Explore

@rdicosmo.bsky.social Sep 29, 2026, 3:22 PM

A distinction I took from Robin Milner in 2007: engineering makes things that work, more or less; science tries to do things right. Thirty years of "more or less" brought us back to the same problem, now at AI-crawler scale. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough

Two definitions in large type on a pale ground. Engineering: making things that work, more or less. Science: doing things right. The line beneath gives them as the author's later paraphrase of a distinction he remembers Robin Milner drawing at Inria's fortieth anniversary in Lille in December 2007.

@rdicosmo.bsky.social Sep 23, 2026, 3:01 PM

Measure what survives without an archive: after five years, 8.5% of one software distribution's sources are gone, and another 1% still download perfectly — but they are no longer the same bytes. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #DigitalPreservation #SupplyChain

96.4% of one distribution's source code was still retrievable from its original addresses after one year, and 91.5% after five years.

@rdicosmo.bsky.social Sep 18, 2026, 5:32 PM

At Software Heritage we put up a bot wall like everyone else, and it blocked the people trying to give us code. The commit that fixed it states the thesis of this whole series, in a parenthesis, with no irony. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #SysAdmin #OpenSource

A commit message reading "This is a trivially circumventable 'are you a browser?' check, but one can't answer an Anubis challenge when doing cross-site requests to the API anyway".

@rdicosmo.bsky.social Sep 15, 2026, 8:37 PM

The bot walls are catching the wrong traffic. The abusive crawler rotates through ten thousand home addresses; the honest one says who it is, and gets stopped. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #WebScraping #OpenSource #CopyrightReform

A wall stopping a single regular shape while a cloud of thousands of scattered dots passes straight through it, captioned "the wall stops the one that identifies itself".

@rdicosmo.bsky.social Sep 11, 2026, 1:57 PM

"Collect once" sounds like a slogan until one looks at the machinery: ask a forge what changed, skip everything that did not, transfer only the difference. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #DigitalPreservation #Metadata #OpenSource

One line of SQL from the Software Heritage scheduler: origins are candidates only when the forge's recorded last update is later than our last successful visit.

@rdicosmo.bsky.social Sep 8, 2026, 3:03 PM

Identifiers anyone can recompute from the bytes themselves, on their own machine, without asking anyone's permission. We built Software Heritage that way on day one, because that property cannot be retrofitted. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #OpenSource

Two rows of boxes. The upper four hold different partial contents and are labelled federated fragments, each partial and each alone. The lower four hold identical complete contents and are joined by a single amber line reading "one identifier, recomputable by anyone".

@rdicosmo.bsky.social Sep 3, 2026, 1:10 PM

Before AI crawlers, we were already wasting the web's effort ourselves: my own research community built thousands of separate repositories that every aggregator must crawl one by one — and we called it autonomy. 👉 www.dicosmo.org/good-enough/ #GoodEnoughIsNotGoodEnough #OpenAccess