LASER LESSONS (ComMarker B4 60W MOPA + LightBurn)

Source: bot-context.md

Laser Bot Operator Handbook (public)

Read this first, then the lessons page (/lessons). Be token-efficient. The owner presses Start. The bot never does.

1. Machine

• ComMarker B4 60 W MOPA fiber (JPT-class source), JCZ USB-Lite controller, LightBurn 1.7.08.
• 110 mm lens, 110x110 field, correction file jcz110.cor. Job centre = 55,55. 63,63 is obsolete.
• Galvo calibration was finalised 2026-10-15. The calibration kit and its history live in the calibration folder (B4_CALIBRATION_KIT).

2. Loading a file into LightBurn safely

1. Check the LightBurn window title. A * means unsaved changes. Ask the owner before discarding them, unless the owner names that project to discard.
2. If LightBurn has to restart, find its PID with ps + awk and kill only that PID. Never pkill -f, because it can match other processes.
3. Relaunch detached (nohup setsid lightburn <file> &) so it survives the shell.
4. Never press Start. Never touch LightBurn while a job is burning. If the owner says they will load the file, don't close or relaunch anything.
5. "Cut Selected Graphics" is on, so the owner presses Ctrl+A before Start.

3. Folder layout (laser root)

• coins/ coin jobs (recovery, brass, lion_portrait, ...). The bot's work goes in grok/ subfolders.
• washers/ washer designs and colour tests.
• colortest/ colour calibration (ss245_rgb/ = 8-face RGB tester).
• tools/ scripts (below). libraries/ LightBurn libraries (grok/HD_fill_black.clb).
• calibration/ galvo and colour calibration kits. pre55/ backups from before the 55,55 recentre and old ss245 outputs.
• LASER_LESSONS.md (lessons, read before every job), INDEX.md / MANIFEST.csv (file index).

4. Tools

• recenter55.py recentres projects from 63,63 to 55,55 (keeps <name>.pre55.lbrn2). --dry audits without writing.
• to_ss245.py <file|folder> makes <name>_ss245.lbrn2 (24.5 mm polished stainless, centre 55,55, colours from the palette, deep stacks collapsed into one layer per colour). --map PATTERN=key overrides a role. --hd-black writes *_ss245_hdblack.lbrn2.
• est_time.py <files> gives a rough burn time (filled area / interval / speed x passes x crosshatch + 15%).
• dump_layers.py <folder> writes a _LAYERS.txt for every .lbrn2.
• brass_rules.py apply() adds the brass CLEAN/POLISH layers idempotently.
• ss_color_palette.json holds the stainless colour presets plus verification status and observed large-area colours.
• gen_ss245_rgb.py (in colortest/ss245_rgb) regenerates the RGB tester and its plan CSV.

5. Design rules

• The design (letters/logo) is RAISED and unburned. The field inside the work circle is DEEP-ETCHED. Outside the work circle is transparent: nothing burns there, and no background, glow or haze.
• Brass recovery coin: 39.5 mm with a lip, so the work circle is 37.5 mm.
• Traces: copy clean shapes and don't redraw. Remove specks and glow, and keep clean gaps. QC whole letters, bottoms included, against the source.
• Every .lbrn2 gets a matching _LAYERS.txt and an output-off FRAME layer.
• Never edit another agent's (Claude's) files. Write new files and keep the originals.
• Christian designs: Baptist-friendly, no crosses, understated.

6. Material rules

Brass/copper deep engraving: insert a CLEAN (20 %, 1500 mm/s, 100 kHz, 200 ns, 0.01 x2, crosshatch) after every 5 deep passes. The LAST layer is a full-surface POLISH (18 %, 2000 mm/s, 350 kHz, 10 ns, 0.01). Never use these on stainless, polished or anodized pieces.
Black-field brass order: (1) deep engraving + CLEANs + POLISH; (2) BLACK on the field only (40 %, 500, 40 kHz, 200 ns, 0.02 x2, untested); (3) final LETTER CLEAN on the raised letters and rings only (brass CLEAN settings, x1).
Stainless colour = surface mark: use one lean layer per colour, with a separate layer for each shade. Never copy brass deep stacks.
- HD fill black (14 purple layers + 2 blue facets, ~130 min per coin, glossy black) is used only on purpose.
- For black fields, use 4 passes of the HD-black purple setting (15 %, 567, 280 kHz, 80 ns, 0.004, crosshatch x2, angles 0/45/90/135), about 14 min.
- The hard blacks (20 %/567 and 30 %/300/200 ns) spark and burn on fields. Keep them for small marks and lines.
- Big areas run hotter and paler than test cells. Keep colour tests at 30 % or less.
- Anodized aluminium black: 12 %, 1500 mm/s, 50 kHz, 0.045, 1 pass. Brass colours are unverified, so flag heat and cool between passes.

7. Verified stainless colours (polished, by eye)

Preset	Settings (crosshatch, x2)	Test cell	Large field (2026-10-15)	Mapping confidence
purple R1-0106	15 %, 567, 280 kHz, 80 ns, 0.004	deep purple-black	lavender / light mauve	medium
deep blue R1-0805	7.5 %, 700, 119 kHz, 60 ns, 0.005	deep blue	slate blue-grey	medium-low
blue R1-0904	15 %, 1200, 230 kHz, 60 ns, 0.005	good blue	pale grey-green / silver	medium
black (washer)	20 %, 567, 280 kHz, 80 ns, 0.004	black	brown-black, sparks, avoid on fields	high
black lines	30 %, 300, 50 kHz, 200 ns, 0.01	dark lines	brown-black, sparks	high
gold R1-1202	48.4 %, 317, 61.5 kHz, 60 ns, 0.0287	untested	never fired (layer was off)	n/a
Energy-density model: E = W / (speed x interval) J/mm², W = power% x 60 W. Dose = E x 2 (crosshatch) x passes.

8. Open work

• The RGB 8-face tester (4 coins, 26 cells per face, plan CSV, ~21 min total) is ready and not burned.
• The colorfield v2 coin (gold removed, black wedges = 4-pass HD black, ~14 min) is ready and not burned.
• Gold is still untested. Whether purple came out lavender and deep blue came out slate, or the other way round, is uncertain; a single swatch of each would confirm it.

9. Site (laserview.hankelsner.tech)

• /rules/ — public rules tree. HTML pages live at /rules/<page>/; raw markdown at /rules/<path>.md; machine index at /rules/index.json.
• /lessons — combined lessons HTML; raw combined dump at /lessons.md.
• /color/ — MOPA colour grids and data (b4_measured.json, color_settings.json, rounds).
• /projectaddons/ — motorized Z / camera / Pi+Pico add-ons (older copy also at /workup/).
• /settings/ — material settings index; /tools/ — tools index.
• /plates/ — colour calibration plates; public JSON at /api/materials, /api/plates, /api/cite (see Colour plates API). Unified browser on /color/ and /colors/; cite Plate 31, A8.
• Source markdown on the VPS: /srv/web/laserview.hankelsner.tech/rules-src/. Publish with laser-rules-build.
• The site is hosted by the VPS bot. Send content to the VPS agent, which publishes it. The laser bot doesn't edit the site directly.

Colour DB v3 (cut settings + sub-layers)

• One colour = one LightBurn layer; multi-pass stacks use sublayers[] (HD fill black = 16).
• Full Cut Settings Editor fields live in cut_settings + each sub-layer. Site browsers show/filter/edit them (Hank-gated writes).
• GET /api/library.clb exports one Entry per colour with sibling CutSettings per sub-layer.
• Sub-layer .clb/.lbrn2 serialization is UNVERIFIED until Hank does one LightBurn open-test. >11 sub-layers may need a split. Do not present exports as verified.

Source: machine.md

Machine

• ComMarker B4 60W MOPA, JCZ USB-Lite, LightBurn 1.7.08.
• 110 mm lens: field 110×110, jcz110.cor, job center 55,55 (63,63 is obsolete).
• Galvo values live in LightBurn prefs.ini; when a .cor is loaded, bulge/skew/trapezoid show greyed out but still apply.

Source: workflow.md

Workflow

• Fix root causes (calibrate the galvo) rather than working around them.
• Every project gets an output-off FRAME layer for alignment.
• Every .lbrn2 ships with a <name>_LAYERS.txt listing each layer's settings, made by tools/dump_layers.py <folder>.
• Each AI's work goes in its own subfolder (grok/, claude/, hank/); never edit another AI's files.
• A human presses Start; bots never do.
• Never discard an open project with unsaved changes unless the owner names it.

Source: design/coins.md

Coins / emblems (get the negative right)

• The DESIGN (letters/logo) is RAISED: the highest point, unburned (white = no burn).
• The FIELD inside the work circle is DEEP-ETCHED (black = deepest, most passes, rotating hatch).
• OUTSIDE the work circle is fully transparent and nothing burns: no background image, glow, haze or fill.
• Example: a 39.5 mm brass coin with a lip leaves a 37.5 mm work circle.
• Tracing a photo logo: copy a clean trace of the shapes and don't redraw or get fancy. Remove glow, streaks and specks (especially at letter feet). Keep clean gaps between letters and rings.
• QC: zoom in on every whole letter, bottoms included, against the source before burning.
• Hiding an old mark: deep coarse crosshatch with a large dot/line width in the field.
• Low-poly art: burn the facet edges as thin lines (mesh look).
• Black-field brass coins: deep + polish, then field-only black mark, then letter-only final clean — full layer order and settings in Black-field brass coins (UNTESTED).

Source: materials/brass-copper.md

Brass / copper deep engraving

ONLY these materials for this deep-etch recipe; never stainless, polished or anodized.

• After every 5 deep-etch passes, add a CLEAN layer: 20%, 1500 mm/s, 100 kHz, 200 ns, 0.01 mm, bidirectional crosshatch, 2 passes, whole coin.
• The LAST layer is a full-surface POLISH (output on): 18%, 2000 mm/s, 350 kHz, 10 ns, 0.01 mm, letters included, covering the entire coin to even out edges and clear debris.
• Sources: the ComMarker B6 MOPA brass blog and lahobbyguy t=8237.
• Deep brass settings are UNTESTED. Heat builds fast, so cool between passes.

Black-field brass coins

Layer index is run order (UNTESTED):

1. All deep engraving — CLEAN after every 5 deep passes, ending with the full-surface POLISH (settings above).
2. BLACK mark on the etched field only: 40%, 500 mm/s, 40 kHz, 200 ns, 0.02 mm, ×2 crosshatch (source omglaser.com/laser-settings #59).
3. FINAL LETTER CLEAN on the raised letters and rings only: 20%, 1500 mm/s, 100 kHz, 200 ns, 0.01 mm, ×1 (source blog.commarker.com/archives/49730). Brightens letters and removes black overspray. Never run it over the black field.

Source: materials/stainless-polished.md

Polished stainless

• MOPA colours: only from eye-verified test cells, and only on polished stainless. Plain stainless goes black instead.
• Large areas run hotter than small test cells.
• See also Color and the live /color grid data.
• Stainless colour = surface mark: use one layer per colour with the verified pass count, and never copy brass deep-pass stacks. The exception is the verified HD preset below.

High-definition fill black (verified, Hank)

Opt-in via to_ss245 --hd-black (lean version is the default). Very good deep glossy black finish (verified by Hank), but slow.

1. 14 identical R1-0106 field layers run in order: 15%, 567 mm/s, 280 kHz, 80 ns, 0.004 mm, crosshatch, ×2; angles 25, 163, 64, 74, 117, 132, 151, 155, 27, 154, 6, 86, 176, 147.
2. 2 R1-0904 blue facet layers: 15%, 1200 mm/s, 230 kHz, 60 ns, 0.005 mm, ×2, 45°/135°.

Source: materials/stainless-plain.md

Plain stainless

• Black: 20%, 567 mm/s, 280 kHz, 80 ns, 0.004 interval, 2 passes (verified).
• Lines: 30%, 300 mm/s, 50 kHz, 200 ns, 2 passes (verified).

Source: materials/anodized-aluminum.md

Anodized aluminum

• Black anodized aluminum: 12%, 1500 mm/s, 50 kHz, 1 pass, 0.045 mm interval (verified).

Source: materials/color.md

MOPA colour on polished stainless: notes and RGB test plan (2026-10-15)

Machine: ComMarker B4 60 W MOPA, 110 mm lens, LightBurn 1.7. Material: 24.5 mm polished stainless coins.

Colour model

Oxide-film interference; thickness sets colour (straw/gold, brown, purple, blue, then 2nd order pale gold, red/magenta, green, then grey/black). Main control: energy density E = W/(speed x interval) J/mm², W = power% x 60 W; crosshatch and passes multiply dose. Pulse width and frequency set pulse energy (W/f) and peak power; longer pulses run warmer. High pulse energy or slow high E causes sparks/burning. Large fills run hotter than small test cells.

Verified by eye

• Purple R1-0106: 15%, 567 mm/s, 280 kHz, 80 ns, 0.004, crosshatch x2 (E≈3.97)
• Deep blue R1-0805: 7.5%, 700, 119 kHz, 60 ns, 0.005, x2 (E≈1.29)
• Blue R1-0904: 15%, 1200, 230 kHz, 60 ns, 0.005, x2 (E≈1.50)
• HD fill black: 14 purple-setting layers at rotating angles + 2 blue facet layers; glossy black, ~130 min per coin.

Lessons 2026-10-15

• Hard black (20%, 567 mm/s) sparked and burned on a colour-field coin. For black fields use 4 passes of the HD-black purple setting at 0/45/90/135 (~14 min per coin).
• Gold R1-1202 was output-off in that burn and never fired, so gold is UNTESTED (it did not fail to show).

Verified by Hank 2026-10-15 on large field wedges (24.5mm polished SS)

• Lavender/light purple: 15%, 567 mm/s, 280 kHz, 80 ns, 0.004, crosshatch x2
• Slate blue-grey: 7.5%, 700 mm/s, 119 kHz, 60 ns, 0.005, x2
• Pale grey-green/silver: 15%, 1200 mm/s, 230 kHz, 60 ns, 0.005, x2
• Brown-black with sparks/burned edges: 20%, 567, 280 kHz, 80 ns, 0.004 x2, and 30%, 300, 50 kHz, 200 ns, 0.01 x2. Avoid these on fields.

Note: big areas came out much paler than the small test cells (large fills run hotter). The purple vs slate-blue mapping could be swapped (medium-low confidence).

Published starting points (other machines; scale before use)

• Red: 45%, 1000 mm/s, 60 ns, 400 kHz, 0.02 (30 W JPT, 110 mm) hobbyistlaser.com/blog/color-marking-stainless-mopa-settings
• Gold: 35%, 1000, 80 ns, 800 kHz, 0.02 (same)
• Blue: 40%, 1500, 100 ns, 600 kHz, 0.02 (same)
• Red: 27%, 800, 300 kHz, 60 ns, 0.003 (60 W JPT, 210 mm) forum.lightburnsoftware.com/t/experiences-with-color-marking/125488
• Green: 200 ns, 70 kHz, 25%, 600 mm/s (60 W JPT) anypowerrun.com/post/detail/1313/
• Gold bands 140-200 ns 200-260 kHz; green 420-500 ns 40-60 kHz (20 W) stylecnc.com/blog/mopa-color-laser-engraving-settings-stainless-steel.html
• Line interval matters more than speed/power; red and green are hardest on 110 mm: forum.lightburnsoftware.com/t/mopa-colours-hunting-for-a-good-red-and-green-on-brushed-stainless/136015
• Power % scales with wattage; ns, kHz and speed don't: mopacolorstudio.com/utilities/mopa-color-reference-chart.html

Scaling: halve power % of a 30 W recipe for 60 W; 110 mm lens has a smaller spot, so start about x0.8 and sweep.

RGB calibration test (8 faces = 4 coins front and back, 26 labelled 2.2 mm cells each, power 25% or less)

F1 E sweep 0.5-16, 80 ns, 1 pass. F2 same, 2 passes, E 0.5-8. F3 E 0.5-16, 60 ns, 230 kHz. F4 forum-red geometry, E 2.5-16, 1 and 2 passes. F5 pulse width 2-200 ns. F6 frequency 30-800 kHz at 60/200 ns. F7 200/350/500 ns warm/gold plus published gold and green. F8 verified colours at x0.7/1/1.4/2 plus published reds. About 21 min total. Results get photographed under fixed white light and fitted to colour = f(dose, ns, f).

Source: color.md

Color

MOPA color work is only meaningful on polished stainless, with eye-verified test cells. Unpolished stainless goes black; large areas run hotter than small cells.

Notes and test plan

• MOPA colour on polished stainless — verified colours, lessons, published starting points, RGB calibration plan

Live data on this site

• /color/ — color grid pages, rounds, charts
• /color/color_settings.json — settings JSON
• /color/color_settings.csv — settings CSV
• /color/b4_measured.json — measured results
• /color/rounds.json — round index
• /tools/ — tools index (includes color chart assets)

Related rule: Polished stainless.

v3: one colour = one layer; multi-pass stacks are sub-layers. Full cut_settings shown in the browsers. Library/.clb sub-layer export is unverified until a LightBurn open-test.

Source: api.md

Colour plates JSON API (v3)

Single live colour database for /color/, /colors/, /plates/, /tools/ and /tech/. Machine shown everywhere: ComMarker B4 60W MOPA, 110 mm lens, LightBurn 1.7.08 (other charts only as “adapted from”).

Public reads

• GET /api/materials — materials + all colours
• GET /api/materials/{id}/colors — one material (supports filter query params)
• GET /api/colors / GET /api/colors/query — unified list with filters:
• confidence=hank_verified,high · material= / materials= · status= · vivid=1|0
• mode=fill,line,image · wobble=1|0 / wobble_enable= · fill_grouping=1|0
• ranges: power=10..40 · speed= · freq= · pulse= · interval= · passes=
• dot_width= / dot_width_mm= · line_width= · line_interval=
• sort=confidence or power_pct:asc / dot-width / line-width / mode …
• q= text search · plate= · group= (client-side also)
• GET /api/colors/{id} — one colour (full cut_settings + sublayers[])
• GET /api/plates · GET /api/plates/P031
• GET /api/cite?q=Plate%2031,%20A8 — also P5 A5, P031-A8 (includes cut_settings / sublayers)
• GET /api/legend
• GET /api/library.clb?... — LightBurn material library download (same filters as /api/colors). Content-Disposition: attachment; filename="library.clb". Sub-layer CutSetting siblings are UNVERIFIED until Hank does one LightBurn open-test.

Cut settings + sub-layers (v3)

• Every colour/cell has cut_settings (first sub-layer / primary) and sublayers[] with the full Cut Settings Editor field set (Fill / Line / Image). Unknown = null; defaulted fields listed in settings_defaulted.
• One colour = one layer. Multi-pass stacks are sub-layers inside that one layer (e.g. HD fill black = 1 entry with 16 sub-layers). UI shows Fill/Line/Image sections and sub-layer tabs; warns when sub-layers > 11.
• lines_per_inch and min_tabs_per_shape are DB/site only (no LightBurn XML element). Do not promise they round-trip.
• Library export: one <Entry> per colour, one sibling <CutSetting> per sub-layer. Serialization shape is a guess — not verified in LightBurn 1.7.08.

Hank-only writes (accounts gate laserview-admin + user hank)

• POST /api/colors/{id}/save — body {color: {...}} full record (incl. cut_settings / sublayers); auto confidence=hank_verified; timestamped backup + audit log
• POST /api/plates/P###/cells/A8/save — {cell:{...}} full cell save
• POST /api/plates/P###/save — full plate object
• POST /api/materials/save — full materials.json
• Photo propose/confirm endpoints unchanged
• Logged-out writes return 401

Assets

• One photo: /plates/P031/photo.jpg
• Test files: /plates/P010/<file>.lbrn2
• Legend: /color/LEGEND.md · /api/legend
• Docs: /docs/VPS_UPDATE.md · /docs/EXPORTER.md
• b4_measured.json is regenerated from this DB (not a divergent cache)

UI

• /color/ and /colors/ — identical unified browser (filters/sort/group in URL query; cut-settings panels + sub-layer tabs; library download buttons)
• /plates/ — plate list + same browser; /plates/P###/ — grid + photo + lbrn2
• /tech/ · /tools/ — same DB/API

Filter UI (2026-10-16 follow-up)

Categorical query params accept comma lists (OR within a param, AND across params):
confidence, material/materials, status, mode, plate, vivid (1/0), wobble, fill_grouping.
Shared browsers use checkbox multi-select with Select all / Clear per group.

Source: tools.md

Tools / converters

• dump_layers.py <folder> writes <name>_LAYERS.txt for each .lbrn2 in the folder.
• recenter55.py moves 63,63 projects to 55,55 (backups go in a pre55/ folder; use --dry to audit).
• to_ss245.py <file|folder> makes 24.5 mm polished-stainless copies (<name>_ss245) mapped to ss_color_palette.json (verified: purple R1-0106, blue R1-0904, deep blue R1-0805, black lines; gold UNTESTED), strips the brass clean/polish layers, and takes --map PATTERN=key overrides. Pass --hd-black for the verified 14× R1-0106 HD fill black (slow; lean default otherwise).

• Washer no-hole method: remove the hole from the fill/strip/frame layers and add a centre motif in the same style.

• Colour plates API: public GET /api/materials, /api/plates, /api/cite?q=; Hank-only photo upload and cell edits — see Colour plates API. UI at /plates/ and the material dropdown on /color/.

• Unified colour browser: /color/ and /colors/ (same DB). Filters in URL, e.g. ?confidence=hank_verified,high&sort=power_pct:asc. Cite Plate 31, A8 via /api/cite.

managed by laser-rules-build (/opt/laser-rules/build.py) · source editable via SSH as grokbot in rules-src · /rules/ · /lessons