Remove em dashes where the replacement is unambiguous

The project has a standing style rule against em dashes that was missed
when these docs were written. This removes the 42 occurrences where the
substitution is mechanical and cannot change meaning:

- table cells using an em dash as a placeholder, now a hyphen
- em dashes immediately followed by a conjunction or subordinator
  (but, and, or, so, though, while, which, where, because, since),
  where a comma is simply correct

115 remain. Those join independent clauses, where a comma would create a
splice and a colon often reads wrong, so each needs a judgement call
rather than a pattern. A blanket regex over them was attempted and
reverted: it produced double colons and broken sentences.

docs/finance/phase1-construction-rfq.md

@@ -30,7 +30,7 @@ class, not a premium — it is what the presses and pallet racking need, and
 stepping up from a light-duty slab adds only on the order of 100–200 THB/m².
 The real cost variable is the ground underneath (see soil, below). A later
 value-engineering option is to zone the slab — 30 kN/m² under press hall and
-warehouse only, lighter under offices — but for rough pricing the whole
+warehouse only, lighter under offices, but for rough pricing the whole
 floor is specified at 30 so the number is conservative.
 
 ### 1a. Panel core: where PIR is not enough
@@ -109,7 +109,7 @@ Simple industrial enclosures beside the main building.
 | Boiler hall | 66 m² | 6.0 m |
 | Biomass fuel store | 42 m² | 5.0 m |
 | Chiller room | 60 m² | 5.0 m |
-| Cooling tower pad (concrete pad only) | 10 m² | — |
+| Cooling tower pad (concrete pad only) | 10 m² | - |
 
 The 14 m flue stack comes with the boiler — excluded here. The boiler hall
 envelope should be non-combustible (mineral-wool panels or masonry), not
@@ -165,7 +165,7 @@ transformer (PEA), and land works beyond normal grading.
   soft-to-medium Lamphun valley soil and to **state the foundation
   assumption they price** (footings vs driven piles; a piled slab can add
   several hundred to 1,000+ THB/m²). The soil, not the slab spec, is the
-  main price risk in the whole package — and it now applies to 4,387 m² of
+  main price risk in the whole package, and it now applies to 4,387 m² of
   pavement and a 1,711 m³ fill platform as well as the building.
 - **Tank proximity to the building**: the fire water tank sits about 4.5 m
   from the main building. Full, it is ~250 t on a 9 m ring foundation. On

docs/finance/revenue-model.md

@@ -2,7 +2,7 @@
 
 A first-pass economic model at mature capacity. Most numbers here are
 estimates from industry pricing norms and this repo's own cost research.
-**One set of real market prices now exists** — see the next section — and
+**One set of real market prices now exists** (see the next section), and
 it does not agree with the model. The point is the shape of the economics
 and which levers matter, not precision.
 
@@ -104,14 +104,14 @@ as output.
 All rows are on the same 300-day basis so they can be compared. CNX does
 not overtake them on installed capability until the Duplex expansion, and
 does not clearly lead until the mature fleet. The fleet is real and
-should not be discounted — but nothing here is a measurement of what they
+should not be discounted, but nothing here is a measurement of what they
 actually ship, and an unconditioned floor may not hold top-end
 effectiveness through the wet season.
 
 What remains true is the **quality and capability gap**, and it is where
 the whole differentiation now sits: no plating, no climate control, no
 metrology, no traceability. There is still no *industrial* vinyl plant in
-Thailand — but the gap CNX is closing is capability, not capacity, and it
+Thailand, but the gap CNX is closing is capability, not capacity, and it
 will be outgunned on presses for its first two phases.
 
 ### The incumbent is press-rich and metal-poor
@@ -138,7 +138,7 @@ these would shape a price list the same way:
 
 - **Printed jackets.** Their price carries one, and a jacket print run
   takes 10–20 working days from approved artwork. A 3,000-unit jacket
-  order is a bigger, slower print job — and this constraint **CNX would
+  order is a bigger, slower print job, and this constraint **CNX would
   share**, which would remove the competitive read entirely.
 - **Floor space.** Storing, finishing and shipping 3,000 records is an
   area problem before it is a metal problem.
@@ -178,7 +178,7 @@ full capex, it addresses the market's actual bottleneck, it earns before
 the press hall exists, and it is the one capability no one in the country
 has. The counter-argument is real — galvanics still needs a building, a
 clean room and zero-discharge water treatment, and it is the hardest skill
-to hire — but the option deserves costing rather than assuming.
+to hire, but the option deserves costing rather than assuming.
 
 **The number that decides it is ResurRec's actual annual output and title
 count**, which is not known. At 300 titles the stamper opportunity is
@@ -188,7 +188,7 @@ talks.
 **No HVAC turns the biggest cost line into the product.**
 [facility-security.md](../facility-security.md) specifies whole-factory
 climate control — 20–26 °C, 35–60% RH, PM filtration sized for
-burning-season PM2.5 — and the utilities budget treats it as overhead.
+burning-season PM2.5, and the utilities budget treats it as overhead.
 Against a competitor pressing in uncontrolled Bangkok air it is a quality
 argument instead:
 
@@ -219,7 +219,7 @@ walk. Two things to ask on the back of it, both free: ask **Newbilt** for
 their Thai
 install base and a service-response precedent — the northern-dispatch SLA
 question in [equipment-suppliers.md](../suppliers/equipment-suppliers.md)
-is much easier to answer with an in-country reference — and, if the
+is much easier to answer with an in-country reference, and, if the
 collaboration talks progress, ask **ResurRec** directly what commissioning
 and support actually cost them.
 
@@ -260,7 +260,7 @@ would be 430–500 USD per title:
 
 Every cell above divides capex by the **top** of the saving range. At the
 bottom the same table gives 2.5 and 5.0 years at 300 titles, 1.2 and 2.5
-at 600 — so the honest span is **one to five years**, and the adverse
+at 600, so the honest span is **one to five years**, and the adverse
 corner stretches past six once the ±30–50% band that
 [total-investment.md](total-investment.md) puts on the galvanics capex is
 applied. It is nonetheless driven by **title count, not record volume**,
@@ -272,7 +272,7 @@ needs the Stamper Discs quote behind it.
 evidence is internally inconsistent.** The model
 bundles mastering, cutting, stampers and test pressings into 19,000–57,000
 THB per title. At observed market rates a ten-track album masters at
-50,000–200,000 THB *before* the 22,000 THB stamper set — so the real
+50,000–200,000 THB *before* the 22,000 THB stamper set, so the real
 per-title value is more like **72,000–222,000 THB (2,180–6,730 USD)**.
 
 At 300–600 titles/yr that is **0.65M to 4.0M USD/yr of title services**,
@@ -297,8 +297,8 @@ for **300 units**. The published discounts:
 
 | Run | Discount | THB/unit | USD/unit |
 |---|---|---|---|
-| 150 (stated minimum) | — | rarely accepted, "cost is way too high" | |
-| **300 (reference)** | — | **650 – 950** | 19.70 – 28.80 |
+| 150 (stated minimum) | - | rarely accepted, "cost is way too high" | |
+| **300 (reference)** | - | **650 – 950** | 19.70 – 28.80 |
 | 500 | −5% | 618 – 903 | 18.70 – 27.35 |
 | 1,000 – 3,000 | −15% | 553 – 808 | 16.75 – 24.45 |
 | >3,000 | −30% | 455 – 665 | 13.80 – 20.15 |
@@ -306,7 +306,7 @@ for **300 units**. The published discounts:
 
 **The curve is remarkably flat.** A tenfold increase in run size buys only
 30% off. Whatever the exact split, that shape means a large per-title
-fixed cost that does not fall with volume — and it is the shape, not any
+fixed cost that does not fall with volume, and it is the shape, not any
 solved number, that is informative.
 
 Solving the first three tiers as a fixed-plus-variable structure gives
@@ -350,7 +350,7 @@ worth testing next, not the finding most worth relying on.
 
 **The fixed component looks small against the stated inputs.** At
 34,000–50,000 THB it barely covers a ≤22,000 THB stamper set plus one to
-three songs of mastering — though at the other end of the solve range
+three songs of mastering, though at the other end of the solve range
 (~41,800 THB low ladder) the tension eases, so this may be an artefact of
 the endpoint chosen rather than a real gap. Either the 650–950 THB does
 not fully carry album mastering, or albums are mastered far below the
@@ -362,7 +362,7 @@ problem.** At nameplate the two presses clear 3,000 units in about two
 days, so a two-to-four-week extension cannot be press time. The likely
 cause is in the numbers already in this repo: a stamper is good for
 **1,000–2,000 pressings**, so a 3,000-unit run needs **two to three
-stamper sets** — and with no plating in-house, every additional set is an
+stamper sets**, and with no plating in-house, every additional set is an
 overseas order. "Blocking out the entire factory" reads like scheduling
 around an imported-metalwork wait.
 
@@ -504,7 +504,7 @@ security, overtime, bonus and PPE:
 
 That is 3.7M–5.4M THB/yr (113k–163k USD), or **0.08–0.16 USD per record**
 over 1.0–1.5M records. Mastering and cutting engineers are excluded —
-they are title-services labour, not per-record — and admin, sales and
+they are title-services labour, not per-record, and admin, sales and
 logistics sit in overheads below.
 
 ## Margin per record
@@ -603,7 +603,7 @@ profits where there are any; see [boi.md](../boi.md).
   data point above is buy-side, from one plant and one buyer, and it
   contradicts the model on per-title services and probably on run size.
   The Making Vinyl trip and early conversations should test the 3.50 to
-  4.60 USD blended assumption — and the run-size question — before either
+  4.60 USD blended assumption, and the run-size question — before either
   anchors a financing case
 - Demand risk is real: vinyl demand is cyclical and capacity has been
   added globally; the model assumes the plant can sell what it presses