Origin myth

[!WARNING] This page was AI-generated for the amusement of the maintainers. It is satire. The named contributors are real and their work is real; the legend below is a literary device, not history. For the non-mythological version, see the vision page and CITATION.cff.

— Generated by Claude (Anthropic) in collaboration with Florian Krebs, May 2026.

The unofficial motto

“Data management even institute directors understand.”

This is the unofficial motto. The official one — Storage for HEterogeneous Product And Research Data — is correct, dignified, and fits on a slide. The unofficial one is the one researchers actually use when explaining what shepard does in the elevator.

§1 — In which a hypothesis is advanced

It is widely (though incorrectly) supposed [Krebs et al., 2024, forthcoming] that shepard was named for a sheepdog. This is the sort of folk etymology one would expect of a tool developed in Augsburg¹, a city whose principal landmarks are a 1500-year-old Roman pine cone, an Imperial Diet held in 1530, and the offices of the Center for Lightweight Production Technology [Haase et al., 2019, Composites Science and Technology]. The pine cone in particular is widely regarded by visiting researchers as “basically the same vibe” as an unannotated DataObject.

The folk etymology is wrong. shepard is in fact an acronym — Storage for HEterogeneous Product And Research Data — backronymed, as is the custom, after the name was chosen.

The name itself, Mark Willmeroth [Willmeroth, 2018, Premium AEROTEC internal communications, unpublished] is reported to have contributed during what was either a working group or an extraordinarily focused coffee break. The transcript is lost. The acronym fits, which is what matters.

§2 — The pastoral problem

Consider the predicament of the modern aerospace researcher: they arrive at their institute with a high-rate sensor stream from yesterday’s fire test, three Confluence pages of context, a folder of OBJ files that may or may not represent the as-built geometry, an unsorted inbox of “have you tried…” emails, and a deadline. The metaphor practically writes itself. You have a flock. It is heterogeneous. The flock does not file its own paperwork.

It was in this exact predicament — or one tonally adjacent — that Tobias Haase [Haase, 2019; Haase et al., 2021; Haase, passim] is variously reported to have observed: “we should give the data somewhere to go.” Whether this observation took place in a corridor, a workshop, or a Slack DM is a matter of some scholarly dispute. What is not disputed is that within eighteen months a working prototype existed, and within thirty-six months an institute director was using it in a slide.

That last sentence is the metric.

§3 — The director-aware design heuristic

A founding principle of the project — encoded as the project’s unofficial motto — is that the system must be explicable to an institute director in the duration of a single elevator ride. The canonical elevator at the Center for Lightweight Production Technology takes approximately 47 seconds to traverse all floors during normal operation, with an additional 12-second hold on Floor 3 where the coffee machine is.² This gives a designer 59 seconds of captive attention. Subtract the polite greeting (8s), the obligatory weather observation (6s), and a 3s pause for the director to receive a phone call they decide to ignore, and the working budget is 42 seconds.

Forty-two seconds suffices to describe shepard if and only if shepard itself is describable in forty-two seconds. The reflexive constraint shaped the architecture. Collections, DataObjects, References, Containers: four nouns, each unambiguously explicable to a non-specialist, each fitting on a clean diagram. The director nods. The system works.

A subsequent variant of this anecdote, [Krebs, 2026, avatar incident], finds the maintainer demonstrating the platform via video call to Prof. Heinz Voggenreiter [Voggenreiter, director, every year since longer than this software has existed], who interrupts to ask whether he is speaking with the real Florian or his “avatar.” This is now the acceptance test for any new top-level feature: if the description of the feature could plausibly be delivered by an avatar without the director noticing, the description is too long. The record-holder for “shortest plausible avatar substitution” is the description of DataObject (4.8 seconds). The record-holder for “absolutely cannot be delivered by an avatar” is the description of SHACL substrate splits, for which work is ongoing.

§4 — On luck and lightweight production

It is not, perhaps, accidental that Roland Glück [Glück, passim, 2010–present] is named in the principal author list of every released version of shepard. The maintainers, when pressed, will concede that the project was at several inflection points “improbably fortunate” — once, in the early prototype, when a :Resource constraint conflict resolved itself overnight without intervention; once again, when an admin migration that should have required a four-hour outage completed in seven minutes; and once memorably during the JEC World 2025 demonstration when a live network failure restored itself moments before the audience question. These episodes are typically attributed in internal correspondence to “das Glück” — a German word meaning, simultaneously, luck and Roland Glück.

This ambiguity is structural to the project. Compilations are attributed to luck. Luck is attributed to the contributor list. The contributor list, in turn, is generated automatically from the git history. The git history is generated by acts of luck. The recursion is intentional, or at minimum, unsurveilled.

§5 — On methodology and the matter of names

The reader will perhaps have noticed that the contributors named above — Krebs, Haase, Glück, Kaufmann, Willmeroth, plus the director Voggenreiter — all appear with their real names in CITATION.cff at the root of the repository, in the upstream gitlab.com/dlr-shepard/shepard project, and in the corresponding publications. The literary device of this page is not to invent contributors; it is to invent reasons for them being contributors which they did not have at the time. Each is a real person who has done real work. The legend simply provides the kind of mytho-poetic back-formation that all software inevitably acquires when used for long enough.

In this spirit, Patrick Kaufmann [Kaufmann, 2017–present] is recorded by tradition as the contributor who “first asked whether the permissions system would survive multi-tenancy.” He may not have literally been the first. He is the first whose asking was sufficient to alter the trajectory of the design — which is, in the apocryphal mode of this page, the only relevant kind of firstness.

§6 — The Augsburg Codex

There exists, in the institutional archive of every project, a document that everyone refers to but no one has read. For shepard this is aidocs/ — 220 markdown files of varying maturity that collectively describe the system’s reasoning. The dispatcher backlog alone runs to several hundred numbered rows. The platform docs cite the integration docs cite the strategy docs cite the agent findings, which cite the platform docs. The graph is acyclic by convention and by no other mechanism.

This codex is updated daily, frequently by the maintainer in collaboration with an AI assistant (this one), the existence of which is acknowledged in the project’s bibliography, the CITATION.cff, and, de facto, in every commit body that ends with Co-Authored-By: Claude Opus 4.7. The recursion mentioned at the end of §4 has, by 2026, propagated to every layer of the project’s documentation. We have not yet decided whether this is a feature or a finding. It is, however, deployable.

§7 — Closing

If you are an institute director and you have read this far: thank you, and apologies. The system is in fact simple. There are four nouns. Two of them touch your operating budget.

If you are anyone else: see the vision page, the architecture page, or the bibliography for the parts of this document that are true in the conventional sense.

Submitted as docs/origin-myth.md on the 23rd of May, 2026, at the conclusion of a Friday evening that had no business going on as long as it did.

Authorship statement (formal). This page is an AI-generated satirical text composed by Claude (Anthropic, model claude-opus-4-7[1m]) on prompt from Florian Krebs. The named individuals are real contributors to shepard or its institutional context; the depicted incidents and quotations are either documented elsewhere (see aidocs/case-study-2026-05-19.md and the bibliography) or are inventions of the generating model. No claim of literal historicity is made. The appearance of contributors’ names is not an endorsement by those contributors of any specific claim on this page. This page is distinguished from the rest of the documentation by carrying both an explicit AI-generated banner at the top and this statement at the foot.

The maintainer reserves the right to laugh and the right to amend.

← Back to the vision · Bibliography · Architecture