Skip to content

Course draft 2: class 1.3, routings, filmed - #255

Merged
kalwei merged 4 commits into
mainfrom
feat/course-draft2-class-13
Nov 15, 2026
Merged

kalwei merged 4 commits into
mainfrom
feat/course-draft2-class-13

Conversation

@kalwei

@kalwei kalwei commented Nov 15, 2026

Copy link
Copy Markdown
Contributor

What

Class 1.3, Routings — the path a material takes, is the first class of module 1, the core. It is filmed (292 s, 1280×720, no audio).

  • scripts/scenes/course/02-03-routings.toml. It opens on a word card of a fork's road (mould, bin, degate, mark) and the fork line on the floor. Then the ideas: operation, sequence, routing, and "the recipe of steps, not of ingredients". Then the plant's routings table in Admin, filtered to its four pallet routings. Then RT-PALLET is typed into the form: 10 Palletizer on PAL1_Palletizer, 20 StretchWrap on PAL1_StretchWrap. Each step shows cost center CC-PAL01, read from the machine's line, not typed. The form refuses two things, a routing with no operation and a code already in use, and the class films both. Three "you now know" lines close it.
  • scripts/course/course.toml: 1.3's outcomes and scene, and nothing else.
  • scripts/record-scene.py learns to fill in a form:
    • fill types into a field one key at a time, and choose picks an option from a list.
    • Before a click, fill or choose, it scrolls the control clear of the caption bar. The Add operation button sat under the bar and the first look take could not press it.
    • The product's toast is lifted above the caption bar, which otherwise covers it.
  • Tests:
    • tests/test_class_1_3_defines_a_routing_whose_steps_bring_their_cost_centers.py (16) holds the scene to the outline.
    • tests/test_a_scene_can_type_into_a_form_and_choose_from_its_lists.py (5) covers the two new verbs and the lifted toast.

Why

It is the 1.3 handoff of course draft 2. A routing is where the floor's work starts, so the core opens with it.

How verified

  • Filmed with record-scene.py on a scratch cutlery plant (labs/cutlery, 10×) on 127.0.0.1:8040. The plant was stopped and its home removed afterwards.
  • The badge is in 5845 of 5845 mp4 frames and 258 of 258 GIF frames. The teaser GIF is 25.8 s. I looked at all 14 stills, and the bottom-right corner is empty in every one.
  • Regenerated with --course, course.json shows 7 modules and 25 lessons, with 0.1, 0.2 and 1.3 recorded. The already-filmed folders are byte-for-byte unchanged (76 files checksummed before and after).
  • The class tests and the recorder's tests pass: 163 fast, plus 10 in the browser tier.
  • In the fast suite, one test fails: test_the_calibration_the_supervisor_records_is_on_the_gauges_own_register. It is the known failure on clean main after 18:00 Central (the date in UTC has turned), and this PR does not cause it.
  • Namegrep over the diff: 0 hit(s) over 3333 names.
  • Windows: there is no subprocess, console output or signal in the new tests. Paths go through pathlib.

What the class shows in passing, filmed as the product draws it:

  • After a create, the form's machine list starts on the company itself ("CUTLERYCO, cost center unknown"), and it offers areas and lines as well as machines.
  • The material list has no blank.
  • The code and name inputs are narrow and cut off what is typed.

🤖 Generated with Claude Code

The first class of the core. A fork's road told as a word card, then the
three words (operation, sequence, routing) and the recipe of steps, not of
ingredients. Then the plant's routings in Admin. RT-PALLET is typed into
the form, and each step's cost center is read from its machine's line. The
two things the form refuses close it out. Filmed on a scratch cutlery plant
on loopback at ten times.

The recorder learns to fill in a form: `fill` types into a field one key
at a time, and `choose` picks an option from a list. It scrolls a control
clear of the caption bar before it touches it. It also lifts the product's
toast above that bar, which otherwise covers it.

course.toml gains 1.3's outcomes and its scene, and nothing else.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: scottkalw <scottkalwei@gmail.com>
@github-actions

Copy link
Copy Markdown

Thanks for the first pull request. Checks: every commit signed off (git commit -s), ruff and pytest green, and the house-rules checklist in the template. The maintainer reviews within a few days.

kalwei and others added 3 commits November 14, 2026 20:54
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: scottkalw <scottkalwei@gmail.com>
Signed-off-by: scottkalw <scottkalwei@gmail.com>
Signed-off-by: scottkalw <scottkalwei@gmail.com>
@kalwei
kalwei merged commit e2a50af into main Nov 15, 2026
27 checks passed
@kalwei
kalwei deleted the feat/course-draft2-class-13 branch November 15, 2026 05:21
kalwei added a commit that referenced this pull request Nov 15, 2026
Class 1.4 of the course's second draft: a work order's day, filmed on a
scratch cutlery plant.

**Stacked on #255.** #255 (class 1.3) was still open when this branch
was made, so its two commits are the first two here. Once #255 merges I
will rebase this onto main and they will drop out. The 1.4 work is the
six commits after them.

## What

- `scripts/scenes/course/02-04-a-work-orders-day.toml` is the new scene
(lesson 1.4, about 5 minutes). It opens with a card on a bin and the
four words: order, release, dispatch, finish. Then it:
- types a new fork order on the Orders page, leaving it planned, then
releases it;
  - finds it in FORK1_Mold's queue on the station screen;
- reads an order's steps and shows that its count is the last step's,
not the sum;
- shows that the plant's own WO-FORK1-0001 books past its 2,778 and
stays open (decision 0029);
  - has a person finish it step by step;
- shows the confirmation waiting in the ERP outbox, and the completed
order with its over-run.
- `scripts/course/course.toml` adds 1.4's three outcomes and points
`scene =` at the new file. The draft-1 scene stays on disk.
-
`tests/test_class_1_4_releases_an_order_and_a_person_not_its_count_finishes_one.py`
has 18 tests, in the shape of 1.3's.
- **Recorder fix.** `clear_of_caption` looked its selector up with
`document.querySelector`, which throws on Playwright's `:has-text(...)`,
so the first take stopped on the Release button. It now uses
Playwright's own locator. A new browser test,
`tests/test_a_scene_brings_a_button_named_by_its_words_clear_of_the_caption_bar.py`,
fails on the old code and passes on the new.

No product code changed.

## How it was verified

- **Filmed** (take 5 of 5): 1280×720, 301.5 s, no audio, plus a 25.4 s
teaser GIF. The badge is on every frame (6,030 of 6,030). I looked at
all 14 stills. The plant ran at 127.0.0.1:8040 / 4844 and was stopped
afterwards, with its home removed.
- **course.json**: 7 modules / 25 lessons; exactly 0.1, 0.2, 1.3 and 1.4
are `recorded`. Before and after checksums of the course tree: only the
new 1.4 folder and course.json changed.
- **Tests**: the scene's tests and the recorder and course tests pass,
browser tier included (120 tests). The fast suite passes except
`test_the_calibration_the_supervisor_records_is_on_the_gauges_own_register`,
the known failure after 18:00 Central (the run was at 22:49 CST). Ruff
is clean.
- **Windows**: the new tests use pathlib, start no subprocess and send
no signals. The recorder does not run on Windows CI.
- **namegrep** over the diff: `0 hit(s) over 3333 names`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Signed-off-by: scottkalw <scottkalwei@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
kalwei added a commit that referenced this pull request Nov 15, 2026
Class 2.1 of the course, draft 2: the station screen, one machine at
arm's length. This adds its scene, its tests, its three outcomes and its
`scene =` line in `course.toml`, and one recorder fix. It is the fifth
class filmed.

**Stacked on #256.** This branch starts from
`feat/course-draft2-class-14`, because #256 is still open (#255 has
merged). The four commits after #256's tip are this PR. Merge #256
first.

## What the film shows

Scratch cutlery plant (`labs/cutlery`) on loopback, 10× replay.
FORK1_Mold's station screen, one panel per caption:

1. state;
2. the queue;
3. booking 500 good / 3 scrap against WO-FORK1-0001, with the queue's
count rising;
4. issuing 2 kg of the order's own resin lot, LOT-PP-FORK1-001 (4,485.5
left);
5. a length reading of 165.1 mm, in spec, beside the simulated floor's
readings;
6. maintenance owed: "Nothing owed on this machine." (the pack has no
maintenance plan).

**What it refuses:** WO-FORK1-0002, on the same mould but only planned,
is not in the Against list. Nothing can be booked against it.

**What it does not refuse:** a typed 5,000 good, more than the counter
has seen, is booked as typed, and the queue then reads 6,597 of 2,778.
The scene says so in its captions and in `honesty`. No product code
changes.

## Recorder fix

`scroll_to` puts a row `offset` px from the top of the window as near as
the page allows. A row at the foot of a short page cannot get there, and
it was left under the caption bar ("Nothing owed on this machine" in
take 2).

The recorder now adds room below the page, the way `clear_of_caption`
does, and lifts the row clear, never higher than the asked-for offset.
Rows that already reached their offset do not move.

New test:
`tests/test_a_scene_scrolled_to_the_foot_of_a_short_page_lifts_the_row_clear_of_the_caption_bar.py`.
It fails with the fix stashed.

## How it was verified

- **Filmed** to
`~/Work/gifs/course/03-the-floor/01-the-station-screen/`.
  - 1280×720, 317.7 s, no audio, GIF teaser 28.7 s.
  - Badge on 6,354/6,354 mp4 frames; corner check clean.
  - Every one of the 13 stills looked at.
- Four takes; three of them caught real faults: a card over the corner,
captions flashing, the queue out of frame, a row under the caption bar.
- **Filmed on the commit before the rebase onto #256's new tip.** That
rebase brings in #253 (3D models and layout for the cutlery pack). It
changes nothing the station screen draws.
- **`course.json`:** 7 modules / 25 lessons; exactly 0.1, 0.2, 1.3, 1.4
and 2.1 are `recorded`. Outside the 2.1 folder only `course.json`
changed (checksums before and after).
- **Tests:**
  - Class 2.1's 20 tests pass.
  - The scene, course and recorder tests pass under `-m browser` (94).
- In the fast suite the only failure is the known
`test_the_calibration_the_supervisor_records_is_on_the_gauges_own_register`,
which fails on clean main after 18:00 Central.
  - `ruff check .` is clean.
- **Name check:** `namegrep.py` over the diff: `0 hit(s) over 3333
names`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Signed-off-by: scottkalw <scottkalwei@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant