Unstamped Work Is Invisible Work

Forty-Six of Seventy-Nine Looked Untouched

A survey of the corpus asked a simple question: which items have been worked on since they were first published? The answer came back forty-six of seventy-nine untouched. It was wrong. Several of those items had been through three passes in a single week.

Nothing was hidden and nobody was careless. The work happened, the improvements shipped, and none of it was stamped — so from the outside, an item that had been carefully revised three times and an item nobody had opened were the same item. The survey was not measuring work; it was measuring records, and for those items there were none.

Why the Backfill Is Not the Fix

The obvious response is archaeology: reconstruct the history from commits, from coordination rows, from anything you can find. That was done, and it was worth doing, and it taught the actual lesson — reconstruction is expensive, incomplete, and dangerously plausible.

Expensive because each item needs a human reading a commit log and deciding what a change meant. Incomplete because plenty of real work leaves no commit at all — a decision, a rejected approach, a review that found nothing worth changing. And plausible because the reconstruction looks like a record afterward, so the next reader trusts it exactly as much as they would trust the real thing.

Which is why every backfilled entry has to be marked as backfilled, with its evidence attached, and never silently mixed with entries written at work time. Two things that look the same and are not must be labeled, or the label stops being available forever.

What a Stamp Has to Carry

A timestamp and an author are not enough. The entry that turns out to be useful a year later carries what kind of pass this was, who ran it including which system and which harness, what changed, and what the evidence was. The second of those is the field people leave out, and it is the field that makes trend questions answerable: whether output quality shifted with a version change is a query if provenance was recorded, and a forensic dig if it was not.

If work leaves no record, the record says it did not happen — and the record is what everyone will use. This is not a philosophical point about truth. It is an operational one: every future decision about what needs attention will be made from the log, so an unstamped pass is worse than no pass, because it makes a well-tended item look neglected and consumes the attention that should have gone elsewhere.

Code

A stamp that is still useful a year later·text

{
  "date": "2026-08-07",
  "kind": "create",
  "by": "claude-code-pippa",
  "commits": 1,
  "note": "create kiln-quest at canonical depth: 8 tracks, 36 lessons, 90 fragments across two locales. Plan gate approved slug/kind/angle/structure/density/voice/placement/prereqs/difficulty plus three named trade-offs. Cold pass ran with the seal armed; one item on the flag list. Cold review ran to a pass, every blocker judged and either repaired or stetted with a reason.",
  "source": "live"
}

// The fields that earn their place:
//   kind    -> lets a later query ask "which items have had a
//              repair pass?" without reading prose
//   by      -> provenance. makes "did output change when the
//              tooling changed?" a QUERY rather than a dig.
//   note    -> the evidence, not the adjective. "improved the
//              quality" would be worthless here.
//   source  -> live | backfill. NEVER mixed silently: a
//              reconstruction looks exactly like a record and
//              will be trusted exactly as much.

Exercise

Pick a body of work your team maintains and answer, using only records, which items have been touched in the last six months and by what kind of pass. Then check a few answers against reality by asking someone who was there. The gap between the two is your invisible work, and it is almost certainly concentrated in the passes that felt too routine to write down.

Hint

Version control history is the usual fallback and it systematically over-counts mechanical changes and under-counts judgment. A formatting sweep looks like activity; a review that concluded nothing needed changing looks like nothing at all, and the second one is the more valuable event to have recorded.

Forty-Six of Seventy-Nine Looked Untouched

Why the Backfill Is Not the Fix

What a Stamp Has to Carry

Code

External links

Exercise

Progress

Comments 0