Lossy Data Conversion¶
A completed change of data representation drops a distinction present in the source, so the converted result alone cannot recover it.
Core Idea¶
A lossy data conversion changes source data into a target type or format and produces a result that cannot, by itself, recover at least one distinction the source contained. The loss can be rounded numeric precision or a dropped document feature. The key is the actual conversion result: a target type's limited capacity is only a warning until a particular input loses something. A retained original or sidecar may preserve information elsewhere, but it does not put that information back into the converted result.[ref-1b76b6e3ec12][ref-2dca320af0b3]
Scope of Application¶
The pattern applies when a particular conversion produces target data that omits a present source distinction.
- Numeric values: Java's worked
int-to-floatconversion loses precision, as the-46return difference shows.[^ref-1b76b6e3ec12] - Feature-bearing documents: Word's Plain Text export drops formatting, macro code, or graphics present in the source.[^ref-2dca320af0b3]
Clarity¶
Ask what distinction the input really contained, what the converter produced, and whether that output alone determines the original distinction. A return conversion can reveal a mismatch, but byte differences alone need not mean data-model loss. The W3C's XML/binary-XML round-trip methodology distinguishes byte-exact copies from lossless equivalents within that specific format setting.[ref-1b76b6e3ec12][ref-e33d82b52939]
The Word example is conditional on a feature-bearing source, not a report of one observed named document.[^ref-2dca320af0b3]
Digital-preservation format migration can lose different amounts of content, formatting, and functionality depending on the formats and source; migration by itself does not guarantee a loss.[^ref-e1968fb2e0bd]
Manages Complexity¶
The fidelity audit becomes four checks: source features → target capacity → actual conversion → target-only recovery. If a necessary feature cannot fit the target, choose a richer target or retain the source outside the converted result. If the target could carry it but a particular converter discards it, the converter or settings may need change. These are diagnostic options, not claims that the cited Java or Word examples were repaired.[ref-1b76b6e3ec12][ref-2dca320af0b3]
Abstract Reasoning¶
Before calling a conversion lossy, name the preservation criterion and inspect the particular input. Then check the output against that criterion. Java's worked integer example establishes an actual loss; a Word-to-text example does so only when the source contained a feature the text output drops. A conversion request that fails without target data is not a completed lossy conversion for that input.[ref-1b76b6e3ec12][ref-2dca320af0b3]
Knowledge Transfer¶
Numeric casts and document exports use different formats but the same test: source distinction, target result, and failed recovery from that result alone. The broader Transformation pattern overlaps where a conversion preserves some source property while altering form, but a target result can drop all substantive source features. That is why the two patterns should not be treated as interchangeable.
Neighborhood in Abstraction Space¶
Lossy Data Conversion sits in a moderately populated region (55th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Program Execution & Runtime Concepts (34 abstractions)
Nearest neighbors
- Data Migration — 0.87
- Incremental Backup — 0.86
- One-Sided Communication — 0.85
- Data Backup Method — 0.85
- In-Place Algorithm — 0.85
Computed from structural-signature embeddings · 2026-10-17
References¶
[^ref-1b76b6e3ec12]: Oracle. Java Language Specification Java SE 26 Chapter 5 Conversions and Contexts. Especially the opening examples, §5.1.2 and Example 5.1.2-1, and §5.1.3. Official full HTML inspected; the worked int-to-float round trip prints -46.
[^ref-2dca320af0b3]: Microsoft. How to troubleshoot damaged documents in Word. Method 6, Step 4 and following Note. Official full HTML inspected; Plain Text save drops formatting, macro code, and graphics. Here this supports a conditional feature-bearing document case, not a named observed conversion.
[^ref-e33d82b52939]: W3C XML Binary Characterization Working Group. “XML Binary Characterization Measurement Methodologies,” §6.14, Round Trip Support. https://www.w3.org/TR/xbc-measurement/. W3C Working Group Note inspected in full; its exact-byte versus lossless-equivalence distinction is used only for XML/binary-XML round-trip methodology.
[^ref-e1968fb2e0bd]: PREMIS Editorial Committee. PREMIS Data Dictionary for Preservation Metadata version 3.0. Glossary, “Migration,” printed p. 270. Official full PDF inspected; migration creates a new-format version and loss varies by format and content type.