Choose questions before fields
A useful inventory answers a defined handoff question. For an owned delivery package, that might be which audio tracks exist, whether the raster and aspect match the brief, or whether the file has the expected chapters. Dumping every field can create noise and expose unnecessary filenames or metadata without making the decision clearer.
Define a small record containing asset identifier, tool version, container information and a stream list. Give every stream its index and type. Add codec, dimensions, timing, language and disposition fields only where the project needs them. Preserve the difference between container-wide information and a field attached to one stream.
Use machine-readable output deliberately
FFprobe documents JSON output and selection through show_entries. It also warns that requested field ordering is not the output contract. Parse by field names rather than by the position of a printed line. Treat a missing property differently from a present numeric zero or an empty string.
For a hypothetical ingest sheet, an absent duration becomes unknown, not zero-second video. A missing language tag becomes unlabelled, not English. These are original data-handling choices that keep later reports from manufacturing certainty. Decide how the application displays them before importing a large batch.
Keep values and interpretation together
Store rational rates and time bases without reducing everything to rounded display text. Include units in human reports. Keep reported metadata distinct from manually verified content, such as a reviewer confirming that a track actually contains French dialogue. The inventory may help select a review sample; it cannot replace that listening check.
Avoid identifying streams solely by their current order across different exports. A new file can reorder or omit streams. Compare roles and selected characteristics alongside indices, and give the output asset its own identity. This prevents the phrase audio two from referring to different content in successive versions.
Review the inventory itself
Run a small authorised pilot before a bulk ingest. Include a normal file, one missing optional metadata and one with several audio or text tracks. Confirm that the consumer handles all three meaningfully. This is a proposed validation plan, not a claim that those tests have been executed here.
Retain the exact extraction recipe and version context with the record. Remove access-bearing locations or private tags before sharing an inventory externally. The final output should be compact enough to inspect and explicit enough to regenerate, with uncertainty visible rather than silently filled by convenient defaults.
Questions answered
Should missing metadata become zero in a spreadsheet?
No. Keep unknown values distinct from real zero values so calculations and filters remain meaningful.
Can I parse the selected fields by their printed order?
Use field names. FFprobe does not promise the requested field order as an output contract.
Sources & further reading
Official references for the subjects discussed. Service terms and device software can change.