simcon-toolkit 0.1.0__py3-none-any.whl

simcon_toolkit/_kit/templates/quoting/shop/commercial.md

@@ -0,0 +1,86 @@
+# Commercial — rates, allowances, margin, terms
+
+> Fictional example. Replace with your controlling department's actual figures.
+
+## The piece-price build-up
+
+We build a piece price in this order, and the quotation should show it in the same order:
+
+```
+material  (shot mass incl. runner share, minus regrind credit, plus scrap allowance)
++ machine (loaded hourly rate x cycle time / cavities / OEE)
++ labour  (operator share + any manual work)
++ secondary (packaging, QC, drying)
+= manufacturing cost
++ overhead (on material + labour only — the machine rate already carries it)
++ margin
+= piece price
+```
+
+**Overhead is not applied to the machine line.** The machine rates in [`machines.md`](machines.md)
+are loaded rates that already recover the shop's fixed costs. Applying overhead again is the
+classic double-count and it makes us uncompetitive by 10–15 % without anyone noticing.
+
+## Scrap and start-up
+
+- **Process scrap 2.0 %** on an established tool, **3.5 % for the first 3 months** of a new tool.
+  We quote the established number and note the start-up allowance separately.
+- **Start-up scrap** per production order: 15 shots, charged into the first order only.
+- Glass-filled grades get **no regrind credit** (see `materials.md`).
+
+## Labour
+
+One operator covers **4 presses** on average, so the operator share per machine-hour is
+`labour_rate / 4`. Manual work (deburring, insert loading, assembly) is quoted per part in
+seconds and is *not* covered by that share.
+
+## Packaging and freight
+
+Standard is a returnable KLT with interleaves; one-way cartons on request. Freight is EXW, so it
+is not in the piece price — but we state the packaging density because the customer's logistics
+cost usually depends on it.
+
+## Volume ladder and validity
+
+- Piece prices are quoted at **10k / 50k / 250k parts per year** unless the customer states a
+  volume. Two things actually move along the ladder, and only two: **setup amortisation** over the
+  order size, and the **material price break** we get on annual tonnage (3 % above 2 t/year, 6 %
+  above 10 t/year — what our own supplier gives us, passed through). Nothing else is
+  volume-dependent, so a ladder that is nearly flat between two rungs is a correct answer, not a
+  missing discount.
+- **Setup:** 3.5 h per production order at the machine rate, plus one drying charge. We plan up
+  to 12 orders a year but never below a **5 000-part minimum batch** — at low annual volume that
+  means fewer, larger runs, not twelve tool changes for a few hundred parts each.
+- **Validity 30 days.** Material prices are indexed to the polymer index at order date; a change
+  above ±5 % is passed through.
+
+```json commercial
+{
+  "labour_rate_eur_h": 38.0,
+  "machines_per_operator": 4,
+  "overhead_pct_on_material_and_labour": 12.0,
+  "margin_pct": 15.0,
+  "scrap_pct_established": 2.0,
+  "scrap_pct_startup": 3.5,
+  "startup_shots_per_order": 15,
+  "setup_hours_per_order": 3.5,
+  "packaging_eur_per_part": 0.012,
+  "qc_eur_per_part": 0.008,
+  "orders_per_year": 12,
+  "min_order_parts": 5000,
+  "material_price_breaks": [
+    {"min_kg_per_year": 0, "discount_pct": 0.0},
+    {"min_kg_per_year": 2000, "discount_pct": 3.0},
+    {"min_kg_per_year": 10000, "discount_pct": 6.0}
+  ],
+  "price_index_passthrough_pct": 5.0,
+  "amortisation_default_parts": 250000
+}
+```
+
+## Tool amortisation
+
+We show the tool as a lump sum **and** as an amortised cent-per-part figure over
+`amortisation_default_parts`, so the customer can compare us with a supplier who hides the tool
+in the piece price. We do not actually amortise unless the customer asks for it in writing —
+an amortised tool that never reaches its volume is a loss we cannot recover.

simcon_toolkit/_kit/templates/quoting/shop/lessons.md

@@ -0,0 +1,94 @@
+# Lessons — what we got wrong, and what changed because of it
+
+**The agent reads this file first and writes to it last.** Every quote ends with a review: the
+estimator corrects what the agent decided, says *why*, and the agent records it here — the reason
+as prose, the number in the `calibration` block below. `quoting/costing.py` reads that block, so
+a correction changes the **next** quote instead of being forgotten.
+
+Do not delete entries. A superseded lesson is history, and knowing that a factor moved twice in
+the same direction is itself information.
+
+## The calibration block
+
+These are the knobs the cost model exposes to this loop. Everything else lives in the other
+`shop/` files — change those directly when the *shop* changes (a new press, a new rate). Change
+**this** block when the *estimate was wrong*.
+
+```json calibration
+{
+  "clamp_safety_factor": 1.3,
+  "use_simulated_clamp_pressure": true,
+  "sim_peak_to_mean_pressure": 0.5,
+  "cooling_time_factor": 1.15,
+  "pack_time_s_per_mm_wall": 1.5,
+  "cycle_time_factor": 1.0,
+  "complexity_bonus": 0.0,
+  "tooling_price_factor": 1.00,
+  "piece_price_factor": 1.0
+}
+```
+
+| Key | Meaning | Moved by |
+|---|---|---|
+| `clamp_safety_factor` | multiplies the computed clamp force before picking a press | L-002 |
+| `use_simulated_clamp_pressure` | if false, clamp comes from the material rule of thumb alone | — |
+| `sim_peak_to_mean_pressure` | simulated **peak** cavity pressure → **mean** over the projected area | — |
+| `cooling_time_factor` | multiplies the analytic cooling time | L-001 |
+| `pack_time_s_per_mm_wall` | holding time per mm of **nominal** wall (gate freeze) | — |
+| `cycle_time_factor` | last-resort factor on the whole cycle | — |
+| `complexity_bonus` | added to the tooling complexity factor for every quote | — |
+| `tooling_price_factor` | multiplies the final tool price | — |
+| `piece_price_factor` | multiplies the final piece price | — |
+
+## Entry format
+
+```markdown
+## L-00N · YYYY-MM-DD · <part> — <one-line title>
+**Agent decided:** …
+**Corrected to:** …
+**Why (estimator's words):** …
+**Rule for next time:** …
+**Changed:** `calibration.<key>` A → B  (or: nothing numeric — judgement only)
+```
+
+---
+
+## L-001 · 2026-05-12 · bracket-A2 (PP TV20, 4 cav.) — cycle time was 12 % optimistic
+
+**Agent decided:** cycle 21.4 s, from the analytic slab-cooling time on the p95 wall thickness.
+
+**Corrected to:** 24.0 s. The tool ran at 24.3 s in production.
+
+**Why (estimator's words):** "The textbook cooling time assumes perfect contact and a cooling
+channel where you want it. On a ribbed part the ribs are the last thing to cool and we can't get
+a channel into the rib root, so real cooling is always longer than the slab formula. On thin
+flat parts the formula is fine — it's ribs and bosses that break it."
+
+**Rule for next time:** apply a cooling-time factor above 1.0 by default; revisit it per part
+family once we have three data points on flat parts vs ribbed parts.
+
+**Changed:** `calibration.cooling_time_factor` 1.00 → 1.15
+
+---
+
+## L-002 · 2026-06-03 · sensor-housing (PA6 GF30, 2 cav.) — clamp force undersized
+
+**Agent decided:** 1100 kN required → Engel victory 110, from the simulated peak cavity pressure.
+
+**Corrected to:** 1500 kN → Arburg 470 A 1500. The part flashed on the victory at trial.
+
+**Why (estimator's words):** "The simulation models one cavity with no hot runner, so it doesn't
+see the pressure drop through the manifold — the machine has to make that up and the cavity
+pressure at the end of pack is far above the fill pressure the AI shows you. For glass-filled PA
+we've always used 650 bar on the projected area and it has never let us down. Use the simulation
+to decide *gates*, not to decide *tonnage*."
+
+**Rule for next time:** clamp = max(simulation-derived, family rule of thumb) × safety factor,
+and the safety factor goes up, not down. The simulation number is a lower bound, never the
+answer. Say in the quotation which of the two governed.
+
+**Changed:** `calibration.clamp_safety_factor` 1.20 → 1.30
+
+---
+
+<!-- New entries are appended here by `main.py learn`. -->

simcon_toolkit/_kit/templates/quoting/shop/machines.md

@@ -0,0 +1,68 @@
+# Machines — the press list
+
+> Fictional example fleet. Replace with yours. The cost model picks the **smallest machine that
+> fits** (clamp force, shot volume, tie-bar opening) and bills its hourly rate.
+
+## The fleet
+
+| Press | Clamp | Shot (PS) | Screw | Tie-bar clear | Rate | Notes |
+|---|---:|---:|---:|---|---:|---|
+| Engel e-mac 50 | 500 kN | 62 cm³ | 25 mm | 320 × 320 mm | 42 €/h | all-electric, small precision work |
+| Arburg 370 S 800 | 800 kN | 118 cm³ | 30 mm | 370 × 370 mm | 48 €/h | workhorse, hydraulic |
+| Engel victory 110 | 1100 kN | 178 cm³ | 35 mm | 420 × 420 mm | 55 €/h | tie-bar-less — good for wide molds |
+| Arburg 470 A 1500 | 1500 kN | 257 cm³ | 40 mm | 470 × 470 mm | 68 €/h | all-electric, our best cycle stability |
+| KraussMaffei 250 CX | 2500 kN | 490 cm³ | 50 mm | 570 × 570 mm | 82 €/h | 2 units |
+| Engel duo 400 | 4000 kN | 1080 cm³ | 70 mm | 720 × 720 mm | 105 €/h | largest tool we can run |
+
+Rates are **machine + operator-share + energy + overhead recovery**, i.e. a loaded rate. They
+already include the shop's fixed-cost recovery, so do **not** add overhead on top of the machine
+line — overhead in [`commercial.md`](commercial.md) applies to material and labour only.
+
+## Rules the agent must apply
+
+- **Clamp:** the selected press needs `clamp_kN ≥ required_clamp_kN`. Required clamp comes from
+  mean cavity pressure × projected area of all cavities **plus the runner's projected area**,
+  times a safety factor. Never size a press at more than 80 % of its rated clamp — see
+  `max_clamp_utilisation` below.
+- **Shot:** the shot volume (all cavities + runner) must sit between **20 % and 80 %** of the
+  press's rated shot volume. Under 20 % the residence time degrades the melt; over 80 % there is
+  no cushion. A part that only fits above 80 % gets the next press up.
+- **Tie-bar:** the mold's plate footprint must clear the tie-bars. The victory 110 is tie-bar-less,
+  so it takes molds up to 700 × 700 mm despite the small clamp — that is why it is on the list.
+- **Two shifts** are the planning assumption (see `shop-profile.md`). A quote that needs a third
+  shift or the weekend shift is a flagged note, not a silent price.
+- If **no press fits**, say so plainly and quote the nearest external option as an assumption.
+  Do not silently scale the cavity count down — the cavity count is the customer's.
+
+```json machines
+{
+  "max_clamp_utilisation": 0.80,
+  "shot_utilisation_window": [0.20, 0.80],
+  "machines": [
+    {"name": "Engel e-mac 50",     "clamp_kN": 500,  "shot_cm3": 62,   "screw_mm": 25, "tiebar_mm": [320, 320], "tiebarless": false, "rate_eur_h": 42},
+    {"name": "Arburg 370 S 800",   "clamp_kN": 800,  "shot_cm3": 118,  "screw_mm": 30, "tiebar_mm": [370, 370], "tiebarless": false, "rate_eur_h": 48},
+    {"name": "Engel victory 110",  "clamp_kN": 1100, "shot_cm3": 178,  "screw_mm": 35, "tiebar_mm": [700, 700], "tiebarless": true,  "rate_eur_h": 55},
+    {"name": "Arburg 470 A 1500",  "clamp_kN": 1500, "shot_cm3": 257,  "screw_mm": 40, "tiebar_mm": [470, 470], "tiebarless": false, "rate_eur_h": 68},
+    {"name": "KraussMaffei 250 CX","clamp_kN": 2500, "shot_cm3": 490,  "screw_mm": 50, "tiebar_mm": [570, 570], "tiebarless": false, "rate_eur_h": 82},
+    {"name": "Engel duo 400",      "clamp_kN": 4000, "shot_cm3": 1080, "screw_mm": 70, "tiebar_mm": [720, 720], "tiebarless": false, "rate_eur_h": 105}
+  ]
+}
+```
+
+## Non-productive time per cycle
+
+Mold open + eject + close + part removal, added to fill + pack + cool to get the cycle. Robot
+take-out is standard on everything above 800 kN; the small presses drop parts to a conveyor.
+
+```json cycle
+{
+  "mold_open_close_s": 1.6,
+  "ejection_s": 0.8,
+  "part_removal_s": {"robot": 1.2, "freefall": 0.4},
+  "robot_from_clamp_kN": 800,
+  "oee": 0.85
+}
+```
+
+`oee` is the realistic availability factor applied to the theoretical cycle when converting to
+capacity and to the piece price — 0.85 is our two-shift long-run average on established tools.

simcon_toolkit/_kit/templates/quoting/shop/materials.md

@@ -0,0 +1,92 @@
+# Materials — house grades and our bias
+
+> Fictional example. Replace with your own grades and prices. This file is where the shop's
+> **material bias** lives: what we buy, what we know how to run, and what we avoid.
+
+## Our bias, stated plainly
+
+We are a **PP / PA / POM shop**. Roughly 45 % of our shots are polypropylene (mostly talc- or
+glass-filled automotive interior parts), 25 % polyamide (PA6 GF30 for under-hood and structural
+brackets), 15 % POM (gears, latches, anything that has to slide), and the rest ABS/ASA and TPE
+overmolds. We hold PP, PA6-GF30 and POM in silo or big-bag; everything else is bought per order
+with a 3-week lead time and a minimum of 500 kg.
+
+**What that means for a quote:**
+
+- If the part shape allows it and the customer has not specified a grade, **propose a house
+  grade** — we run it better and buy it cheaper. Say that we proposed it.
+- A non-house grade is quotable, but add the **procurement surcharge** and the lead time, and
+  flag the minimum order quantity — a 20k/year part in a bought-in grade carries a lot of
+  leftover material.
+- **Never silently substitute.** If the customer named a grade, quote that grade. If we would
+  recommend another, put it on the assumptions slide as a proposal.
+- Drying is not optional for PA and POM — it is in the piece price (see `drying_eur_per_kg`).
+
+## When the customer has not named a material
+
+Ask **once**, with a recommendation, and give the price impact. The question to ask:
+
+> "The geometry doesn't imply a material. We'd default to **PP TV20 (talc-filled)** for a part
+> like this — it's a house grade, ~1.85 €/kg, and it runs on our standard presses. If it needs
+> stiffness or heat, **PA6 GF30** at ~3.40 €/kg. Which should I quote?"
+
+If the user does not care, quote the house default (`default_grade` below) and mark it as an
+assumption. Do **not** stall the quote waiting for a material answer.
+
+## The grades we quote
+
+| House grade | Family | Density | Price | Stock | Catalogue query | Stand-in |
+|---|---|---:|---:|---|---|---|
+| PP Generic | PP unfilled | 0.91 g/cm³ | 1.55 €/kg | silo | `PP Generic` | — |
+| PP TV20 | PP + 20 % talc | 1.05 g/cm³ | 1.85 €/kg | silo | `PP T20` | `PP Generic` |
+| PP GF30 | PP + 30 % glass | 1.13 g/cm³ | 2.60 €/kg | big-bag | `AKROLEN PP GF30` | `PP Generic` |
+| PA6 GF30 | PA6 + 30 % glass | 1.36 g/cm³ | 3.40 €/kg | big-bag | `ALCOM PA6 900/8 GF10` | `PA66 Generic` |
+| POM C | POM copolymer | 1.41 g/cm³ | 3.10 €/kg | big-bag | `POM Generic` | — |
+| ABS | ABS | 1.05 g/cm³ | 2.35 €/kg | per order | `ABS Generic` | — |
+
+**Catalogue names are not house names, and our compounds are often not in the catalogue at all.**
+Resolve the grade against the platform's material catalogue before simulating
+(`GET /materials?q=…`), take the `material_id` and the version's recommended melt/mould
+temperatures, and record *which* catalogue entry you used.
+
+When the grade is not there — `PP TV20` is not, and the catalogue holds no talc-filled PP at all —
+the search falls back to the **stand-in named above**, then to the family's `… Generic` grade. It
+never silently takes "the first search hit": a search for `PP` returns a GF30 compound first, and
+simulating a glass-filled rheology for an unfilled part would be worse than not simulating. Any
+substitution is carried onto the quote's assumptions slide.
+
+## Cavity-pressure rule of thumb (used for clamp sizing)
+
+The AI filling solver models **one cavity with no runner system**, and is known to under-read
+absolute pressure on multi-gate parts. So clamp force is **never** sized from the simulation
+alone. We take the larger of (simulation-derived mean pressure) and (the family rule of thumb
+below), then apply the safety factor from [`lessons.md`](lessons.md).
+
+| Family | Mean cavity pressure for clamp |
+|---|---|
+| PP unfilled / talc | 350 bar |
+| PP GF / ABS | 450 bar |
+| PA6 GF30 | 650 bar |
+| POM | 550 bar |
+
+```json materials
+{
+  "default_grade": "PP TV20",
+  "procurement_surcharge_pct": 8.0,
+  "non_stock_lead_weeks": 3,
+  "non_stock_moq_kg": 500,
+  "drying_eur_per_kg": 0.06,
+  "grades": [
+    {"name": "PP Generic", "family": "PP",    "density_g_cm3": 0.91, "price_eur_kg": 1.55, "stock": "silo",      "catalogue_query": "PP Generic",       "catalogue_fallback": "PP Generic",  "clamp_pressure_bar": 350, "regrind_pct": 15, "dry": false},
+    {"name": "PP TV20",    "family": "PP",    "density_g_cm3": 1.05, "price_eur_kg": 1.85, "stock": "silo",      "catalogue_query": "PP T20",          "catalogue_fallback": "PP Generic",  "clamp_pressure_bar": 350, "regrind_pct": 15, "dry": false},
+    {"name": "PP GF30",    "family": "PP-GF", "density_g_cm3": 1.13, "price_eur_kg": 2.60, "stock": "big-bag",   "catalogue_query": "AKROLEN PP GF30", "catalogue_fallback": "PP Generic",  "clamp_pressure_bar": 450, "regrind_pct": 0,  "dry": false},
+    {"name": "PA6 GF30",   "family": "PA-GF", "density_g_cm3": 1.36, "price_eur_kg": 3.40, "stock": "big-bag",   "catalogue_query": "ALCOM PA6 900/8 GF10", "catalogue_fallback": "PA66 Generic", "clamp_pressure_bar": 650, "regrind_pct": 0, "dry": true},
+    {"name": "POM C",      "family": "POM",   "density_g_cm3": 1.41, "price_eur_kg": 3.10, "stock": "big-bag",   "catalogue_query": "POM Generic",     "catalogue_fallback": "POM Generic", "clamp_pressure_bar": 550, "regrind_pct": 10, "dry": true},
+    {"name": "ABS",        "family": "ABS",   "density_g_cm3": 1.05, "price_eur_kg": 2.35, "stock": "per order", "catalogue_query": "ABS Generic",     "catalogue_fallback": "ABS Generic", "clamp_pressure_bar": 450, "regrind_pct": 10, "dry": true}
+  ]
+}
+```
+
+`regrind_pct` is how much of the sprue/runner we are allowed to feed back for that family — it
+reduces the material line on a cold-runner tool. Zero for glass-filled grades (fibre breakage)
+and for anything the customer has spec'd as virgin.

simcon_toolkit/_kit/templates/quoting/shop/shop-profile.md

@@ -0,0 +1,87 @@
+# Shop profile — Feldmann Formenbau & Spritzguss GmbH
+
+> **`provenance: example` — this is a fictional shop.** Every press, rate and margin below is
+> invented. Until someone runs the interview (`python main.py setup`, script in
+> [`../INTERVIEW.md`](../INTERVIEW.md)) and flips `provenance` to their own shop's name, every
+> quote is watermarked as built on example data, because a quote against someone else's fictional
+> rates is worse than no quote at all.
+>
+> The agent quotes as *this* shop, so every sentence here changes what it offers and what it
+> refuses. Keep it as a worked reference for what "filled in" looks like; replace the values.
+
+## What we are
+
+A 62-person **moldmaker with an attached molding shop** in Lüdenscheid (NRW). Two thirds of
+revenue is tooling; one third is molded parts. We build the tool, sample it on our own presses,
+and — for most customers — also run the series.
+
+## What we do in-house
+
+| | In-house | Notes |
+|---|---|---|
+| Mold design (CAD) | ✅ | Siemens NX, 2 designers |
+| Cavity/core milling, EDM, wire | ✅ | 5-axis HSC, 3 sinkers |
+| Hardening, coating | ❌ | bought in (a hardening shop, +2 weeks) |
+| Texturing / graining | ❌ | bought in (a texturing house, +3 weeks) |
+| **Mold trials (sampling)** | ✅ | on our own presses — **always priced into the tool**, 2 shots included |
+| **Series production** | ✅ | 12 presses, 2-shift + optional weekend shift |
+| Assembly / pad printing | partial | manual assembly cells only |
+| Metrology | ✅ | Zeiss CMM, first-article inspection reports |
+
+**Consequence for quoting:** we quote **tool + piece price together** by default. A tool-only
+quote is possible but we then add a sampling package as a separate line — we never hand over an
+unsampled mold.
+
+## What we are good at, and what we decline
+
+- **Sweet spot:** technical parts, 50–400 g shot weight, 1–8 cavities, PP / PA / POM / ABS,
+  annual volumes 20k–500k.
+- **We decline:** Class-A cosmetic exteriors (no paint line, no texture bench of our own),
+  micro-molding under 1 g, LSR, and multi-component tools with more than 2 materials.
+- **We are cautious with:** unfilled PA (moisture handling), parts with wall thickness under
+  0.8 mm, and anything with a tolerance tighter than IT10 on a shrinking dimension — those get
+  a flagged assumption, not a silent price.
+
+## How we quote (the house rules the agent must follow)
+
+1. **Piece price is quoted at a volume ladder**, not at one number — 10k / 50k / 250k parts per
+   year — unless the customer states a volume. We do not ask for a volume just to produce a
+   number; we state the ladder and the assumption.
+2. **Tooling is quoted as a lump sum** with a lead time in weeks, amortisation shown separately
+   (we do not hide tool cost in the piece price unless the customer asks for it).
+3. **The cavity count is the customer's**, not ours to change. If the simulation says the count
+   is a bad idea (machine too large, pressure marginal, weld line on a functional face), we say
+   so as a flagged note and quote what was asked anyway.
+4. **Every assumption is written down.** A quote with an unstated assumption is how we lose
+   money — stating the assumptions in the quotation is not optional.
+5. **Validity 30 days**, material prices indexed (see [`commercial.md`](commercial.md)).
+
+## Standard commercial position
+
+- Currency **EUR**, prices ex works Lüdenscheid, Incoterms EXW.
+- Tooling payment 50 % order / 40 % first article / 10 % approval.
+- Series: 30 days net, minimum call-off 1 month of forecast.
+- Tool ownership passes to the customer on final payment; the tool stays in our shop.
+
+```json shop
+{
+  "provenance": "example",
+  "name": "Feldmann Formenbau & Spritzguss GmbH",
+  "location": "Lüdenscheid, DE",
+  "currency": "EUR",
+  "services": {
+    "mold_design": true,
+    "mold_making": true,
+    "mold_trials": true,
+    "series_production": true,
+    "texturing_in_house": false,
+    "hardening_in_house": false
+  },
+  "volume_ladder_parts_per_year": [10000, 50000, 250000],
+  "quote_validity_days": 30,
+  "shifts_per_day": 2,
+  "production_hours_per_year": 3800,
+  "declines": ["class-a cosmetic exterior", "micro molding < 1 g", "LSR", "3+ component tools"],
+  "caution": ["wall thickness < 0.8 mm", "unfilled PA", "tolerance tighter than IT10"]
+}
+```

simcon_toolkit/_kit/templates/quoting/shop/tooling.md

@@ -0,0 +1,145 @@
+# Tooling — how we price a mold
+
+> Fictional example cost model. Replace the numbers with your own estimating basis; keep the
+> structure, because [`quoting/costing.py`](../quoting/costing.py) reads the block at the bottom.
+
+## The estimating basis
+
+We estimate a mold as **base + cavities + the things that make it complicated**, not as a
+€/kg-of-steel figure. The five lines that decide the price:
+
+1. **Mold base** — driven by the plate footprint, which comes from the part's projected size, the
+   cavity layout, and the space the runner and cooling need.
+2. **Cavity + core machining** — per cavity, scaled by a complexity factor derived from the
+   geometry (surface area, wall thinness, feature count).
+3. **Side actions** — every undercut that cannot be released in the draw direction is a slide or
+   a lifter. This is the single most common reason a tool quote goes wrong: **count them from the
+   geometry, do not assume there are none.**
+4. **Feed system** — a hot runner costs real money per drop, and **the number of drops is the
+   number of gates the filling study decides**. This is why the study runs before the quote and
+   not after it: one extra gate per cavity on a 4-cavity tool is four more nozzles.
+5. **Cooling, ejection, finish, sampling** — smaller lines, but they add up to 20–30 % of the
+   tool.
+
+## Rules
+
+- **Standard construction is a 2-plate tool with a hot runner** for anything above 2 cavities;
+  below that we compare against a cold runner and quote the cheaper total (tool + material over
+  the volume ladder). A cold runner is not "cheaper" once you price the runner scrap over
+  250k parts.
+- **Cavity layout** is a balanced grid (1, 2, 4, 6, 8…). An unbalanced layout (3, 5) needs a
+  naturally-balanced runner and costs a slide-equivalent in design time — flag it.
+- **We include 2 sampling shots**; further loops are chargeable and quoted per shot.
+- **Hardening and texturing are bought in** — they add cost *and* weeks, and the weeks are
+  usually what the customer actually cares about.
+- **Steel choice** follows the volume ladder: 1.2311 pre-hardened up to 100k parts/year,
+  1.2343 ESU hardened above that, glass-filled grades always hardened.
+- **Tool lifetime guarantee:** 500k shots on hardened steel, 250k on pre-hardened, subject to
+  our maintenance contract.
+
+## Complexity factor — how we judge "complicated"
+
+The factor multiplies the cavity machining line. It starts at 1.0 and moves with what the
+geometry actually shows:
+
+| Signal | Effect |
+|---|---|
+| thin walls (nominal < 1.2 mm) | +0.15 — harder to fill, tighter tolerance on the steel |
+| deep draw (depth > 60 mm) | +0.15 |
+| **shape factor** above 1.0 | +0.25 per unit — part surface vs. its bounding-box surface, i.e. how much rib, boss and pocket is in there |
+| undercuts present | already priced as slides; +0.05 each for the fitting work |
+| textured or polished finish | priced separately, see `finish_eur_per_cm2` |
+
+> **Not** surface-to-volume ratio, which looks like the same thing and is not: it mostly measures
+> wall thinness, so a plain 1.5 mm plate scores 15 on it and would be priced as a hard cavity.
+> A flat part has a shape factor near or below 1; a ribbed housing goes well above it.
+
+Cap it at 1.8. Above that we are not estimating any more, we are guessing — say so and quote
+after a design review.
+
+```json tooling
+{
+  "layout": {
+    "cavity_spacing_mm": 45,
+    "edge_margin_mm": 60,
+    "min_plate_mm": [246, 246],
+    "max_plate_mm": [720, 720],
+    "preferred_grids": {"1": [1, 1], "2": [2, 1], "4": [2, 2], "6": [3, 2], "8": [4, 2], "12": [4, 3], "16": [4, 4]}
+  },
+  "base": {
+    "fixed_eur": 1200,
+    "eur_per_cm2_plate": 2.2,
+    "guided_ejector_eur": 850
+  },
+  "cavity": {
+    "fixed_eur_per_cavity": 2500,
+    "eur_per_cm2_part_surface": 15.0,
+    "complexity_cap": 1.8,
+    "hardened_surcharge_eur_per_cavity": 900,
+    "hardened_lead_weeks": 2
+  },
+  "side_actions": {
+    "slide_eur": 3200,
+    "lifter_eur": 1900,
+    "undercut_area_mm2_per_action": 400
+  },
+  "feed_system": {
+    "hot_runner_controller_eur": 2400,
+    "hot_runner_nozzle_eur": 1850,
+    "hot_runner_manifold_eur": 2800,
+    "cold_runner_eur": 900,
+    "cold_runner_volume_cm3_per_drop": 4.5,
+    "hot_runner_from_cavities": 3
+  },
+  "cooling": {
+    "circuit_eur": 420,
+    "cm2_plate_per_circuit": 450,
+    "min_circuits": 4,
+    "conformal_surcharge_eur": 0
+  },
+  "ejection": {
+    "fixed_eur": 1500,
+    "pin_eur": 55,
+    "pins_per_cavity_per_cm2_projected": 0.06,
+    "min_pins_per_cavity": 4
+  },
+  "finish": {
+    "technical_eur_per_cm2": 1.2,
+    "polished_eur_per_cm2": 4.5,
+    "textured_eur_per_cm2": 2.8,
+    "texturing_lead_weeks": 3,
+    "default": "technical"
+  },
+  "engineering": {
+    "design_rate_eur_h": 85,
+    "design_hours_base": 24,
+    "design_hours_per_cavity": 6,
+    "design_hours_per_side_action": 8,
+    "project_mgmt_pct": 6.0
+  },
+  "sampling": {
+    "shots_included": 2,
+    "eur_per_shot": 1600,
+    "extra_shot_eur": 1200
+  },
+  "lead_time": {
+    "base_weeks": 8,
+    "weeks_per_extra_cavity": 0.4,
+    "weeks_per_side_action": 0.8,
+    "hot_runner_weeks": 2,
+    "sampling_weeks": 2
+  },
+  "margin_pct": 18.0,
+  "risk_contingency_pct": 5.0
+}
+```
+
+## What is *not* in the model
+
+Deliberately, so the agent flags them instead of pricing them wrong:
+
+- **Multi-component / 2K tools** — out of scope (see `shop-profile.md`).
+- **Unscrewing cores** for threads — priced by design review, roughly 2× a slide.
+- **Conformal cooling** — we buy it in per project; `conformal_surcharge_eur` is 0 because we
+  quote it separately when the cooling study says the part needs it.
+- **Spare parts package** — quoted on request, typically 8 % of the tool.

simcon_toolkit/_kit/templates/run-one-simulation/.gitignore

@@ -0,0 +1,16 @@
+# What this template writes when it runs. A generated project receives THIS file and not the
+# kit's root one, so the rules have to live here or a customer's first commit carries their
+# own simulation results.
+#
+# build/ holds the .h5 the run produces and the inputs recorded beside it. Safe to delete.
+build/
+*.h5
+
+.venv/
+__pycache__/
+*.pyc
+.pytest_cache/
+
+# never commit credentials
+*token*.json
+.env

simcon_toolkit/_kit/templates/run-one-simulation/AGENTS.md

@@ -0,0 +1,41 @@
+# Run one simulation — how to drive this template
+
+The smallest end-to-end run: mesh a part, upload it, pick a material, run one AI filling
+simulation, read the peak pressure back. Start here when someone wants to see the platform
+work at all, or when a single answer is the whole task.
+
+## What to run
+
+```
+python main.py                          # the bundled sample part
+python main.py --part path/to/part.stl  # your own geometry
+```
+
+There is one script and no subcommands. A run takes seconds and writes the result under
+`build/`.
+
+## What is in this folder
+
+- `main.py` is the whole workflow, top to bottom, meant to be read.
+- `cadmould_geometry/` meshes the part; `cadmould_results/` opens the result file.
+- `sample/` holds the default geometry. It is ours — replace the file with your own and
+  `python main.py` keeps working, or delete the folder and always pass `--part`.
+- `build/` is where the downloaded result lands. It is yours and it is safe to delete.
+
+## Things to get right here
+
+**Everything happens inside one licence session**, `set_base_url` included. The script is
+written that way and it is the shape to copy — see the rule above on nesting.
+
+**The result opens in the web viewer.** Every simulation returns a viewer URL; this kit
+renders nothing locally, so hand the engineer the link rather than trying to draw the field.
+
+**Reading the result is the interesting part.** `main.py` pulls one number out. If you are
+asked for more, open the file with `cadmould_results` rather than reaching for h5py, and
+apply the result-reading rules above.
+
+## Going further
+
+This template uses the `cadmould.cloud` SDK, which does not wrap projects, groups or the
+decision log. If the task grows into a real study, that needs a REST client — the gate-study
+workflow is the worked example.