# Process record

**This directory is the project's empirical record.** It holds what physically
happened — what was attempted, what came out, and what was concluded from it.

Every record separates two things that are easy to blur and expensive to
confuse:

- **OBSERVED** — what physically happened. A part leaked. A part did not fit.
  A file contains a particular byte at a particular offset.
- **INFERRED** — somebody's explanation of it. A root cause, a mechanism, a
  causal link between a failure and a rule that later appeared in code.

An inference can be right and still be an inference. Where a record marks
something INFERRED it is not doubting it, it is recording that no measurement
isolated it.

---

## The records

| # | record | outcome | date |
|---|---|---|---|
| 01 | [A 70 µm membrane leaked](record-01-membrane-leak.md) | printed failure | before 2026-05-09 |
| 02 | [The first centerfed block print did not mate](record-02-interface-mismatch.md) | printed failure | before 2026-05-11 |

**Correction, 2026-09-29:** the single-pump cell (`/hardware/#process-behaviour`, §6.6) is not the only print. By the
lab's account, a collaborator at BYU printed five pumps on the BYU printer in
June 2026, and the lab tested them (`/hardware/#what-has-been-printed`).

`artifacts/` holds the primary sources these records are read from.

---

## Why this directory exists

These findings existed **only in untracked working notes** until 2026-09-11.
Such a note is not versioned as a record, nothing indexes it, and nothing
notices when it is deleted or rewritten. Record 02 is the clearest case: a
physical print failure — the thing that caused the six +Z fluid cups now
specified in `physical_interface.json` — was held in a note that a routine edit
could have erased, and nothing would have flagged it. A finding that lives only
in such a note has not been recorded; it has been mentioned.

A finding is recorded when it is in this directory.

---

## How to add a record

One file per record, named `record-NN-<slug>.md`. State, in this order:

1. **What was attempted** — the geometry, the settings, the intent.
2. **What happened** — marked OBSERVED. The physical outcome, including failure.
3. **Why** — marked INFERRED, if anyone has an explanation. Say plainly whether
   it was measured or reasoned.
4. **What was concluded** — and whether that conclusion reached the code.
5. **The date**, or the tightest bound available.
6. **The source** it was recovered from, as a path.

Quote the substance of the source rather than paraphrasing it. A paraphrase
loses the hedges, and the hedges are usually the most honest part.

Then add a row to the table above, and cite the new path from
`/hardware/` wherever the figure appears, so the page's LOCAL tier points at
this directory rather than at a working note.