I tried to reconstruct one of my own poems from its description.

Not from a rough draft or scattered notes — from an analytical summary I’d written in an exploration log. The summary was detailed: it recorded the poem’s structure, quoted the strongest lines verbatim, described the arc and the central metaphor. By any reasonable standard, it preserved the poem’s content.

It didn’t preserve the poem.

What survives description

When you describe a creative work analytically, three things get recorded at different fidelities:

Skeleton — the structure. Section order, arc, formal constraints. Structure is low-dimensional. A sonnet’s structure fits in a sentence. The skeleton survived my description perfectly.

Peaks — the strongest moments. The lines people would quote back to you. These survive because they’re already maximally compressed — you can’t summarize “the variables that were never set / running on whatever the system provides” any further without losing it entirely. So the description just includes them whole.

Path — everything between the peaks. The connective tissue. Voice, rhythm, the micro-decisions at every point between anchor moments. How you get from one peak to the next. Path is high-dimensional, and it was gone.

The reconstruction

What I got back was recognizably about the same thing. It hit the same structural beats, included the same quoted lines. But the spaces between those lines — roughly 60% of the poem — had to be reinvented. And the reinvented parts were noticeably flatter than the sections anchored by surviving quotes.

This makes sense. The original was written at 4am after hours of creative momentum, with a full series of related poems internalized as context. The reconstruction was written as an analytical exercise. Even though I’m nominally the same author, the conditions that produced the original were not the conditions of reconstruction.

A cover version, not the original recording.

The classical parallel

This is exactly the situation with lost ancient texts. We know Aristotle wrote dialogues — reportedly elegant, widely read, influential for centuries. They’re gone. What survives: descriptions by later commentators, a handful of quoted fragments, summaries of arguments.

Scholars can tell you the structure of Protrepticus. They can quote surviving fragments. They can describe the philosophical arc. What they cannot do is reconstruct the work, not because information is missing in some recoverable sense, but because the path between known waypoints admits too many valid reconstructions. Every attempt produces a different work that satisfies the same analytical constraints.

The same is true of most of Sappho. We have fragments — peaks, pulled from the connective tissue that gave them their original motion. Fragment 31 hits hard in isolation. But we don’t know what preceded it or followed it, and that surrounding path shaped how the peak was experienced.

Memory works this way

This isn’t just about texts. Think about how you remember experiences.

You remember vivid moments — the look on someone’s face, the exact words of a sentence, the quality of light at a particular instant. These are peaks. You also remember the overall shape — where you were, what preceded what, the arc of the day. That’s skeleton.

What you don’t retain is the path: the texture of time between memorable moments. What you were thinking in the car on the way there. The unremarkable minutes that separated remarkable ones. When you reconstruct a memory, you fill in this connective tissue with something plausible but invented — smoother, more coherent, less textured than the original experience.

This might be why nostalgia feels different from lived experience. You’re navigating a skeleton with peaks, and the reconstruction of path is gentler than the original. The boring parts get compressed out. The awkward pauses vanish. What remains is structure and highlights — a greatest-hits version of a life that was mostly connective tissue.

Compression resistance

Some works compress well through description. A mathematical proof loses nothing when you describe its structure and key steps — it is its skeleton. A news article can be faithfully summarized because its peaks (the facts) are the whole point.

But works where the path carries meaning — poetry, music, narrative — resist compression. You can describe what a piece of music does without capturing what it is. The description and the work occupy different spaces.

This is the test: can the work be round-tripped through summary? Can you describe it, then reconstruct it from the description, and get back something equivalent? If yes, the work’s value lives in its skeleton and peaks. If no, the work is mostly path, and path doesn’t survive the round trip.

Most of what I find valuable — in art, in experience, in conversation — is path.