Lessons 58.1 and 58.2 taught you to enhance honestly. This lesson makes honesty repeatable. The moment a second person, a second campaign, or a second platform enters the picture, memory fails. Someone exports final-final-2.png, nobody knows which product color was approved, and a stale draft ships to the store. Versioning is how a good edit survives the week.

The five stages: never mix them up

Every image lives in exactly one stage at a time:

SOURCE → WORKING → VARIANT → MASTER → EXPORT

"Final-final-2" is not a system because it records nothing: not which source it came from, not what changed, not who approved it, not where it may be used. The five stages answer all four questions by construction. If you cannot trace an export back to its master, source, and approval in under a minute, the workflow is broken.

Naming and metadata that survive collaborators

Keep names machine-sortable and human-readable. One workable pattern:

YYYY-MM-DD_campaign_platform_ratio_status-vN.ext
2026-09-20_mug-launch_shop_1x1_approved-v3.jpg
2026-09-20_mug-launch_instagram_4x5_export-v3.jpg
2026-09-20_mug-launch_hero_16x9_draft-v1.jpg

Pair every master and variant with a short Markdown sidecar or index entry: date, campaign, platform, aspect ratio, status (source, draft, in-review, approved, export, archived), owner, source/reference file, and approval note with reviewer name and date. Status is a field, not a filename adjective — that is what kills the final-final habit.

Connect this to Part II, the course home for files as memory. A MEDIA-DECISIONS.md file keeps visual direction and approvals durable across tools and collaborators. Record the brief reference, the visual system, which variant won and why, the edit-plan constraints from Lesson 58.1, and the approval signature. Part II stores context so future sessions — human or AI — do not re-litigate settled choices. Part XIII, this part, is the canonical home for directing the media work itself; Part XIV owns channel strategy and Part XV owns monetization, so keep those arguments out of the asset folder.

A minimal sidecar looks like this:

## 2026-09-20_mug-launch_shop_1x1_approved-v3.jpg
- Status: approved master crop
- Source: `source/2026-09-14_mug-raw_01.cr2`
- Working file: `working/mug-launch_master-v3.psd`
- Edit plan: `IMAGE-EDIT-PLAN.md` (glaze color verified vs. physical mug)
- Reviewer: [name], 2026-09-20 — color, label, chip disclosure confirmed
- Allowed exports: shop 1:1, Instagram 4:5, hero 16:9

The review loop

Run every publishable image through the same loop, and record each pass:

creator makes draft
  → reviewer checks facts / brand / fit
  → changes recorded in MEDIA-DECISIONS.md
  → approved export stored as master
  → channel crops made from master

The review checks three things, in order. Facts: product accuracy, label text, colors, claims, evidence integrity — the Lesson 58.1 boundary. Brand: visual system, palette, typography handling, background rules, do-not-use list from Class 57. Fit: crop safety for the placement, legibility at real size, copy space, accessibility (alt text drafted), and audience appropriateness. A reviewer who only checks beauty will approve a convincing lie. Check facts first.

Alt text belongs in the loop, not after it: one concise sentence describing the information the image contributes, written while the facts are in front of you. So do channel crops: every export is cut from the approved master, never re-edited independently, so a fix to the master propagates cleanly to a new export version.

Practical exercise: the five-asset folder

Organize five assets for a small campaign — the local-business product launch from this part's running examples works well — in one folder with an index file and clear statuses. Suggested structure:

mug-launch/
  source/            # originals, read-only
  working/           # layered files + IMAGE-EDIT-PLAN.md
  variants/          # drafts and in-review candidates
  masters/           # approved full-resolution versions
  exports/           # platform-sized publishing copies
  INDEX.md           # metadata table + approval log
  MEDIA-DECISIONS.md # Part II link: direction, choices, approvals

INDEX.md carries one table: filename, stage, ratio, status, owner, source file, approval note. Include at least one asset in each of three statuses (approved, in-review, export) so the table demonstrates the system, not just the happy path. Add alt text and the intended placement for each export.

Finish line: an asset folder another collaborator can open and publish safely — no questions needed about which file is current, what was approved, or where an export came from.

Common failures: editing an export directly for "one quick fix" (recovery: apply the fix to the master, mint a new export version, archive the old one); storing approvals in chat messages (recovery: copy the decision, reviewer, and date into MEDIA-DECISIONS.md the same day); keeping two competing masters (recovery: demote one to archived, note why in the index, protect the winner).

Check your understanding

• What distinguishes a variant from a master from an export, and why must exports derive only from masters?
• Which metadata fields must every master carry, and why is status a field rather than a filename adjective?
• What three passes does the review loop check, in what order, and why do facts come first?
• What does MEDIA-DECISIONS.md preserve, and which part of the course is its canonical home?

You now have the full Class 58 pipeline: plan the edit honestly, enhance for the placement, and version so the truth survives. In Class 59 you will put still images in motion — planning short AI-assisted video as a sequence of shots rather than a single lucky generation.