unreadable card. Copy the whole card into a folder named for the roll (A001, A002…) and do the tidying one level above. Second, sort into a junk folder, never into the trash — false starts, lens caps and corrupt files are moved aside where they cannot clutter a bin, but every sorting decision stays reversible.

Stage 3: Sync the same night

Dual-system sound should be married to picture during dailies, not deferred to the edit, for one practical reason: tonight you still know what happened. The recorder that lost power after lunch, the take where the boom op started late, the camera that was never jammed — tonight these are memories; in three weeks they are mysteries.

The mechanics are the same two paths as always: timecode when camera and recorder were jammed from a common source (fast, whole day in one pass, trustworthy exactly as long as the jam is fresh — the arithmetic of how clocks wander is in the timecode drift guide), and waveform correlation against the camera’s scratch track when timecode is absent or lying. The full decision tree, including what to do when both paths fail, is in the no-clapper sync guide — dailies is simply the right night to run it.

Whatever performs the sync, the result should land in the package as interchange files — FCPXML, Premiere XML or ALE with synced relationships the NLE imports directly — plus a sync report that states per clip what was matched, by which method, and what was left unmatched. Unmatched is information: a clip honestly listed as “no usable audio” saves the editor an hour of looking for audio that does not exist.

Stage 4: Proxies — only where needed

Proxies are a triage decision, not a default. Long-GOP H.264 from a mirrorless camera edits acceptably on any recent Mac; 8K RAW does not edit acceptably on anything without help. The dailies question is per format: will the edit machine scrub this natively? If yes, skip the render and save the hours. If no — BRAW, R3D, CinemaDNG and their relatives — render ProRes Proxy or ProRes LT mirrors with the same filename, start timecode and reel as the original, because that identity triple is what lets the NLE swap proxies for originals at conform time. The per-format recipes and the GB-per-hour arithmetic are in the RAW proxy workflow guide.

On a small crew the practical constraint is render time on the same machine that just spent the evening offloading. The honest answer is to queue proxies last and let them run overnight — which only works if the queue is set up before you sleep, not remembered at breakfast.

Stage 5: Transcripts while the machine is busy anyway

If the day contained interviews or usable dialogue, dailies night is when transcription belongs: it is unattended machine time, and it turns tomorrow’s “where did she say the thing about the merger?” into a text search instead of a scrub. On-device models transcribe roughly in real time or faster on Apple silicon; the mapping from transcript timestamps to actual timecode — the detail most setups get wrong — is covered in the searchable footage guide.

The evening, timed

Concrete numbers for a plausible documentary day — 1 TB across two cameras and a recorder, offloading through a USB‑C NVMe enclosure that sustains 700–800 MB/s:

• Copy pass, first destination: 1 TB at ~750 MB/s ≈ 22–25 minutes.
• Verify pass (destination read back and hashed): the same again — call it 25 minutes.
• Second destination, copy + verify: another ~50 minutes if sequential; overlapping the two destinations compresses it.
• Organise: 10–15 minutes when the structure is a habit; an hour when it is improvised.
• Sync: minutes by timecode; longer by waveform, but unattended.
• Proxies + transcripts: queued, overnight, zero attended minutes.

Attended total: roughly two hours, most of it waiting on verification passes — which is exactly the part that cannot be skipped, because the verify pass is the offload. The realistic optimisation is not to cut steps; it is to overlap them, so organising, syncing and queueing happen while the second copy verifies.

How dailies nights die

• The card was formatted after one copy. The single most expensive habit in small-crew production. Two verified copies, then format — no exceptions for “it’s late”.
• The copy was never verified. A Finder copy that “worked” hides corruption until the edit. If no checksum ran, the offload has not happened yet.
• Sorting happened on the card. Deleting false starts in-camera or renaming on the card breaks card structure and, for some formats, playability. Cards are read-only.
• Originals were renamed. Filenames are one third of the identity triple that relinking and conform depend on. Rename in bins and in proxies’ display names, never on source media.
• Sync was deferred. Three weeks later, nobody remembers which recorder file covers which take, and the scratch-track mystery costs half a day. Sync while you remember.
• The second copy lives in the same bag as the first. One stolen backpack, zero copies. The 3‑2‑1 rule’s “one elsewhere” is about geography, not devices.

Where Launchr Post fits

Every stage above is a solved problem individually; the small-crew pain is running five of them, in order, at the end of a twelve-hour day. Launchr Post runs the chain as one pass on your Mac: drop the cards and it offloads with checksum verification, sorts cameras, sound and junk into the structure editors expect, syncs by timecode and sample-accurate waveform, drives the proxy renders, transcribes on-device and exports the editor package — interchange files, sync report and transcripts included. The two hours of attended dailies become the ten minutes of inserting cards.