Design record · not part of the site

The two views on HAER sheet 4, cut into separate assets and put where each one's proportions want to go: the deck plan turned bow-up down the right column, the inboard profile level across the foot. Mock of /architecture; nothing installed.

Cut rather than CSS-cropped from one sheet, for three reasons that all matter at page load: the rail pulls 212 kB instead of dragging in 555 kB of sheet to show a quarter of it; the rotation is baked in, so there is no transform for a blend to get trapped inside; and each asset carries intrinsic dimensions, so the column and the band reserve their space before the image arrives — no layout shift.

LatentWinds — Agentic Systems Shipyard Architectural Work

Drawings · Principles · Evidence

Begin with the system, not the spectacle.

This section is designed for engineers who are interested in AI but unwilling to suspend the disciplines they already trust.

Plate E.1

The Engineer’s Path

A staged route from familiar system boundaries to responsible use of models and agents.

Plate E.2

Identity and Scope

Workload identity, authority and least-capability participation before conversational freedom.

Plate E.3

Evaluation and Evidence

Define success, test actual workflows and inspect failures before increasing autonomy.

Plate E.4

Provenance

Trace who acted, what evidence was used, what changed and what followed.

Lower deck plan of a fishing schooner seen from above, bow uppermost: forecastle, main hold, engine room and captain's cabin divided by bulkheads, with framing drawn along both sides.
Lower deck plan · HAER MA-168
Schooner Ernestina, 1894
Inboard profile of a fishing schooner: the hull cut lengthways to show captain's cabin, engine room, main hold, galley and forecastle, with masts, keel and framing.
Inboard profile · Historic American Engineering Record MA-168 · Library of Congress · public domain
LatentWinds · Portland, Maine Captain’s Log · Observatory · Contact

What to look for

The rail is sticky, so the vessel is still there at the bottom of the reading. It holds still while the text moves, which is the opposite of a parallax trick — nothing animates, and there is nothing here for reduced-motion to switch off.

The rail dies below 900px and is removed rather than shrunk. A 1:4 drawing at phone width is 1300px of column with nothing beside it. The band survives every width, because a band is the one shape that does not care how narrow the page gets — which is an argument for the band being the safer of the two commitments.

The band becomes a silhouette at phone width, and this is the one thing the arrangement does not solve. At 390px the profile is about 300px across: the vessel's line survives, every internal detail does not. It still works as a closing device — the detail is not what a band above a footer is for — but nobody should install it believing a reader on a phone can see the galley. The alternative is a horizontally scrolling band, which is worse: a footer is the wrong place to ask someone to drag.

Both drawings are the same vessel, which is the quiet part. The band shows her cut lengthways and the rail shows the same hull from above; a reader who notices that the forecastle is forward in both has been told something true about drawings, on a page whose subject is drawing systems before claiming them.