ballistics_engine/

card_service.rs

//! Transport-free card-generation services (come-ups, range table, wind card).
//!
//! These replicate the CLI's card commands exactly — same zero solve, same sampled
//! trajectory, same nearest-sample row selection, same adjustment/bias/CF/click
//! ordering via [`crate::adjustment`] — as versioned request/response types the
//! bridge can expose to embedded consumers. The CLI remains the reference
//! implementation; `tests/card_bridge_golden.rs` asserts row-for-row agreement.
//!
//! ## Unit convention
//!
//! Unlike solve-json (explicit SI), card requests are denominated in the shooter's own
//! units, exactly like the CLI flags they mirror: imperial = fps, grains, inches, yards,
//! °F, inHg, mph; metric = m/s, grams, mm, meters, °C, hPa, m/s. A DOPE card is a display
//! artifact; its inputs and outputs share the shooter's unit world, and this keeps the
//! request shape identical to the documented CLI surface.
//!
//! ### Per-dimension units (MBA-1519)
//!
//! Nobody shoots in one system. A Finnish shooter ranges in metres, dials MIL, and buys
//! bullets labelled in grains, because that is how the industry sells them; European makers
//! dual-print `grams / grains` rather than replacing one with the other, and mix within a
//! single table. So the denominating state on this request is a unit PER DIMENSION, and
//! [`CardUnits`] is a one-shot preset that fills in every dimension the request does not
//! state for itself — not a mode the card is "in".
//!
//! Concretely: an explicit per-dimension field wins; an absent one falls back to what
//! `units` implies. A request written before those fields existed states none of them, so
//! every dimension falls back and the card is byte-for-byte the card it always was
//! (`tests/card_units_per_dimension.rs` pins that against a stored document).
//!
//! The response needed no new FIELD for any of this: [`CardUnitsBlockV1`] has always carried
//! seven independent label fields, so a card whose distance is metres and whose velocity is
//! fps already had somewhere to say so. It did need the two angular labels to become
//! requestable: `elevation_adjustment`/`windage_adjustment` print `"MIL"`/`"MOA"`/`"SMOA"`/
//! `"IPHY"`/`"CLICKS"` and [`AdjustmentUnit`] deserialized lowercase only, so those two were
//! the only fields of the block a caller could not echo back. They take the uppercase
//! spellings as aliases now, which makes "echo the units block into the next request" true
//! of all seven fields rather than five of them.
//!
//! Three candidates do NOT get a new field, and each for its own reason:
//!
//! * **Scope adjustment** is already per-dimension, and was first: `adjustment_unit` and
//!   `windage_unit` are separate fields with separate click graduations, because a shooter
//!   with a MIL rifle and an MOA rifle has two scopes rather than a preference. SMOA (and
//!   its `iphy` spelling) is already one of the values they take.
//! * **Clicks** is not a units dimension. A click is meaningless without a graduation, and
//!   this module already enforces that: `adjustment_unit == Clicks` without an
//!   `elevation_click_value` is a refusal. The graduation belongs beside the turret's
//!   MIL/MOA/SMOA marking, not in a units picker.
//! * **Twist rate** is a dimension of the design but not of this request — a card request
//!   carries no twist, so there is nothing here for a twist unit to denominate.
//!
//! And the linear-at-distance spellings are NOT units: "inches@100yd" IS SMOA — the same
//! 1/3600 rad, the same printed number, no conversion — and "cm/100m" is 0.1 mrad, i.e. MIL
//! with a factor-of-ten display. They are alternate HEADER TEXT for units this request
//! already has, and adding them as values would give a caller two spellings that produce
//! identical numbers.
//!
//! ### What the response does NOT tell a caller
//!
//! Like every other default on this request — `sight_height`, `temperature`, `pressure`,
//! `humidity`, `windage_unit` — a per-dimension unit taken from the preset is applied
//! silently, and this surface has no notice channel at all (there is no `assumptions` array
//! on [`CardResponseV1`], unlike solve-json's `SolveSuccessV1`).
//!
//! Be precise about what that costs, because a paragraph of prose can argue it away. The `units`
//! block names the unit of the SEVEN OUTPUT COLUMNS and nothing else. It says nothing about
//! the six dimensions that are input-only — `mass`, `diameter`, `sight_height`,
//! `temperature`, `pressure`, `altitude` — so for those a caller cannot read back off the
//! card which unit the engine settled on.
//!
//! `altitude` is the sharp one, because it is the dimension whose fallback is deliberately
//! NOT the preset's: an imperial `card.range_table` carrying `altitude: 5000` and no
//! `altitude_unit` — which a US caller writing an imperial card means as feet — is solved at
//! 5000 METRES, and no field of the response says so. On a .308 175 gr that is ~1.9 MIL of
//! elevation at 1000 yd against the same card with `altitude_unit: "feet"`. The BEHAVIOUR
//! predates the per-dimension fields (`altitude` has always reached
//! [`AtmosphericConditions`] unconverted); what is new is that the request can now name the
//! dimension, so the gap is describable instead of merely present.
//!
//! No notice was added here, and the reason is narrow rather than principled: any notice —
//! an `assumptions` array, or extra label fields on the units block — is a change to the
//! RESPONSE bytes of every card, including the stored documents the reprint path compares
//! label-for-label and the six goldens that are the whole evidence for this change being
//! additive on the wire. A response-schema change belongs to a ticket that can move those
//! goldens honestly. Until then the remedy is on the request: a caller that cares which
//! altitude it got states `altitude_unit`, and a card built by an app should state every
//! dimension rather than lean on a preset.

use serde::{Deserialize, Serialize};

use crate::adjustment::{
    adjustment_display, adjustment_unit_label, parse_click_value, windage_adjustment_display,
    AdjustmentUnit, ClickValue,
};
use crate::hold_curve::{
    card_sample_max_range_m, range_not_sampled_message, run_sampled_flight, sample_at_range,
    CardTruncation, SampledFlight, CARD_SAMPLE_INTERVAL_M,
};
use crate::{AtmosphericConditions, BallisticInputs, BCSegmentData, DragModel, WindConditions};

/// Schema version of the card request/response contract.
pub const CARD_SCHEMA_VERSION_V1: u32 = 1;

const GRAINS_TO_KG: f64 = crate::constants::GRAINS_TO_KG;

/// The unit PRESET for a card request: a one-shot bulk action that fills in every
/// per-dimension unit the request does not state for itself, not a mode the card is in.
///
/// It stays because it is the whole of the pre-MBA-1519 contract and because "give me an
/// imperial card" remains the overwhelmingly common request. What changed is its standing:
/// it is now the FALLBACK for each dimension rather than the denominator of all of them.
#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardUnits {
    #[default]
    Imperial,
    Metric,
}

// --- Per-dimension units (MBA-1519) ------------------------------------------------------
//
// One enum per dimension the card request actually denominates. Each is a plain value list;
// the mapping from a value to a conversion factor and to a printed label lives in exactly
// one place, `Units` below, so a dimension cannot convert one way and be labelled another.
//
// Every enum carries serde aliases for the spelling the RESPONSE prints for that unit
// (`"m/s"`, `"ft-lb"`, `"in"`, `"yd"`, ...), and so does [`AdjustmentUnit`] over in
// `adjustment.rs` for the two uppercase dial labels. A caller that reads a card's `units`
// block and writes those strings straight back into the next request is doing the obvious
// thing, and it now works — for all seven fields of the block — instead of failing on a
// request that denies unknown values.

/// Range axis: what `zero_distance`, `start`/`end`/`step` and every row's `range` are in.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardDistanceUnit {
    #[serde(alias = "yd", alias = "yard")]
    Yards,
    #[serde(alias = "m", alias = "metres", alias = "meter", alias = "metre")]
    Meters,
}

/// Speed of the bullet: `muzzle_velocity`, `bc_segments` band edges, the `velocity` column.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardVelocityUnit {
    #[serde(alias = "ft/s")]
    Fps,
    #[serde(alias = "m/s")]
    Mps,
}

/// Bullet weight. Grains and grams differ by 15.43x, and on the BC path mass never enters
/// the drag denominator — it touches energy and stability only — so a grams-for-grains slip
/// yields a trajectory that looks entirely normal. Which is exactly why the unit has to be
/// stated per-dimension rather than inferred: the dual-printed `g / gr` box a European
/// shooter is holding is the normal case, not an edge one.
///
/// The one-character wire tokens `"g"` and `"gr"` are deliberately NOT accepted, though
/// `"gram"`/`"grams"` and `"grain"`/`"grains"` are. Those two tokens are the slip this
/// dimension exists to prevent: a 175-grain load sent as `"g"` solves to bit-identical
/// dials, drift, velocity and time — only the energy column moves, 1009 ft-lb to 15575 — and
/// the response has no mass label for a reader to catch it on. Nor can a plausibility guard
/// catch it: bullets run roughly 5–800 grains and roughly 0.3–52 grams, so the two ranges
/// OVERLAP across 5–52, where 48 grains (a .22 varmint bullet) and 48 grams (a .50 BMG) are
/// both ordinary. There is no threshold to put a guard at, so the defence has to be the
/// spelling, and the spelled-out name is the one thing a caller cannot typo into the other
/// unit.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardMassUnit {
    #[serde(alias = "grain")]
    Grains,
    #[serde(alias = "gram")]
    Grams,
}

/// A small linear length. Shared by the request's two TARGET-SIDE linear dimensions
/// (sight-to-bore geometry and the linear drop column) because they take the same values,
/// never because they are the same setting: they are separate fields and resolve
/// independently.
///
/// Bullet diameter is NOT one of them — it has [`CardDiameterUnit`], a narrower vocabulary.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardLinearUnit {
    #[serde(alias = "in", alias = "inch")]
    Inches,
    #[serde(alias = "centimeters", alias = "centimetres")]
    Cm,
    #[serde(alias = "millimeters", alias = "millimetres")]
    Mm,
}

/// Bullet diameter: inches or millimetres, the two units a calibre is ever quoted in.
///
/// Its own enum rather than a third user of [`CardLinearUnit`], which would also accept
/// `cm`. The design's dimension table lists inches and mm for this dimension and no third
/// value, and a wire vocabulary on a `deny_unknown_fields` paid contract is far cheaper to
/// widen later than to narrow: nobody quotes a calibre in centimetres, so `0.782` cm is
/// likelier to be a slipped `7.82` mm than a request anyone meant.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardDiameterUnit {
    #[serde(alias = "in", alias = "inch")]
    Inches,
    #[serde(alias = "millimeters", alias = "millimetres")]
    Mm,
}

/// Wind and crossing-target speed. Four values, not two: a European card reads km/h and a
/// coastal one reads knots, and neither is expressible by picking a system.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardWindSpeedUnit {
    Mph,
    #[serde(alias = "m/s")]
    Mps,
    #[serde(alias = "km/h", alias = "kmh")]
    Kph,
    #[serde(alias = "kn", alias = "kt", alias = "knot")]
    Knots,
}

/// Air temperature.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardTemperatureUnit {
    #[serde(alias = "f", alias = "degf")]
    Fahrenheit,
    #[serde(alias = "c", alias = "degc")]
    Celsius,
}

/// Station pressure. `hpa` and `mbar` are the same number, which is why mbar is an alias
/// rather than a value of its own.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardPressureUnit {
    #[serde(alias = "in-hg", alias = "inchesofmercury")]
    InHg,
    #[serde(alias = "mbar", alias = "millibar")]
    HPa,
}

/// Terminal energy column.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardEnergyUnit {
    #[serde(alias = "ft-lb", alias = "ftlbs", alias = "footpounds")]
    FtLb,
    // "J" as well as "j": the response's energy label is the capitalized SI symbol, and the
    // aliases exist precisely so a caller can echo a card's own labels back at it.
    #[serde(alias = "j", alias = "J", alias = "joule")]
    Joules,
}

/// Site elevation. See [`CardRequestV1::altitude_unit`] for why this one dimension's
/// default does not come from the preset.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum CardAltitudeUnit {
    #[serde(alias = "ft", alias = "foot")]
    Feet,
    #[serde(alias = "m", alias = "metres", alias = "meter", alias = "metre")]
    Meters,
}

/// Everything the three card surfaces share: load, zero, atmosphere, display axes.
///
/// Field values are in this request's PER-DIMENSION units: each dimensional field names the
/// `*_unit` field that denominates it, and each of those falls back to what `units` implies
/// when it is absent (see the module docs). Stating none of them is the pre-MBA-1519 shape
/// and reproduces the pre-MBA-1519 card exactly.
///
/// ## Adding a field here is additive on the WIRE and breaking in RUST
///
/// This is a deserialization target: the supported way to build one is
/// `serde_json::from_str`/`from_value` on a JSON document, which is what the bridge and every
/// binding do, and an optional new field costs such a caller nothing. A downstream Rust crate
/// that constructs it with a struct literal has no such luck — every added field is an
/// `E0063: missing fields` against code that compiled before, which is exactly what MBA-1519's
/// eleven fields do to a crate written against 0.39.x. The remedy is one line per field
/// (`altitude_unit: None`, and so on for the rest); the intended path is to deserialize.
///
/// `#[non_exhaustive]` was considered and NOT applied. It would break the same callers in the
/// same release, and — with no `Default` and no builder on this type — it would leave a
/// downstream crate no way to construct the request at all, trading one fixable compile error
/// for a permanent one. So the breakage is stated in the CHANGELOG for the release that causes
/// it rather than encoded in the type, and a future field will be stated the same way.
#[derive(Debug, Clone, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct CardRequestV1 {
    /// The preset that fills in every `*_unit` field below that this request leaves absent.
    #[serde(default)]
    pub units: CardUnits,

    // Load
    /// Muzzle speed, in `velocity_unit`.
    pub muzzle_velocity: f64,
    pub ballistic_coefficient: f64,
    /// Bullet weight, in `mass_unit`.
    pub mass: f64,
    /// Bullet diameter, in `diameter_unit`.
    pub diameter: f64,
    #[serde(default)]
    pub drag_model: DragModelV1,
    /// Sight height over bore, in `sight_height_unit`; CLI default 1.5 in.
    #[serde(default = "sight_height_unset")]
    pub sight_height: f64,

    // Zero
    /// Zero range, in `distance_unit`.
    pub zero_distance: f64,
    /// Deliberate POI offset at the zero range, in `drop_unit` — it is a displacement on
    /// the target, which is the dimension the drop column reports; 0 = zeroed dead-on.
    #[serde(default)]
    pub zero_poi_vertical: f64,
    /// Deliberate lateral POI offset at the zero range, in `drop_unit`.
    #[serde(default)]
    pub zero_poi_horizontal: f64,
    /// Lateral sight-to-bore mount offset, in `sight_height_unit` — it is rifle geometry
    /// measured beside the sight height, not a figure read off a target.
    #[serde(default)]
    pub sight_offset_lateral: f64,

    // Atmosphere (CLI defaults: 59 °F / 15 °C, 29.92 inHg / 1013.25 hPa, 50 %, 0 alt)
    /// Station temperature, in `temperature_unit`.
    #[serde(default)]
    pub temperature: Option<f64>,
    /// Station pressure, in `pressure_unit`.
    #[serde(default)]
    pub pressure: Option<f64>,
    /// Relative humidity, percent. Dimensionless, so no unit field governs it.
    #[serde(default = "default_humidity")]
    pub humidity: f64,
    /// Site elevation, in `altitude_unit` — which defaults to METRES whatever `units` says.
    #[serde(default)]
    pub altitude: f64,

    // Wind for the flight (in `wind_speed_unit`; wind-FROM degrees). The wind card ignores
    // these and sweeps its own speed list.
    #[serde(default)]
    pub wind_speed: f64,
    #[serde(default)]
    pub wind_direction_deg: f64,

    // Card domain, in `distance_unit`
    pub start: f64,
    pub end: f64,
    pub step: f64,

    // --- Per-dimension units (MBA-1519) --------------------------------------------------
    //
    // Every one of these is `Option`al and `#[serde(default)]`: `CardRequestV1` denies
    // unknown fields and backs paid surfaces (the Pro DOPE card, the saved-card reprint, the
    // PDF export), so a stored request written before they existed has to keep parsing and
    // keep producing the identical document. `None` means "take it from `units`", which is
    // what every such request says about every dimension.
    //
    // What is deliberately NOT here: a scope-adjustment unit (`adjustment_unit` /
    // `windage_unit` below already are that dimension, SMOA included), a clicks unit (a
    // click needs a graduation, which is `elevation_click_value`), a twist-rate unit (this
    // request carries no twist), and the linear-at-distance spellings (header text for SMOA
    // and MIL, not units). The module docs give the full reasoning.
    /// Unit of `zero_distance`, `start`, `end`, `step` and every row's `range`.
    /// Default: yards (imperial) / metres (metric).
    ///
    /// Distance is the dimension where reinterpreting the number does the most damage the
    /// most quietly: "300" yards and "300" metres are targets 9.36% apart, and a stored zero,
    /// a DOPE row and a target-size hold are all attached to a PHYSICAL distance rather than
    /// to a unit-free label. So a card's distance unit is stated, never inferred from a
    /// neighbouring dimension.
    #[serde(default)]
    pub distance_unit: Option<CardDistanceUnit>,
    /// Unit of `muzzle_velocity`, the `bc_segments` band edges and the `velocity` column.
    /// Default: fps (imperial) / m/s (metric).
    #[serde(default)]
    pub velocity_unit: Option<CardVelocityUnit>,
    /// Unit of `mass`. Default: grains (imperial) / grams (metric).
    #[serde(default)]
    pub mass_unit: Option<CardMassUnit>,
    /// Unit of `diameter`. Default: inches (imperial) / mm (metric). Those two values and no
    /// others — see [`CardDiameterUnit`].
    #[serde(default)]
    pub diameter_unit: Option<CardDiameterUnit>,
    /// Unit of `sight_height` and `sight_offset_lateral` — the sight-to-bore geometry.
    /// Default: inches (imperial) / mm (metric).
    ///
    /// This is a dimension the design's table does not contain, added deliberately rather
    /// than by oversight: without it, the only home for the sight height would be
    /// `drop_unit`, and then asking for a drop column in centimetres would silently reread a
    /// 1.5-inch sight height as 1.5 cm. It is a twelfth field on a `deny_unknown_fields`
    /// contract that cannot be withdrawn without a breaking change, which is the cost being
    /// accepted here.
    ///
    /// Separate from `drop_unit` on purpose. They are the same unit today only because both
    /// presets happen to give them the same value; they are different measurements (one is
    /// taken on the rifle with calipers, the other is read off a target downrange), and
    /// folding them together would mean a caller who asks for a drop column in centimetres
    /// silently reinterprets a 1.5-inch sight height as 1.5 cm.
    #[serde(default)]
    pub sight_height_unit: Option<CardLinearUnit>,
    /// Unit of the `drop_linear` / `wind_linear` columns and of `zero_poi_vertical` /
    /// `zero_poi_horizontal`. Default: inches (imperial) / mm (metric).
    ///
    /// The ANGULAR half of the design's drop dimension is `adjustment_unit` /
    /// `windage_unit`, which have always been separate fields; this is the linear half, and
    /// `cm` is available here because a metric card prints 250 cm of drop, not 2500 mm.
    #[serde(default)]
    pub drop_unit: Option<CardLinearUnit>,
    /// Unit of `wind_speed`, of every entry in `wind_speeds`, and of `pdf.target_speed`.
    /// Default: mph (imperial) / m/s (metric).
    #[serde(default)]
    pub wind_speed_unit: Option<CardWindSpeedUnit>,
    /// Unit of `temperature` and of the temperature default. Default: °F (imperial) / °C
    /// (metric).
    #[serde(default)]
    pub temperature_unit: Option<CardTemperatureUnit>,
    /// Unit of `pressure` and of the pressure default. Default: inHg (imperial) / hPa
    /// (metric).
    #[serde(default)]
    pub pressure_unit: Option<CardPressureUnit>,
    /// Unit of the `energy` column. Default: ft-lb (imperial) / J (metric).
    #[serde(default)]
    pub energy_unit: Option<CardEnergyUnit>,
    /// Unit of `altitude`. Default: **metres, whatever `units` says** — this is the one
    /// dimension whose fallback is not the preset's.
    ///
    /// `altitude` has ignored the `units` scalar since this module was written: it is passed
    /// to [`AtmosphericConditions`] unconverted, and that field is documented metres, so an
    /// imperial card's altitude has always been metres too (the PDF header divides it by
    /// 0.3048 regardless of `units` for exactly that reason). Letting the imperial preset
    /// start filling this with `feet` would therefore not be additive: every stored imperial
    /// card carrying `altitude: 1000` would silently become a card shot at 304.8 m instead
    /// of 1000 m, which is a different density altitude and a different trajectory.
    ///
    /// So the preset does not reach this field. A caller that wants feet says `"feet"`, and
    /// gets the first way this request has ever had to express one.
    #[serde(default)]
    pub altitude_unit: Option<CardAltitudeUnit>,

    // Display axes
    #[serde(default)]
    pub adjustment_unit: AdjustmentUnit,
    /// Windage axis unit; defaults to the elevation axis.
    #[serde(default)]
    pub windage_unit: Option<AdjustmentUnit>,
    /// Turret graduations as suffixed strings ("0.1mil", "0.25moa"); required when
    /// the corresponding axis unit is `clicks`. Windage falls back to elevation.
    #[serde(default)]
    pub elevation_click_value: Option<String>,
    #[serde(default)]
    pub windage_click_value: Option<String>,
    /// Scope tracking correction factors (MBA-1358); 1.0 = tracks true.
    #[serde(default = "default_cf")]
    pub elevation_cf: f64,
    #[serde(default = "default_cf")]
    pub windage_cf: f64,
    /// Selected zero set's dial corrections, true angular MIL (MBA-1360).
    #[serde(default)]
    pub zero_set_elevation_bias_mil: f64,
    #[serde(default)]
    pub zero_set_windage_bias_mil: f64,

    /// Wind card only: the sweep of wind speeds, each in `wind_speed_unit` (so km/h and
    /// knots are available here too), one output column each.
    #[serde(default)]
    pub wind_speeds: Vec<f64>,
    /// Wind card only: wind-FROM angles in degrees; default `[90]` (full-value from
    /// the right), matching the CLI.
    #[serde(default)]
    pub wind_angles_deg: Vec<f64>,

    /// Explicit velocity-banded BC schedule, mirroring the CLI's repeatable
    /// `--bc-segment VMIN:VMAX:BC`. Velocities are in the request's `velocity_unit`,
    /// exactly like the CLI flag. When supplied it
    /// wins over `bc5d_table_path`, and — CLI parity — the scalar
    /// `ballistic_coefficient` remains the interior-gap fallback unchanged. The
    /// schedule feeds BOTH the zero solve and every sampled trajectory.
    #[serde(default)]
    pub bc_segments: Option<Vec<CardBcSegmentV1>>,
    /// Filesystem path to a caliber-specific BC5D correction table
    /// (`bc5d_<caliber>.bin`, the exact format the CLI's `--bc-table-dir`
    /// consumes). The table is CRC-verified, a velocity-keyed BC schedule is
    /// generated for this load (same ladder as the CLI), and the scalar BC gets the
    /// table's muzzle correction as the interior-gap fallback. Ignored when
    /// `bc_segments` is supplied. Not available on wasm32 builds (no filesystem);
    /// there it is rejected with an invalid-request error.
    #[serde(default)]
    pub bc5d_table_path: Option<String>,

    /// Presentation-only options for the printable PDF dope card (`card.pdf`). Nothing in
    /// here touches the ballistics: the numbers in the PDF are the rows
    /// [`range_table_v1`] returns, computed by the same call.
    ///
    /// It lives on the shared request — rather than in a `card.pdf`-only wrapper type — so
    /// an app stores ONE `CardRequestV1` per saved card and replays that exact document
    /// against either surface: `card.range_table` for the screen, `card.pdf` for the
    /// printout. The on-screen commands ignore this block.
    ///
    /// Deliberately NOT `#[cfg(feature = "pdf")]`, even though everything that consumes it
    /// is: `CardRequestV1` denies unknown fields, so gating the field would make a
    /// pdf-less build REJECT a stored request that carries it, breaking exactly the
    /// round-trip the field exists to support. A pdf-less build parses it and never reads
    /// it; only the `card.pdf` command itself disappears.
    #[serde(default)]
    pub pdf: Option<PdfCardOptionsV1>,
}

/// Presentation knobs for the PDF dope card: the header/footer labels the printed card
/// carries and the table's font size. The ballistics axes it prints (elevation unit,
/// windage unit, click graduations, tracking CFs, zero-set biases) are NOT here — they are
/// already fields on [`CardRequestV1`], shared with the on-screen card, precisely so the
/// two cannot disagree about what a row means.
///
/// Every field is optional and defaults to the CLI dope card's own default.
#[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct PdfCardOptionsV1 {
    /// Leading text of header line 1 — the CLI's rifle/profile name. Default
    /// `"Dope Card"`. Long headers are truncated to fit the page.
    #[serde(default)]
    pub title: Option<String>,
    /// Header `Loc:` label. Default empty.
    #[serde(default)]
    pub location: Option<String>,
    /// Footer `Powder:` label. Default empty.
    #[serde(default)]
    pub powder: Option<String>,
    /// Footer `Bullet:` label. Default empty.
    #[serde(default)]
    pub bullet: Option<String>,
    /// Crossing-target speed for the Lead column, in the request's `wind_speed_unit` — the
    /// same convention as the CLI's `--target-speed`
    /// (MBA-1325). The lead is `speed * time-of-flight` for a full-value 90° crossing,
    /// held on the windage axis, exactly as `trajectory -o pdf` computes it.
    ///
    /// Absent means "this card carries no lead data": the Lead column renders as em-dashes
    /// rather than a column of confident-looking zeroes. An explicit `0.0` is a different
    /// statement — a stationary target, whose lead genuinely is zero — and prints zeroes,
    /// matching the CLI's `--target-speed 0` default.
    #[serde(default)]
    pub target_speed: Option<f64>,
    /// Data-table font scale. Must be finite and within
    /// [`crate::pdf_dope_card::FONT_SCALE_RANGE`] (0.5..=3.0); out-of-band values are
    /// rejected rather than silently clamped, so what a stored request asks for is what it
    /// gets. Mutually exclusive with `font_preset`. Default 1.0.
    #[serde(default)]
    pub font_scale: Option<f64>,
    /// Named font scale: `"small"` (0.8), `"medium"` (1.0), `"large"` (1.4) — the CLI's
    /// `--font-preset` vocabulary, single-letter aliases included. An unrecognized preset
    /// is an error here (the CLI warns and falls back to medium; a stored request should
    /// not silently render at a size it did not ask for). Mutually exclusive with
    /// `font_scale`.
    #[serde(default)]
    pub font_preset: Option<String>,
    /// Render the data rows in bold. Default false.
    #[serde(default)]
    pub bold_data: bool,
}

/// One velocity band of an explicit velocity-keyed BC schedule
/// (`CardRequestV1::bc_segments`). `velocity_min`/`velocity_max` are in the
/// request's `velocity_unit` and must satisfy `velocity_min < velocity_max`;
/// `bc` is the BC (for the request's `drag_model`) that applies inside the band.
#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct CardBcSegmentV1 {
    pub velocity_min: f64,
    pub velocity_max: f64,
    pub bc: f64,
}

/// The `sight_height` sentinel for "not supplied".
///
/// A serde field default is a nullary function: it cannot see the request, so it cannot know
/// which `sight_height_unit` the figure would have to be in. NaN stands in until `resolve()`
/// swaps it for `Units::default_sight_height`, which does know.
fn sight_height_unset() -> f64 {
    f64::NAN
}
fn default_humidity() -> f64 {
    50.0
}
fn default_cf() -> f64 {
    1.0
}

/// Drag model selector mirroring the CLI's `--drag-model` strings.
#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum DragModelV1 {
    G1,
    #[default]
    G7,
    G2,
    G5,
    G6,
    G8,
    GI,
    GS,
    RA4,
}

impl From<DragModelV1> for DragModel {
    fn from(v: DragModelV1) -> Self {
        match v {
            DragModelV1::G1 => DragModel::G1,
            DragModelV1::G7 => DragModel::G7,
            DragModelV1::G2 => DragModel::G2,
            DragModelV1::G5 => DragModel::G5,
            DragModelV1::G6 => DragModel::G6,
            DragModelV1::G8 => DragModel::G8,
            DragModelV1::GI => DragModel::GI,
            DragModelV1::GS => DragModel::GS,
            DragModelV1::RA4 => DragModel::RA4,
        }
    }
}

#[derive(Debug, Clone, Serialize)]
pub struct CardRowV1 {
    /// Range in the request's `distance_unit`; `CardUnitsBlockV1::distance` names it.
    pub range: f64,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub drop_linear: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub drop_adj: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub come_up: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub wind_linear: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub wind_adj: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub velocity: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub energy: Option<f64>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub time: Option<f64>,
    #[serde(skip_serializing_if = "Vec::is_empty", default)]
    pub wind_columns: Vec<f64>,
}

#[derive(Debug, Clone, Serialize)]
pub struct CardUnitsBlockV1 {
    pub distance: &'static str,
    pub velocity: &'static str,
    pub energy: &'static str,
    pub drop: &'static str,
    pub wind_speed: &'static str,
    pub elevation_adjustment: String,
    pub windage_adjustment: String,
}

#[derive(Debug, Clone, Serialize)]
pub struct CardResponseV1 {
    pub schema_version: u32,
    pub kind: &'static str,
    pub zero_distance: f64,
    /// The scalar BC these rows were actually computed with: the request's published
    /// `ballistic_coefficient` unless a `bc5d_table_path` applied its muzzle correction, in
    /// which case it is the corrected value (the same scalar the solve and every sampled
    /// flight ran with).
    ///
    /// A saved card must be able to state the BC its numbers came from — the printed
    /// footer's `BC:` — long after the correction table it used has been replaced on disk.
    /// Without this on the response there is nothing for an app to store, and a reprint
    /// could only quote the BC the request nominated.
    pub bc_for_solve: f64,
    pub units: CardUnitsBlockV1,
    /// Wind card only: the swept speeds, one per `wind_columns` entry.
    #[serde(skip_serializing_if = "Vec::is_empty", default)]
    pub wind_speeds: Vec<f64>,
    /// Wind card only: one block of rows per wind angle.
    #[serde(skip_serializing_if = "Vec::is_empty", default)]
    pub wind_angles_deg: Vec<f64>,
    pub rows: Vec<CardRowV1>,
    /// Wind card with multiple angles: rows for angles beyond the first.
    #[serde(skip_serializing_if = "Vec::is_empty", default)]
    pub extra_angle_rows: Vec<Vec<CardRowV1>>,
    /// Present ONLY when the solved flight stopped short of the rows requested, in which case
    /// `rows` holds every row it did reach and nothing past them (MBA-1476 follow-up).
    ///
    /// Additive and absent on every card that ran to its requested `end`, so a caller reading
    /// an untruncated card sees byte-identical JSON to before. It is a field rather than only
    /// a log line because the apps and the Flask layer consume this shape: a card that quietly
    /// returns fewer rows than were asked for, with nothing on the response saying so, is the
    /// same class of silent failure as the substituted row this work removed.
    #[serde(skip_serializing_if = "Option::is_none", default)]
    pub truncation: Option<CardTruncationV1>,
}

/// The wire form of [`CardTruncation`]: the rows a card could not print and why.
/// Distances are in the response's own `units.distance`.
#[derive(Debug, Clone, Copy, Serialize)]
pub struct CardTruncationV1 {
    /// The furthest row the request asked for.
    pub requested_end: f64,
    /// The last row actually present in `rows`.
    pub last_row: f64,
    /// How far the solved flight got — a property of the load, not of the request.
    pub reach: f64,
}

impl CardTruncationV1 {
    fn new(truncation: CardTruncation) -> Self {
        Self {
            requested_end: truncation.requested_end,
            last_row: truncation.last_row,
            reach: truncation.reach,
        }
    }
}

/// Back to the engine type, so the wording lives on [`CardTruncation`] alone and no surface
/// has to restate the field mapping in order to say the same sentence.
impl From<CardTruncationV1> for CardTruncation {
    fn from(wire: CardTruncationV1) -> Self {
        Self {
            requested_end: wire.requested_end,
            last_row: wire.last_row,
            reach: wire.reach,
        }
    }
}

#[derive(Debug, thiserror::Error)]
pub enum CardServiceError {
    #[error("invalid request: {0}")]
    InvalidRequest(String),
    #[error("zero solve failed: {0}")]
    ZeroSolve(String),
    #[error("trajectory failed: {0}")]
    Trajectory(String),
    /// PDF rendering itself failed (font load/parse, or an empty row set). Gated with the
    /// only code path that can produce it, so a build without `pdf` sees an unchanged enum.
    #[cfg(feature = "pdf")]
    #[error("pdf generation failed: {0}")]
    Pdf(String),
    /// The card is too big to print ([`MAX_PDF_ROWS`] / [`MAX_PDF_PAGES`]). A resource
    /// refusal, not a failure: the request was well formed and the caller can be told
    /// exactly what is true of the card (how many rows, how many pages), so the message is
    /// carried verbatim rather than prefixed.
    #[cfg(feature = "pdf")]
    #[error("{0}")]
    TooLarge(String),
}

// --- Unit conversions: byte-identical constants to the CLI's UnitConverter ---
//
// One resolved unit per dimension, and one place where a unit becomes a factor. Every
// imperial arm and every metric arm below is the SAME EXPRESSION the single-scalar version
// of this module used for that dimension — the imperial arm for the dimensions the imperial
// preset fills, the metric arm for the metric preset's — so a request that states no
// per-dimension unit performs exactly the multiplications it always performed, in the same
// order, and lands on the same bits. The only genuinely new arms are the values neither
// preset produces (cm, km/h, knots, and feet of altitude), which no pre-existing request can
// select.
//
// The reverse conversions are deliberately NOT expressed as `to_metric` composed with a
// division. `Mph -> v` and `Mph -> (v * 0.44704) / 0.44704` are not the same f64, and the
// pre-existing behaviour is the former.

/// m/s per km/h.
const KPH_TO_MPS: f64 = 1.0 / 3.6;
/// m/s per knot: one nautical mile (1852 m) per hour, exactly.
const KNOTS_TO_MPS: f64 = 1852.0 / 3600.0;

struct Units {
    distance: CardDistanceUnit,
    velocity: CardVelocityUnit,
    mass: CardMassUnit,
    diameter: CardDiameterUnit,
    sight_height: CardLinearUnit,
    drop: CardLinearUnit,
    wind_speed: CardWindSpeedUnit,
    temperature: CardTemperatureUnit,
    pressure: CardPressureUnit,
    energy: CardEnergyUnit,
    altitude: CardAltitudeUnit,
}

impl Units {
    /// Settle every dimension: the request's own field where it states one, otherwise what
    /// the [`CardUnits`] preset implies.
    ///
    /// Pure, cheap, and free of validation, so the PDF path can settle the axes it prints in
    /// the header without repeating the request checks `resolve_inner` performs.
    fn resolve(req: &CardRequestV1) -> Self {
        let imperial = req.units == CardUnits::Imperial;
        let linear = if imperial { CardLinearUnit::Inches } else { CardLinearUnit::Mm };
        Self {
            distance: req.distance_unit.unwrap_or(if imperial {
                CardDistanceUnit::Yards
            } else {
                CardDistanceUnit::Meters
            }),
            velocity: req.velocity_unit.unwrap_or(if imperial {
                CardVelocityUnit::Fps
            } else {
                CardVelocityUnit::Mps
            }),
            mass: req.mass_unit.unwrap_or(if imperial {
                CardMassUnit::Grains
            } else {
                CardMassUnit::Grams
            }),
            diameter: req.diameter_unit.unwrap_or(if imperial {
                CardDiameterUnit::Inches
            } else {
                CardDiameterUnit::Mm
            }),
            sight_height: req.sight_height_unit.unwrap_or(linear),
            drop: req.drop_unit.unwrap_or(linear),
            wind_speed: req.wind_speed_unit.unwrap_or(if imperial {
                CardWindSpeedUnit::Mph
            } else {
                CardWindSpeedUnit::Mps
            }),
            temperature: req.temperature_unit.unwrap_or(if imperial {
                CardTemperatureUnit::Fahrenheit
            } else {
                CardTemperatureUnit::Celsius
            }),
            pressure: req.pressure_unit.unwrap_or(if imperial {
                CardPressureUnit::InHg
            } else {
                CardPressureUnit::HPa
            }),
            energy: req.energy_unit.unwrap_or(if imperial {
                CardEnergyUnit::FtLb
            } else {
                CardEnergyUnit::Joules
            }),
            // NOT `imperial`-dependent: `altitude` has always been metres in both systems.
            // See CardRequestV1::altitude_unit for why the preset must never reach it.
            altitude: req.altitude_unit.unwrap_or(CardAltitudeUnit::Meters),
        }
    }

    // --- into SI, for the solver ---

    fn velocity_to_metric(&self, v: f64) -> f64 {
        match self.velocity {
            CardVelocityUnit::Fps => v * 0.3048,
            CardVelocityUnit::Mps => v,
        }
    }
    fn mass_to_kg(&self, v: f64) -> f64 {
        match self.mass {
            CardMassUnit::Grains => v * GRAINS_TO_KG,
            CardMassUnit::Grams => v * 0.001,
        }
    }
    fn distance_to_metric(&self, v: f64) -> f64 {
        match self.distance {
            CardDistanceUnit::Yards => v * 0.9144,
            CardDistanceUnit::Meters => v,
        }
    }
    fn linear_to_metric(unit: CardLinearUnit, v: f64) -> f64 {
        match unit {
            CardLinearUnit::Inches => v * 0.0254,
            CardLinearUnit::Cm => v * 0.01,
            CardLinearUnit::Mm => v * 0.001,
        }
    }
    fn diameter_to_metric(&self, v: f64) -> f64 {
        match self.diameter {
            CardDiameterUnit::Inches => v * 0.0254,
            CardDiameterUnit::Mm => v * 0.001,
        }
    }
    fn sight_height_to_metric(&self, v: f64) -> f64 {
        Self::linear_to_metric(self.sight_height, v)
    }
    fn drop_to_metric(&self, v: f64) -> f64 {
        Self::linear_to_metric(self.drop, v)
    }
    fn wind_to_metric(&self, v: f64) -> f64 {
        match self.wind_speed {
            CardWindSpeedUnit::Mph => v * 0.44704,
            CardWindSpeedUnit::Mps => v,
            CardWindSpeedUnit::Kph => v * KPH_TO_MPS,
            CardWindSpeedUnit::Knots => v * KNOTS_TO_MPS,
        }
    }
    fn temperature_to_metric(&self, v: f64) -> f64 {
        match self.temperature {
            CardTemperatureUnit::Fahrenheit => (v - 32.0) * 5.0 / 9.0,
            CardTemperatureUnit::Celsius => v,
        }
    }
    fn pressure_to_metric(&self, v: f64) -> f64 {
        match self.pressure {
            CardPressureUnit::InHg => v * 33.8639,
            CardPressureUnit::HPa => v,
        }
    }
    /// `altitude` in metres, which is what [`AtmosphericConditions::altitude`] wants and
    /// what every request that predates `altitude_unit` already supplied.
    fn altitude_to_metric(&self, v: f64) -> f64 {
        match self.altitude {
            CardAltitudeUnit::Feet => v * 0.3048,
            CardAltitudeUnit::Meters => v,
        }
    }

    // --- back out, for the printed row ---

    fn velocity_from_metric(&self, v: f64) -> f64 {
        match self.velocity {
            CardVelocityUnit::Fps => v / 0.3048,
            CardVelocityUnit::Mps => v,
        }
    }
    fn distance_from_metric(&self, v: f64) -> f64 {
        match self.distance {
            CardDistanceUnit::Yards => v / 0.9144,
            CardDistanceUnit::Meters => v,
        }
    }
    fn energy_from_metric(&self, v: f64) -> f64 {
        match self.energy {
            CardEnergyUnit::FtLb => v * 0.737562,
            CardEnergyUnit::Joules => v,
        }
    }
    fn drop_linear_from_metric(&self, v: f64) -> f64 {
        match self.drop {
            CardLinearUnit::Inches => v / 0.0254,
            CardLinearUnit::Cm => v * 100.0,
            CardLinearUnit::Mm => v * 1000.0,
        }
    }

    // --- column labels; the response's `units` block IS this, field for field ---

    /// Label for the distance axis, for error messages that must name a range.
    fn distance_label(&self) -> &'static str {
        match self.distance {
            CardDistanceUnit::Yards => "yd",
            CardDistanceUnit::Meters => "m",
        }
    }
    fn velocity_label(&self) -> &'static str {
        match self.velocity {
            CardVelocityUnit::Fps => "fps",
            CardVelocityUnit::Mps => "m/s",
        }
    }
    fn energy_label(&self) -> &'static str {
        match self.energy {
            CardEnergyUnit::FtLb => "ft-lb",
            CardEnergyUnit::Joules => "J",
        }
    }
    fn drop_label(&self) -> &'static str {
        match self.drop {
            CardLinearUnit::Inches => "in",
            CardLinearUnit::Cm => "cm",
            CardLinearUnit::Mm => "mm",
        }
    }
    fn wind_speed_label(&self) -> &'static str {
        match self.wind_speed {
            CardWindSpeedUnit::Mph => "mph",
            CardWindSpeedUnit::Mps => "m/s",
            CardWindSpeedUnit::Kph => "km/h",
            CardWindSpeedUnit::Knots => "kn",
        }
    }

    // --- the CLI's own axes, for the call sites that are pinned to CLI constants ---
    //
    // These do NOT reuse the conversions above. The BC5D path and the PDF header each
    // hardcode a factor that `main.rs` hardcodes at the same call site (3.280_839_895 vs
    // 1/0.3048; 15.4324 vs the exact grains-per-gram), and the golden tests pin the CLI's
    // digits. Sharing one factor between them would change numbers this work must not.
    //
    // Each of these helpers is compiled exactly where its call site is, rather than being
    // blanket-allowed as dead code, so that a helper which stops being called anywhere is
    // still a warning. For most of them that means a cfg attribute; `velocity_to_fps_bc5d`
    // carries none because its call site is the always-compiled `bc_segments` arm of
    // `resolve_bc_schedule`, so it is reached in every build.

    /// Display velocity -> fps using `parse_bc_segment`'s factor.
    fn velocity_to_fps_bc5d(&self, v: f64) -> f64 {
        match self.velocity {
            CardVelocityUnit::Fps => v,
            CardVelocityUnit::Mps => v * 3.280_839_895,
        }
    }
    /// Bullet weight -> grains using the CLI's BC5D/PDF factor.
    #[cfg(any(feature = "pdf", not(target_arch = "wasm32")))]
    fn mass_to_grains_cli(&self, v: f64) -> f64 {
        match self.mass {
            CardMassUnit::Grains => v,
            CardMassUnit::Grams => v * 15.4324,
        }
    }
    /// Bullet diameter -> inches, the BC5D table's caliber axis (which no wasm32 build has,
    /// there being no filesystem to read a table from).
    #[cfg(not(target_arch = "wasm32"))]
    fn diameter_to_inches(&self, v: f64) -> f64 {
        match self.diameter {
            CardDiameterUnit::Inches => v,
            CardDiameterUnit::Mm => v / 25.4,
        }
    }
    /// Muzzle velocity -> fps for the PDF header (`main.rs` divides by 0.3048 here).
    #[cfg(feature = "pdf")]
    fn velocity_to_fps_pdf(&self, v: f64) -> f64 {
        match self.velocity {
            CardVelocityUnit::Fps => v,
            CardVelocityUnit::Mps => v / 0.3048,
        }
    }
    /// Pressure -> inHg for the PDF header (`main.rs` uses 33.8639 on both PDF call sites,
    /// NOT `pdf_dope_card::INHG_TO_HPA`).
    #[cfg(feature = "pdf")]
    fn pressure_to_inhg(&self, v: f64) -> f64 {
        match self.pressure {
            CardPressureUnit::InHg => v,
            CardPressureUnit::HPa => v / 33.8639,
        }
    }
    /// Pressure -> hPa for the PDF header, same factor.
    #[cfg(feature = "pdf")]
    fn pressure_to_hpa(&self, v: f64) -> f64 {
        match self.pressure {
            CardPressureUnit::InHg => v * 33.8639,
            CardPressureUnit::HPa => v,
        }
    }
    /// Temperature -> °F for the PDF header.
    #[cfg(feature = "pdf")]
    fn temperature_to_fahrenheit(&self, v: f64) -> f64 {
        match self.temperature {
            CardTemperatureUnit::Fahrenheit => v,
            CardTemperatureUnit::Celsius => v * 9.0 / 5.0 + 32.0,
        }
    }
    /// Wind or crossing speed -> mph for the PDF header.
    #[cfg(feature = "pdf")]
    fn wind_to_mph(&self, v: f64) -> f64 {
        match self.wind_speed {
            CardWindSpeedUnit::Mph => v,
            CardWindSpeedUnit::Mps => v / 0.44704,
            CardWindSpeedUnit::Kph => v * KPH_TO_MPS / 0.44704,
            CardWindSpeedUnit::Knots => v * KNOTS_TO_MPS / 0.44704,
        }
    }
    /// Altitude -> feet for the PDF header, which reports the altitude the SOLVE used.
    #[cfg(feature = "pdf")]
    fn altitude_to_feet(&self, v: f64) -> f64 {
        match self.altitude {
            CardAltitudeUnit::Feet => v,
            CardAltitudeUnit::Meters => v / 0.3048,
        }
    }

    // --- defaults that are themselves denominated ---

    /// The CLI's 1.5-inch sight height, restated in this request's `sight_height_unit`.
    /// The millimetre figure is the module's long-standing 38.0 (a rounded 1.5 in, not a
    /// conversion of it), and the centimetre figure is that same length, exactly.
    fn default_sight_height(&self) -> f64 {
        match self.sight_height {
            CardLinearUnit::Inches => 1.5,
            CardLinearUnit::Cm => 3.8,
            CardLinearUnit::Mm => 38.0,
        }
    }
    /// The CLI's 59 °F / 15 °C standard temperature, in this request's `temperature_unit`.
    fn default_temperature(&self) -> f64 {
        match self.temperature {
            CardTemperatureUnit::Fahrenheit => 59.0,
            CardTemperatureUnit::Celsius => 15.0,
        }
    }
    /// The CLI's 29.92 inHg / 1013.25 hPa standard pressure, in this request's
    /// `pressure_unit`.
    fn default_pressure(&self) -> f64 {
        match self.pressure {
            CardPressureUnit::InHg => 29.92,
            CardPressureUnit::HPa => 1013.25,
        }
    }
}

struct Resolved {
    u: Units,
    velocity_m: f64,
    mass_kg: f64,
    diameter_m: f64,
    sight_height_m: f64,
    zero_distance_m: f64,
    zero_poi_vertical_m: f64,
    zero_poi_horizontal_m: f64,
    sight_offset_lateral_m: f64,
    temperature_c: f64,
    pressure_hpa: f64,
    /// Site elevation in metres — which is what `altitude` already was for every request
    /// that states no `altitude_unit`, so this is a rename of an identity, not a conversion.
    altitude_m: f64,
    end_m: f64,
    elevation_click: Option<ClickValue>,
    windage_click: Option<ClickValue>,
    windage_unit: AdjustmentUnit,
    drag_model: DragModel,
    /// Scalar BC actually fed to the zero solve and every sampled run. Equals the
    /// request's `ballistic_coefficient` unless a BC5D table applied its muzzle
    /// correction (CLI parity: the scalar is the schedule's interior-gap fallback).
    bc_for_solve: f64,
    /// Velocity-keyed BC schedule (velocities in fps, the unit the solver compares
    /// against), from explicit `bc_segments` or generated from `bc5d_table_path`.
    bc_segments_fps: Option<Vec<BCSegmentData>>,
}

fn resolve(req: &CardRequestV1) -> Result<Resolved, CardServiceError> {
    resolve_inner(req, true)
}

/// [`resolve`] WITHOUT the BC-schedule step — the request's declared axes, click
/// graduations and unit conversions, and nothing that touches the filesystem.
///
/// This is the resolve `card.pdf` performs when it prints caller-supplied rows: those rows
/// were computed elsewhere (they are the stored `card.range_table` response), so opening
/// `bc5d_table_path` here would do nothing but couple a reprint to whether that file is
/// still on the device — which is exactly the coupling printing stored rows exists to
/// break. `bc_for_solve` is left at the request's published BC and is not used for the
/// footer on that path (the stored card's own `bc_for_solve` is).
#[cfg(feature = "pdf")]
fn resolve_axes_only(req: &CardRequestV1) -> Result<Resolved, CardServiceError> {
    resolve_inner(req, false)
}

fn resolve_inner(req: &CardRequestV1, load_bc_schedule: bool) -> Result<Resolved, CardServiceError> {
    let u = Units::resolve(req);

    for (name, v) in [
        ("muzzle_velocity", req.muzzle_velocity),
        ("ballistic_coefficient", req.ballistic_coefficient),
        ("mass", req.mass),
        ("diameter", req.diameter),
        ("zero_distance", req.zero_distance),
        ("start", req.start),
        ("end", req.end),
        ("step", req.step),
    ] {
        if !v.is_finite() || v <= 0.0 {
            return Err(CardServiceError::InvalidRequest(format!(
                "{name} must be finite and positive, got {v}"
            )));
        }
    }
    if req.end < req.start {
        return Err(CardServiceError::InvalidRequest(
            "end must be >= start".into(),
        ));
    }
    if !(0.5..=1.5).contains(&req.elevation_cf) || !(0.5..=1.5).contains(&req.windage_cf) {
        return Err(CardServiceError::InvalidRequest(
            "tracking correction factors must lie in (0.5, 1.5)".into(),
        ));
    }

    // Each of these three defaults is itself denominated, so it is taken from the dimension
    // that governs the field rather than from the preset (`Units` restates the CLI's own
    // figures per unit). `sight_height`'s sentinel is NaN because a serde field default
    // cannot see the request it belongs to.
    let sight_height = if req.sight_height.is_nan() {
        u.default_sight_height()
    } else {
        req.sight_height
    };
    let temperature = req.temperature.unwrap_or_else(|| u.default_temperature());
    let pressure = req.pressure.unwrap_or_else(|| u.default_pressure());

    let parse_click = |label: &str, s: &Option<String>| -> Result<Option<ClickValue>, CardServiceError> {
        s.as_deref()
            .map(|raw| {
                parse_click_value(raw).map_err(|e| {
                    CardServiceError::InvalidRequest(format!("{label}: {e}"))
                })
            })
            .transpose()
    };
    let elevation_click = parse_click("elevation_click_value", &req.elevation_click_value)?;
    let windage_click =
        parse_click("windage_click_value", &req.windage_click_value)?.or(elevation_click);

    if req.adjustment_unit == AdjustmentUnit::Clicks && elevation_click.is_none() {
        return Err(CardServiceError::InvalidRequest(
            "adjustment_unit 'clicks' requires elevation_click_value".into(),
        ));
    }
    let windage_unit = req.windage_unit.unwrap_or(req.adjustment_unit);
    if windage_unit == AdjustmentUnit::Clicks && windage_click.is_none() {
        return Err(CardServiceError::InvalidRequest(
            "windage_unit 'clicks' requires a click graduation".into(),
        ));
    }

    let (bc_for_solve, bc_segments_fps) = if load_bc_schedule {
        resolve_bc_schedule(req, &u)?
    } else {
        (req.ballistic_coefficient, None)
    };

    Ok(Resolved {
        bc_for_solve,
        bc_segments_fps,
        velocity_m: u.velocity_to_metric(req.muzzle_velocity),
        mass_kg: u.mass_to_kg(req.mass),
        diameter_m: u.diameter_to_metric(req.diameter),
        sight_height_m: u.sight_height_to_metric(sight_height),
        zero_distance_m: u.distance_to_metric(req.zero_distance),
        zero_poi_vertical_m: u.drop_to_metric(req.zero_poi_vertical),
        zero_poi_horizontal_m: u.drop_to_metric(req.zero_poi_horizontal),
        sight_offset_lateral_m: u.sight_height_to_metric(req.sight_offset_lateral),
        temperature_c: u.temperature_to_metric(temperature),
        pressure_hpa: u.pressure_to_metric(pressure),
        altitude_m: u.altitude_to_metric(req.altitude),
        end_m: u.distance_to_metric(req.end),
        elevation_click,
        windage_click,
        windage_unit,
        drag_model: req.drag_model.into(),
        u,
    })
}

/// Resolve the velocity-keyed BC schedule for a card request, mirroring the CLI's
/// precedence exactly: explicit `bc_segments` win outright (and leave the scalar BC
/// untouched, like `--bc-segment` restoring the pre-table BC); otherwise a
/// `bc5d_table_path` generates the same segment ladder `--bc-table-dir` does and
/// applies the table's muzzle correction to the scalar fallback BC.
///
/// Returns `(scalar BC for the solve, optional fps-keyed schedule)`.
fn resolve_bc_schedule(
    req: &CardRequestV1,
    u: &Units,
) -> Result<(f64, Option<Vec<BCSegmentData>>), CardServiceError> {
    if let Some(segments) = req.bc_segments.as_ref().filter(|s| !s.is_empty()) {
        let mut converted = Vec::with_capacity(segments.len());
        for (index, segment) in segments.iter().enumerate() {
            if !segment.velocity_min.is_finite()
                || !segment.velocity_max.is_finite()
                || !segment.bc.is_finite()
            {
                return Err(CardServiceError::InvalidRequest(format!(
                    "bc_segments[{index}]: velocity_min, velocity_max, and bc must be finite"
                )));
            }
            if segment.velocity_min >= segment.velocity_max {
                return Err(CardServiceError::InvalidRequest(format!(
                    "bc_segments[{index}]: velocity_min must be < velocity_max"
                )));
            }
            if segment.bc <= 0.0 {
                return Err(CardServiceError::InvalidRequest(format!(
                    "bc_segments[{index}]: bc must be > 0"
                )));
            }
            converted.push(BCSegmentData {
                // Display velocity -> fps with the exact factor the CLI's parse_bc_segment
                // uses; the band edges are in `velocity_unit`, like `muzzle_velocity`.
                velocity_min: u.velocity_to_fps_bc5d(segment.velocity_min),
                velocity_max: u.velocity_to_fps_bc5d(segment.velocity_max),
                bc_value: segment.bc,
            });
        }
        return Ok((req.ballistic_coefficient, Some(converted)));
    }

    let Some(path) = req.bc5d_table_path.as_deref() else {
        return Ok((req.ballistic_coefficient, None));
    };

    #[cfg(target_arch = "wasm32")]
    {
        let _ = (path, u);
        Err(CardServiceError::InvalidRequest(
            "bc5d_table_path is not supported on this target (no filesystem); supply \
             bc_segments instead"
                .into(),
        ))
    }
    #[cfg(not(target_arch = "wasm32"))]
    {
        // The table's CONTENT must be for this shot's caliber, not merely CRC-valid: a
        // wrong-caliber table silently biases every row of the card (see
        // Bc5dTable::ensure_caliber_matches), so a mismatch is refused here rather than
        // applied — and rather than quietly falling back to an uncorrected card, which
        // the caller could not distinguish from a corrected one.
        let diameter_in = u.diameter_to_inches(req.diameter);
        let table = crate::bc_table_5d::path_cache::load_verified_for_caliber(
            std::path::Path::new(path),
            diameter_in,
        )
        .map_err(|e| CardServiceError::InvalidRequest(format!("bc5d_table_path: {e}")))?;

        // The BC5D axes are grains + fps; convert from the request's units the same
        // way the CLI does. The v2 tables carry only G1/G7 planes — anything else is
        // typed as G1, matching the CLI/WASM coercion.
        let weight_grains = u.mass_to_grains_cli(req.mass);
        let muzzle_fps = u.velocity_to_fps_bc5d(req.muzzle_velocity);
        let drag_type = if req.drag_model == DragModelV1::G7 { "G7" } else { "G1" };
        let base_bc = req.ballistic_coefficient;

        match table.generate_segments(base_bc, drag_type, weight_grains, Some(muzzle_fps)) {
            Some(segments) => {
                // CLI parity: the scalar BC becomes the muzzle-corrected fallback for
                // interior coverage gaps (main.rs: trued_bc = base * muzzle_correction).
                let fallback_bc =
                    table.get_effective_bc(weight_grains, base_bc, muzzle_fps, muzzle_fps, drag_type);
                Ok((fallback_bc, Some(segments)))
            }
            // A neutral table (every sampled cell ~= 1.0) carries no correction:
            // keep the published scalar BC, exactly like the CLI.
            None => Ok((base_bc, None)),
        }
    }
}

fn solve_zero(req: &CardRequestV1, r: &Resolved) -> Result<f64, CardServiceError> {
    let zero_inputs = BallisticInputs {
        bc_value: r.bc_for_solve,
        bc_type: r.drag_model,
        bullet_mass: r.mass_kg,
        muzzle_velocity: r.velocity_m,
        bullet_diameter: r.diameter_m,
        bullet_length: crate::truing::fallback_bullet_length_m(r.diameter_m, r.mass_kg),
        sight_height: r.sight_height_m,
        use_rk4: true,
        zero_poi_vertical_m: r.zero_poi_vertical_m,
        zero_poi_horizontal_m: r.zero_poi_horizontal_m,
        sight_offset_lateral_m: r.sight_offset_lateral_m,
        // The zero MUST be solved with the same velocity-keyed BC schedule the
        // sampled flights use (the 0.22.11 auto-zero lesson: a schedule that changes
        // early-flight drag would otherwise mis-zero every row).
        use_bc_segments: r.bc_segments_fps.is_some(),
        bc_segments_data: r.bc_segments_fps.clone(),
        ..Default::default()
    };
    let atmosphere = AtmosphericConditions {
        temperature: r.temperature_c,
        pressure: r.pressure_hpa,
        humidity: req.humidity,
        altitude: r.altitude_m,
    };
    crate::calculate_zero_angle_with_conditions(
        zero_inputs,
        r.zero_distance_m,
        r.sight_height_m,
        WindConditions::default(),
        atmosphere,
    )
    .map_err(|e| CardServiceError::ZeroSolve(e.to_string()))
}

#[allow(clippy::too_many_arguments)]
fn sampled(
    req: &CardRequestV1,
    r: &Resolved,
    wind_speed_m: f64,
    wind_direction_deg: f64,
    zero_angle: f64,
) -> Result<SampledFlight, CardServiceError> {
    run_sampled_flight(
        r.velocity_m,
        r.bc_for_solve,
        r.mass_kg,
        r.diameter_m,
        r.drag_model,
        r.sight_height_m,
        r.temperature_c,
        r.pressure_hpa,
        req.humidity,
        r.altitude_m,
        wind_speed_m,
        wind_direction_deg,
        // MBA-1476: at least one whole grid cell past the last row asked for — see
        // card_sample_max_range_m for why that headroom decides reachability.
        card_sample_max_range_m(r.end_m),
        // MBA-1476: the engine's own card grid, never the request's row spacing.
        CARD_SAMPLE_INTERVAL_M,
        zero_angle,
        // The same schedule as the zero solve above, on every card surface
        // (come-ups, range table, wind card) consistently.
        r.bc_segments_fps.clone(),
        None,
        None,
        r.zero_poi_vertical_m,
        r.zero_poi_horizontal_m,
        r.sight_offset_lateral_m,
        Some(r.zero_distance_m),
    )
    .map_err(|e| CardServiceError::Trajectory(e.to_string()))
}

/// Read one card row's sample off a solved flight, or `None` when the flight does not span it
/// (MBA-1476).
///
/// The nearest-sample-within-one-and-a-half-steps search this replaces answered a range the
/// flight had no sample for with a neighbour's numbers, invisibly. There is no tolerance to
/// tune here: [`sample_at_range`] interpolates exactly at the requested range whenever the
/// flight spans it, and a range it cannot span yields nothing at all. The caller ends the card
/// there and records a [`CardTruncation`] — it never substitutes, and it never drops the row
/// without saying so.
fn row_sample(
    samples: &[crate::trajectory_sampling::TrajectorySample],
    range_m: f64,
) -> Option<crate::trajectory_sampling::TrajectorySample> {
    sample_at_range(samples, range_m)
}

/// The error a card returns when its flight reaches NONE of the rows requested — there is no
/// card to truncate, so this stays a refusal (MBA-1476 follow-up).
fn no_rows_error(first_row_display: f64, reach_m: f64, u: &Units) -> CardServiceError {
    CardServiceError::Trajectory(range_not_sampled_message(
        first_row_display,
        u.distance_from_metric(reach_m),
        u.distance_label(),
    ))
}

/// Record the truncation of a card whose row loop stopped early, in display units.
fn truncation_of(
    requested_end: f64,
    last_row: f64,
    reach_m: f64,
    u: &Units,
) -> Option<CardTruncationV1> {
    Some(CardTruncationV1::new(CardTruncation {
        requested_end,
        last_row,
        reach: u.distance_from_metric(reach_m),
    }))
}

/// The response's per-dimension label block: one label per dimension, read off the same
/// resolved [`Units`] the rows were converted with, so a column cannot be computed in one
/// unit and headed with another.
fn units_block(req: &CardRequestV1, u: &Units, windage_unit: AdjustmentUnit) -> CardUnitsBlockV1 {
    CardUnitsBlockV1 {
        distance: u.distance_label(),
        velocity: u.velocity_label(),
        energy: u.energy_label(),
        drop: u.drop_label(),
        wind_speed: u.wind_speed_label(),
        elevation_adjustment: adjustment_unit_label(req.adjustment_unit),
        windage_adjustment: adjustment_unit_label(windage_unit),
    }
}

/// Come-ups: elevation dial per range plus the incremental come-up between rows.
/// Mirrors the CLI's `come-ups` command row-for-row.
pub fn come_ups_v1(req: &CardRequestV1) -> Result<CardResponseV1, CardServiceError> {
    let r = resolve(req)?;
    let zero_angle = solve_zero(req, &r)?;
    let flight = sampled(
        req,
        &r,
        r.u.wind_to_metric(req.wind_speed),
        req.wind_direction_deg,
        zero_angle,
    )?;

    let mut rows: Vec<CardRowV1> = Vec::new();
    let mut truncation = None;
    let mut prev_drop_adj: f64 = 0.0;
    let mut current_range = req.start;
    while current_range <= req.end + 0.1 {
        let range_m = r.u.distance_to_metric(current_range);
        let Some(sample) = row_sample(&flight.samples, range_m) else {
            let Some(last) = rows.last() else {
                return Err(no_rows_error(current_range, flight.reach_m, &r.u));
            };
            truncation = truncation_of(req.end, last.range, flight.reach_m, &r.u);
            break;
        };
        let drop_yd = r.u.distance_from_metric(sample.drop_m);
        // MBA-1476: the row's OWN range, not a nearby sample's — the angular conversion
        // divides by the distance the row is labelled with.
        let drop_adj = adjustment_display(
            drop_yd,
            current_range,
            req.adjustment_unit,
            r.elevation_click,
            req.zero_set_elevation_bias_mil,
            req.elevation_cf,
        )
        .value;
        rows.push(CardRowV1 {
            range: current_range,
            drop_linear: None,
            drop_adj: Some(drop_adj),
            come_up: Some(drop_adj - prev_drop_adj),
            wind_linear: None,
            wind_adj: None,
            velocity: Some(r.u.velocity_from_metric(sample.velocity_mps)),
            energy: Some(r.u.energy_from_metric(sample.energy_j)),
            time: Some(sample.time_s),
            wind_columns: Vec::new(),
        });
        prev_drop_adj = drop_adj;
        current_range += req.step;
    }

    Ok(CardResponseV1 {
        schema_version: CARD_SCHEMA_VERSION_V1,
        kind: "come_ups",
        zero_distance: req.zero_distance,
        bc_for_solve: r.bc_for_solve,
        units: units_block(req, &r.u, req.adjustment_unit),
        wind_speeds: Vec::new(),
        wind_angles_deg: Vec::new(),
        rows,
        extra_angle_rows: Vec::new(),
        truncation,
    })
}

/// Range table: drop and wind on possibly different display axes.
/// Mirrors the CLI's `range-table` command row-for-row (two solves: with wind for
/// drift, without wind for the pure elevation axis).
pub fn range_table_v1(req: &CardRequestV1) -> Result<CardResponseV1, CardServiceError> {
    // Discards the PDF-only extras; see `range_table_rows`.
    range_table_rows(req, None).map(|(card, _lead, _bc)| card)
}

/// The one range-table computation, shared by the on-screen `card.range_table` and the
/// printable `card.pdf`.
///
/// Returns `(the card, the Lead column, the scalar BC the solve ran with)`:
///
/// * The card is exactly what `card.range_table` serializes — `range_table_v1` is a
///   discarding wrapper, so the PDF's Range/Drop/Wind figures ARE the on-screen figures.
///   They are taken from here rather than recomputed because recomputing them is the one
///   way the printed card could ever disagree with the rows the shooter already read.
/// * The Lead column parallels `card.rows` one-to-one. `lead_target_speed` is in the
///   request's `wind_speed_unit`; `None` (no lead requested) yields all `None`, which the
///   PDF renders as em-dashes rather than fake zeroes.
/// * The BC is `Resolved::bc_for_solve` — the published BC unless a BC5D table applied its
///   muzzle correction — which the PDF footer prints. The footer must state the BC the
///   numbers above it came from, not the one the request nominated.
fn range_table_rows(
    req: &CardRequestV1,
    lead_target_speed: Option<f64>,
) -> Result<(CardResponseV1, Vec<Option<f64>>, f64), CardServiceError> {
    let r = resolve(req)?;
    let zero_angle = solve_zero(req, &r)?;
    let wind_flight = sampled(
        req,
        &r,
        r.u.wind_to_metric(req.wind_speed),
        req.wind_direction_deg,
        zero_angle,
    )?;
    let no_wind_flight = sampled(req, &r, 0.0, 0.0, zero_angle)?;
    // Both flights feed every row, so the card can only run as far as the shorter of them.
    let reach_m = wind_flight.reach_m.min(no_wind_flight.reach_m);
    let lead_speed_mps = lead_target_speed.map(|speed| r.u.wind_to_metric(speed));

    let mut rows: Vec<CardRowV1> = Vec::new();
    let mut lead_adj = Vec::new();
    let mut truncation = None;
    let mut current_range = req.start;
    while current_range <= req.end + 0.1 {
        let range_m = r.u.distance_to_metric(current_range);
        let (Some(nw), Some(w)) = (
            row_sample(&no_wind_flight.samples, range_m),
            row_sample(&wind_flight.samples, range_m),
        ) else {
            let Some(last) = rows.last() else {
                return Err(no_rows_error(current_range, reach_m, &r.u));
            };
            truncation = truncation_of(req.end, last.range, reach_m, &r.u);
            break;
        };
        // MBA-1476: the row's OWN range on both axes.
        let range_display = current_range;
        let drop_yd = r.u.distance_from_metric(nw.drop_m);
        let drop_adj = adjustment_display(
            drop_yd,
            range_display,
            req.adjustment_unit,
            r.elevation_click,
            req.zero_set_elevation_bias_mil,
            req.elevation_cf,
        )
        .value;
        let drift_yd = r.u.distance_from_metric(w.wind_drift_m);
        let wind_adj = windage_adjustment_display(
            drift_yd,
            range_display,
            r.windage_unit,
            r.windage_click,
            req.zero_set_windage_bias_mil,
            req.windage_cf,
        )
        .value;
        // Lead is the windage-axis hold for a full-value 90° crossing target:
        // `speed * time-of-flight` (MBA-1287's shared moving-target math), on the
        // no-wind sample whose time this row already reports. It is a COMPONENT
        // hold composed on top of the wind dial, which already carries the
        // zero-set bias, so it stays bias-free (bias 0.0) while still being
        // divided by the windage tracking CF — the same treatment
        // `main.rs::dope_card_row_from_sample` gives the CLI's Lead column.
        lead_adj.push(lead_speed_mps.map(|speed_mps| {
            let lead_display = r.u.distance_from_metric(
                crate::lead_from_tof(speed_mps, 90.0, nw.time_s, nw.distance_m).lead_m,
            );
            windage_adjustment_display(
                lead_display,
                range_display,
                r.windage_unit,
                r.windage_click,
                0.0,
                req.windage_cf,
            )
            .value
        }));
        rows.push(CardRowV1 {
            range: current_range,
            drop_linear: Some(r.u.drop_linear_from_metric(nw.drop_m)),
            drop_adj: Some(drop_adj),
            come_up: None,
            wind_linear: Some(r.u.drop_linear_from_metric(w.wind_drift_m)),
            wind_adj: Some(wind_adj),
            velocity: Some(r.u.velocity_from_metric(nw.velocity_mps)),
            energy: Some(r.u.energy_from_metric(nw.energy_j)),
            time: Some(nw.time_s),
            wind_columns: Vec::new(),
        });
        current_range += req.step;
    }

    debug_assert_eq!(rows.len(), lead_adj.len(), "Lead column must parallel rows");
    Ok((
        CardResponseV1 {
            schema_version: CARD_SCHEMA_VERSION_V1,
            kind: "range_table",
            zero_distance: req.zero_distance,
            bc_for_solve: r.bc_for_solve,
            units: units_block(req, &r.u, r.windage_unit),
            wind_speeds: Vec::new(),
            wind_angles_deg: Vec::new(),
            rows,
            extra_angle_rows: Vec::new(),
            truncation,
        },
        lead_adj,
        r.bc_for_solve,
    ))
}

/// Wind card: drift matrix, ranges x wind speeds, per wind-FROM angle.
/// Mirrors the CLI's `wind-card` command cell-for-cell. The windage axis uses
/// `adjustment_unit` (the CLI's wind-card has a single unit for the matrix).
pub fn wind_card_v1(req: &CardRequestV1) -> Result<CardResponseV1, CardServiceError> {
    if req.wind_speeds.is_empty() {
        return Err(CardServiceError::InvalidRequest(
            "wind_card requires at least one entry in wind_speeds".into(),
        ));
    }
    let r = resolve(req)?;
    let zero_angle = solve_zero(req, &r)?;

    let angles: Vec<f64> = if req.wind_angles_deg.is_empty() {
        vec![90.0]
    } else {
        req.wind_angles_deg.clone()
    };

    let mut ranges: Vec<f64> = Vec::new();
    let mut current = req.start;
    while current <= req.end + 0.1 {
        ranges.push(current);
        current += req.step;
    }

    // Every angle/speed combination is its own flight and they do not all reach equally far
    // (a crosswind bleeds a little velocity), but the matrix has ONE range axis. So the card
    // runs to the shortest of them: `reachable` is the number of leading ranges every flight
    // supplies, and the ragged tails are trimmed once, below.
    let mut reachable = ranges.len();
    let mut reach_m = f64::INFINITY;
    let mut per_angle_drifts: Vec<Vec<Vec<f64>>> = Vec::new();
    for &angle_deg in &angles {
        let mut all_drifts: Vec<Vec<f64>> = vec![Vec::new(); ranges.len()];
        for &ws in &req.wind_speeds {
            let flight = sampled(req, &r, r.u.wind_to_metric(ws), angle_deg, zero_angle)?;
            reach_m = reach_m.min(flight.reach_m);
            for (ri, &range_display) in ranges.iter().enumerate().take(reachable) {
                let range_m = r.u.distance_to_metric(range_display);
                // MBA-1476: a range the flight cannot supply used to become a 0.0 cell —
                // indistinguishable from "no drift at all" on a wind card.
                let Some(sample) = row_sample(&flight.samples, range_m) else {
                    reachable = ri;
                    break;
                };
                let drift_yd = r.u.distance_from_metric(sample.wind_drift_m);
                let drift_adj = adjustment_display(
                    drift_yd,
                    range_display,
                    req.adjustment_unit,
                    r.windage_click,
                    req.zero_set_windage_bias_mil,
                    req.windage_cf,
                )
                .value;
                all_drifts[ri].push(drift_adj);
            }
        }
        per_angle_drifts.push(all_drifts);
    }

    let truncation = if reachable < ranges.len() {
        if reachable == 0 {
            return Err(no_rows_error(ranges[0], reach_m, &r.u));
        }
        let last = ranges[reachable - 1];
        ranges.truncate(reachable);
        truncation_of(req.end, last, reach_m, &r.u)
    } else {
        None
    };

    let mut per_angle_rows: Vec<Vec<CardRowV1>> = per_angle_drifts
        .into_iter()
        .map(|all_drifts| {
            ranges
                .iter()
                .zip(all_drifts)
                .map(|(&range, wind_columns)| CardRowV1 {
                    range,
                    drop_linear: None,
                    drop_adj: None,
                    come_up: None,
                    wind_linear: None,
                    wind_adj: None,
                    velocity: None,
                    energy: None,
                    time: None,
                    wind_columns,
                })
                .collect()
        })
        .collect();

    let first = per_angle_rows.remove(0);
    Ok(CardResponseV1 {
        schema_version: CARD_SCHEMA_VERSION_V1,
        kind: "wind_card",
        zero_distance: req.zero_distance,
        bc_for_solve: r.bc_for_solve,
        units: units_block(req, &r.u, req.adjustment_unit),
        wind_speeds: req.wind_speeds.clone(),
        wind_angles_deg: angles,
        rows: first,
        extra_angle_rows: per_angle_rows,
        truncation,
    })
}

// ---------------------------------------------------------------------------------------
// Printable PDF dope card (`card.pdf`), feature `pdf`.
//
// Two ways in, one renderer:
//
// * REPRINT — the caller supplies the rows: `StoredCardV1` carries the stored
//   `card.range_table` response verbatim. Nothing is solved, no correction table is opened,
//   and the footer's BC and provenance come from that document. Only under this construction
//   is a reprint of a saved card actually a reprint: it cannot drift when the engine is
//   bumped, or when the BC5D table file at the stored path is overwritten in place by a
//   table-set refresh.
// * SOLVE — no rows supplied: `range_table_rows` runs, which is the same call
//   `card.range_table` makes (`range_table_v1` is a discarding wrapper over it). This is the
//   pre-existing behaviour, unchanged, and what the CLI-shaped caller gets.
//
// Neither path recomputes an adjustment, a unit conversion or a trajectory that the card it
// prints already performed.
// ---------------------------------------------------------------------------------------

/// The one card kind `card.pdf` can print, in the `kind` spelling
/// [`CardResponseV1::kind`] uses.
///
/// A come-ups card's Come-Up column and a wind card's swept drift matrix are not columns a
/// range-table card has, so a request for either is REFUSED rather than answered with a
/// range table — an `ok` response whose defining field was silently ignored is the defect
/// (a stored wind card exported as a range table asserted 0.0 drift on every row while the
/// screen showed up to -0.42 MIL).
#[cfg(feature = "pdf")]
pub const PDF_CARD_KIND: &str = "range_table";

/// Rows a printable card may carry.
///
/// Checked from the row count itself, before a document exists — the byte cap
/// (`bridge::MAX_PDF_BYTES`) can only refuse a 26 MB `Vec<u8>` that has already been built
/// and paginated. At ~0.5 KiB/row over the ~815 KiB font floor this bound keeps every
/// accepted card comfortably inside that cap, which remains the backstop for the other way
/// to make a huge document: header/footer labels, which are drawn verbatim on every page.
#[cfg(feature = "pdf")]
pub const MAX_PDF_ROWS: usize = 5_000;

/// Pages a printable card may run to. Bounds the one row set the row cap above cannot:
/// few enough rows, but at a font scale that fits very few of them per page.
#[cfg(feature = "pdf")]
pub const MAX_PDF_PAGES: usize = 60;

/// Where the printed rows came from — reported so a caller can verify it got a reprint
/// rather than a re-solve, which is otherwise indistinguishable from the document.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum PdfRowSource {
    /// Solved by this call, exactly as `card.range_table` would for the same request.
    Solve,
    /// Supplied by the caller: a stored `card.range_table` response, printed as-is.
    StoredRows,
}

#[cfg(feature = "pdf")]
impl PdfRowSource {
    /// Wire spelling for the bridge's `source` field.
    pub fn as_str(self) -> &'static str {
        match self {
            Self::Solve => "solve",
            Self::StoredRows => "stored_rows",
        }
    }
}

/// A card whose rows already exist: print THESE numbers instead of solving.
///
/// `card` is the stored `card.range_table` response, pasted verbatim — an app keeps that
/// document for every saved card and shows it on screen, so handing the same bytes back is
/// what makes the paper and the screen the same card by construction rather than by
/// coincidence. The two provenance strings are printed in the footer so a card in a
/// shooter's pocket can be reconciled with a screen afterwards; both are optional, and an
/// absent or empty one prints nothing at all rather than a placeholder.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Default, PartialEq, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct StoredCardV1 {
    /// The stored response, exactly as this engine emitted it.
    pub card: StoredCardResponseV1,
    /// Engine version that produced the rows (`engine_version` from the bridge envelope of
    /// the call that produced them). Absent/empty prints no engine line.
    #[serde(default)]
    pub engine_version: Option<String>,
    /// Correction-table version the rows were solved against — the published BC5D table-set
    /// version an app records at save time. Absent/empty prints no table line, which is the
    /// honest rendering of "this card used no correction table".
    #[serde(default)]
    pub bc5d_table_version: Option<String>,
}

/// A stored [`CardResponseV1`], as a deserializable mirror.
///
/// [`CardResponseV1`] is `Serialize` only (and its `units` block holds `&'static str`s), so
/// a stored response is read back through this twin. It accepts the response document
/// unchanged, field for field: `kind` and `units` are load-bearing here — they say what the
/// rows ARE and what a row MEANS — while the descriptive scalars are optional so a card
/// stored by an older engine still prints.
///
/// **Deliberately NOT `deny_unknown_fields`** (unlike [`CardRequestV1`], where the caller
/// really is the author of the document). What arrives here is this engine's own output,
/// round-tripped through an app's storage — never hand-written input — so strictness buys
/// no validation and costs a hard cross-version break: the FIRST field ever added to
/// `CardResponseV1` would make every card saved by the newer platform completely
/// unexportable on the older one. The two apps carry independent engine pins and ship
/// through separate stores, so that staggering is the normal case, not a downgrade. An
/// unknown field is therefore ignored, exactly as the apps' own decoders ignore it, and the
/// stored cells still reach the paper. The same reasoning applies to
/// [`StoredCardUnitsBlockV1`] and [`StoredCardRowV1`] below.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Default, PartialEq, Deserialize)]
pub struct StoredCardResponseV1 {
    /// Which card this is. Anything but [`PDF_CARD_KIND`] is refused.
    pub kind: String,
    /// Column labels and the range unit of the stored rows. The printed headings are
    /// these, not the request's, because these are what the rows were computed in.
    pub units: StoredCardUnitsBlockV1,
    /// The rows to print, in order.
    pub rows: Vec<StoredCardRowV1>,
    #[serde(default)]
    pub schema_version: Option<u32>,
    #[serde(default)]
    pub zero_distance: Option<f64>,
    /// The BC the stored rows were computed with; printed in the footer. Absent (a card
    /// saved by an engine before [`CardResponseV1::bc_for_solve`] existed) falls back to the
    /// request's published `ballistic_coefficient`.
    #[serde(default)]
    pub bc_for_solve: Option<f64>,
    /// Present only on a wind card, and therefore a refusal here.
    #[serde(default)]
    pub wind_speeds: Vec<f64>,
    #[serde(default)]
    pub wind_angles_deg: Vec<f64>,
    #[serde(default)]
    pub extra_angle_rows: Vec<Vec<StoredCardRowV1>>,
    /// Present only when the stored card itself stopped short of the rows its request asked
    /// for — the [`CardResponseV1::truncation`] block, round-tripped (MBA-1477).
    ///
    /// Without this the reprint path loses the notice: `StoredCardResponseV1` ignores unknown
    /// fields by design, so a saved card whose rows stopped short came back through here as an
    /// unremarkable short card and printed with nothing on the paper saying rows were missing.
    /// The screen it was saved from said so; the paper made from that very document did not.
    ///
    /// A card saved before this field existed simply has no block, which is indistinguishable
    /// from an untruncated card and is the best that can be said of it.
    #[serde(default)]
    pub truncation: Option<StoredCardTruncationV1>,
}

/// The truncation block of a stored response (see [`CardTruncationV1`], which this mirrors).
///
/// All three numbers are REQUIRED, unlike the descriptive scalars elsewhere in a stored card:
/// they are printed together as one sentence, and a block missing one of them would print
/// "TRUNCATED at 0 yd" on a field card. A card that was not truncated omits the whole block,
/// which is the only shape that means "nothing to say".
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Copy, PartialEq, Deserialize)]
pub struct StoredCardTruncationV1 {
    pub requested_end: f64,
    pub last_row: f64,
    pub reach: f64,
}

#[cfg(feature = "pdf")]
impl From<StoredCardTruncationV1> for CardTruncationV1 {
    fn from(stored: StoredCardTruncationV1) -> Self {
        Self {
            requested_end: stored.requested_end,
            last_row: stored.last_row,
            reach: stored.reach,
        }
    }
}

/// The `units` block of a stored response (see [`CardUnitsBlockV1`], which this mirrors).
/// Unknown fields are ignored, for the reason [`StoredCardResponseV1`] gives.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Default, PartialEq, Deserialize)]
pub struct StoredCardUnitsBlockV1 {
    /// `"yd"` / `"m"` — the unit the stored `range` values are in.
    pub distance: String,
    /// Drop column label ("MIL"/"MOA"/"SMOA"/"IPHY"/"CLICKS").
    pub elevation_adjustment: String,
    /// Wind and Lead column label; may differ from the elevation label (MBA-1410).
    pub windage_adjustment: String,
    #[serde(default)]
    pub velocity: Option<String>,
    #[serde(default)]
    pub energy: Option<String>,
    #[serde(default)]
    pub drop: Option<String>,
    #[serde(default)]
    pub wind_speed: Option<String>,
}

/// One stored row (see [`CardRowV1`], which this mirrors). Only `range` is required; a
/// column the stored card did not carry stays `None` and prints as an em-dash rather than a
/// fabricated zero. Unknown fields are ignored, for the reason [`StoredCardResponseV1`]
/// gives.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone, Default, PartialEq, Deserialize)]
pub struct StoredCardRowV1 {
    pub range: f64,
    #[serde(default)]
    pub drop_linear: Option<f64>,
    #[serde(default)]
    pub drop_adj: Option<f64>,
    #[serde(default)]
    pub come_up: Option<f64>,
    #[serde(default)]
    pub wind_linear: Option<f64>,
    #[serde(default)]
    pub wind_adj: Option<f64>,
    #[serde(default)]
    pub velocity: Option<f64>,
    #[serde(default)]
    pub energy: Option<f64>,
    /// Time of flight, seconds. The Lead column of a reprint is derived from this — pure
    /// arithmetic on a number the stored card already carries, not a new trajectory. A row
    /// without it prints an em-dash for Lead.
    #[serde(default)]
    pub time: Option<f64>,
    #[serde(default)]
    pub wind_columns: Vec<f64>,
}

/// A rendered PDF dope card: the document plus the facts a transport needs to describe it
/// without re-parsing the bytes.
#[cfg(feature = "pdf")]
#[derive(Debug, Clone)]
pub struct PdfCardV1 {
    pub pdf_bytes: Vec<u8>,
    /// Pages in `pdf_bytes`, from [`crate::pdf_dope_card::dope_card_page_count`] — the
    /// same function the generator paginated by, not a second estimate of it.
    pub page_count: usize,
    /// Rows printed. On the solve path this is the on-screen `card.range_table` row count;
    /// on the reprint path it is the stored row count, so a caller can confirm every stored
    /// row reached the paper.
    pub row_count: usize,
    /// Solved here, or printed from caller-supplied rows.
    pub source: PdfRowSource,
    /// Characters of the header title (`pdf.title`) the card's font could not draw —
    /// distinct, in order of first use, and empty when the title printed in full.
    ///
    /// Not an error: the document is complete and every ROW printed. But a title is how a
    /// shooter tells one card from another, and Liberation Sans covers no CJK, Arabic,
    /// Hebrew, Thai or emoji, so a name in any of those used to vanish from the paper with
    /// `ok: true` and nothing said. Each such character prints as
    /// [`crate::pdf_dope_card::UNPRINTABLE_SUBSTITUTE`] and is named here, so a caller that
    /// accepts any non-empty card name (both apps do) can warn at export time instead of
    /// handing over an unidentifiable card.
    pub unprintable_title_chars: String,
    /// Present ONLY when the printed rows stop short of the card that was asked for — the
    /// same block [`CardResponseV1::truncation`] carries, for the same reason (MBA-1477).
    ///
    /// Both paths fill it: a SOLVE takes the truncation of the very `card.range_table` rows
    /// it printed, and a REPRINT takes the one the stored card was saved with. Until this
    /// existed the PDF was the only card surface with no truncation signal at all — not the
    /// stderr warning a CLI reader gets, not the `truncated` block in CLI JSON, not
    /// `CardResponseV1::truncation` on the bridge — so a caller could not even tell that the
    /// document it was about to hand a shooter was short. It is also printed in the
    /// document's own footer, which is the half that survives being carried to a range.
    pub truncation: Option<CardTruncationV1>,
}

/// Resolve the effective table font scale from the mutually exclusive `font_scale` /
/// `font_preset` options.
#[cfg(feature = "pdf")]
fn resolve_font_scale(opts: &PdfCardOptionsV1) -> Result<f32, CardServiceError> {
    use crate::pdf_dope_card::{FontSizePreset, FONT_SCALE_RANGE};

    match (opts.font_scale, opts.font_preset.as_deref()) {
        (Some(_), Some(_)) => Err(CardServiceError::InvalidRequest(
            "pdf.font_scale and pdf.font_preset are mutually exclusive; supply one".into(),
        )),
        (Some(scale), None) => {
            let scale = scale as f32;
            if !scale.is_finite() || !FONT_SCALE_RANGE.contains(&scale) {
                return Err(CardServiceError::InvalidRequest(format!(
                    "pdf.font_scale must be finite and within {}..={}, got {}",
                    FONT_SCALE_RANGE.start(),
                    FONT_SCALE_RANGE.end(),
                    opts.font_scale.unwrap_or(f64::NAN)
                )));
            }
            Ok(scale)
        }
        (None, Some(preset)) => FontSizePreset::from_str(preset).map(|p| p.scale()).ok_or_else(|| {
            CardServiceError::InvalidRequest(format!(
                "pdf.font_preset '{preset}' is not one of small, medium, large"
            ))
        }),
        (None, None) => Ok(1.0),
    }
}

/// The rows to print plus everything the header/footer states about where they came from.
/// Filled by exactly one of the two paths in [`pdf_card_v1`], which then share one
/// renderer, so a reprint and a solve cannot be laid out or labelled differently.
#[cfg(feature = "pdf")]
struct RowsToPrint {
    rows: Vec<crate::card::CardRow>,
    source: PdfRowSource,
    /// Footer `BC:` — the BC these rows were computed with.
    bc: f64,
    /// Footer `Engine:` — empty prints nothing.
    engine_version: String,
    /// Footer `Table:` — empty prints nothing.
    table_version: String,
    /// Drop column label.
    elevation_unit_label: String,
    /// Wind and Lead column label.
    windage_unit_label: String,
    /// Footer truncation notice — `None` on a card that reached every row it asked for.
    /// Both paths fill it, so the printed notice cannot depend on where the rows came from.
    truncation: Option<CardTruncationV1>,
}

/// Turn a stored response's rows into printable rows, deriving only the Lead column.
///
/// The Drop and Wind cells are the stored values, untouched. Lead is `target_speed x stored
/// ToF` held on the windage axis — the same [`crate::lead_from_tof`] arithmetic
/// `range_table_rows` performs, applied to the time of flight the stored row already
/// carries. No trajectory, no zero solve, no correction table.
#[cfg(feature = "pdf")]
fn stored_rows_to_print(
    stored: &StoredCardResponseV1,
    req: &CardRequestV1,
    r: &Resolved,
    target_speed: Option<f64>,
) -> Vec<crate::card::CardRow> {
    use crate::card::CardRow;

    let lead_speed_mps = target_speed.map(|speed| r.u.wind_to_metric(speed));
    stored
        .rows
        .iter()
        .map(|row| CardRow {
            range: row.range,
            drop_linear: row.drop_linear,
            drop_adj: row.drop_adj,
            come_up: row.come_up,
            wind_linear: row.wind_linear,
            wind_adj: row.wind_adj,
            velocity: row.velocity,
            energy: row.energy,
            time: row.time,
            lead_adj: lead_speed_mps.zip(row.time).map(|(speed_mps, tof_s)| {
                let range_m = r.u.distance_to_metric(row.range);
                let lead_display =
                    r.u.distance_from_metric(crate::lead_from_tof(speed_mps, 90.0, tof_s, range_m).lead_m);
                // Bias-free (it composes on top of the wind dial, which carries the
                // zero-set bias) but still divided by the windage tracking CF — the exact
                // treatment `range_table_rows` and the CLI's Lead column give it.
                windage_adjustment_display(
                    lead_display,
                    row.range,
                    r.windage_unit,
                    r.windage_click,
                    0.0,
                    req.windage_cf,
                )
                .value
            }),
            wind_columns: row.wind_columns.clone(),
        })
        .collect()
}

/// Reject a stored card that is not the range table it must be, cell values included.
#[cfg(feature = "pdf")]
fn validate_stored_card(
    stored: &StoredCardV1,
    req: &CardRequestV1,
    r: &Resolved,
) -> Result<(), CardServiceError> {
    let card = &stored.card;
    if card.kind != PDF_CARD_KIND {
        return Err(CardServiceError::InvalidRequest(format!(
            "card.pdf prints a {PDF_CARD_KIND} card; the stored card's kind is '{}'. Its \
             columns are not a range table's, so it is refused rather than reprinted as one",
            card.kind
        )));
    }
    // The same "this is not a range table" test the REQUEST gets (see the `wind_speeds` /
    // `wind_angles_deg` loop in `pdf_card_v1`), field for field. The two halves used to
    // disagree — `wind_angles_deg` was refused on the request and accepted here — and the
    // stored half is the one that has to hold when a future kind starts emitting it.
    for (field, present) in [
        ("wind_speeds", !card.wind_speeds.is_empty()),
        ("wind_angles_deg", !card.wind_angles_deg.is_empty()),
        ("extra_angle_rows", !card.extra_angle_rows.is_empty()),
    ] {
        if present {
            return Err(CardServiceError::InvalidRequest(format!(
                "the stored card carries {field}, a wind card's defining field, so it is not the \
                 {PDF_CARD_KIND} response it claims to be"
            )));
        }
    }
    if card.rows.is_empty() {
        return Err(CardServiceError::InvalidRequest(
            "the stored card has no rows; an empty dope card is not a document".into(),
        ));
    }

    // A row means something only in a unit. If the stored labels and the request's own axes
    // disagree, the two are not the same card — printing anyway would put one unit's numbers
    // under another unit's heading, which is precisely the failure the shared request shape
    // exists to prevent.
    let elevation = adjustment_unit_label(req.adjustment_unit);
    let windage = adjustment_unit_label(r.windage_unit);
    let distance = r.u.distance_label();
    for (field, stored_label, request_label) in [
        ("units.distance", card.units.distance.as_str(), distance),
        (
            "units.elevation_adjustment",
            card.units.elevation_adjustment.as_str(),
            elevation.as_str(),
        ),
        (
            "units.windage_adjustment",
            card.units.windage_adjustment.as_str(),
            windage.as_str(),
        ),
    ] {
        if stored_label != request_label {
            return Err(CardServiceError::InvalidRequest(format!(
                "the stored card's {field} is '{stored_label}' but this request's own axes say \
                 '{request_label}' — the stored rows and the request are not the same card"
            )));
        }
    }

    // Same rule as the cells below: a non-finite figure would print "NaN"/"inf" — here in a
    // sentence telling a shooter where their card stops, which is worse than in a column.
    if let Some(t) = &card.truncation {
        for (field, value) in [
            ("requested_end", t.requested_end),
            ("last_row", t.last_row),
            ("reach", t.reach),
        ] {
            if !value.is_finite() {
                return Err(CardServiceError::InvalidRequest(format!(
                    "the stored card's truncation.{field} is not a finite number"
                )));
            }
        }
    }

    // A non-finite cell would print "NaN"/"inf" on a field card.
    for (index, row) in card.rows.iter().enumerate() {
        for (field, value) in [
            ("range", Some(row.range)),
            ("drop_adj", row.drop_adj),
            ("wind_adj", row.wind_adj),
            ("time", row.time),
        ] {
            if value.is_some_and(|v| !v.is_finite()) {
                return Err(CardServiceError::InvalidRequest(format!(
                    "the stored card's rows[{index}].{field} is not a finite number"
                )));
            }
        }
    }
    Ok(())
}

/// Render the printable PDF dope card for a card request.
///
/// `stored` decides where the numbers come from, and it is the whole point of this surface:
///
/// * `Some(card)` — REPRINT. The rows are the caller's: the stored `card.range_table`
///   response for this same request. Nothing is solved, `bc5d_table_path` is never opened,
///   and the footer's BC, engine version and table version are the stored card's. A saved
///   card therefore reprints identically after an engine bump, after the correction table at
///   the stored path is overwritten in place, and even after that file is deleted.
/// * `None` — SOLVE. `range_table_rows` runs, i.e. literally the rows [`range_table_v1`]
///   would return for this request, and the footer states THIS build's version. Unchanged
///   behaviour for a caller that has no stored rows.
///
/// Either way this prints a [`PDF_CARD_KIND`] card and nothing else: a request carrying a
/// wind card's `wind_speeds`/`wind_angles_deg`, or a stored card of another kind, is refused.
/// Both paths map their rows onto the CLI's Range/Drop/Wind/Lead dope card, plus the Lead
/// column that `CardRequestV1::pdf`'s `target_speed` asks for. The Range column is
/// denominated in the stored/requested `distance_unit`, unlike `trajectory -o pdf`, whose
/// dope card is always yards.
///
/// The header/footer block is always imperial, matching both CLI PDF call sites: whatever the
/// request's velocity/temperature/pressure/altitude/wind/weight dimensions are denominated in,
/// they are converted for display only. The `Solver:` label reports this build (`online` with
/// the `online` feature, otherwise `offline`) and the timestamp is generation time, so neither
/// is caller-settable — and neither is a number a shooter dials.
#[cfg(feature = "pdf")]
pub fn pdf_card_v1(
    req: &CardRequestV1,
    stored: Option<&StoredCardV1>,
) -> Result<PdfCardV1, CardServiceError> {
    use crate::pdf_dope_card::{
        calculate_density_altitude, dope_card_page_count, generate_dope_card_pdf, DopeCardConfig,
        RangeUnit,
    };

    let opts = req.pdf.clone().unwrap_or_default();
    // Validate presentation options BEFORE spending a zero solve and two trajectories on a
    // request that cannot be rendered anyway.
    let font_scale = resolve_font_scale(&opts)?;
    // A non-finite crossing speed would print "inf"/"NaN" in every Lead cell rather than
    // failing. (A NEGATIVE speed is accepted, matching the CLI: it reads as the mover
    // crossing the other way and simply flips the sign of the hold.)
    if opts.target_speed.is_some_and(|speed| !speed.is_finite()) {
        return Err(CardServiceError::InvalidRequest(
            "pdf.target_speed must be finite".into(),
        ));
    }

    // A wind card's defining fields cannot be honoured by a range-table PDF. Refuse them
    // instead of returning a document whose Wind column contradicts the screen.
    for (field, present) in [
        ("wind_speeds", !req.wind_speeds.is_empty()),
        ("wind_angles_deg", !req.wind_angles_deg.is_empty()),
    ] {
        if present {
            return Err(CardServiceError::InvalidRequest(format!(
                "card.pdf prints a {PDF_CARD_KIND} card; this request carries {field}, a wind \
                 card's defining field, which a range-table PDF cannot show and would silently \
                 ignore. There is no wind_card or come_ups PDF in this build"
            )));
        }
    }

    let to_print = match stored {
        Some(stored) => {
            // Axes and click graduations only: no BC schedule, so no correction table is
            // opened for a card whose numbers are already decided.
            let r = resolve_axes_only(req)?;
            validate_stored_card(stored, req, &r)?;
            RowsToPrint {
                rows: stored_rows_to_print(&stored.card, req, &r, opts.target_speed),
                source: PdfRowSource::StoredRows,
                bc: stored.card.bc_for_solve.unwrap_or(req.ballistic_coefficient),
                engine_version: stored.engine_version.clone().unwrap_or_default(),
                table_version: stored.bc5d_table_version.clone().unwrap_or_default(),
                elevation_unit_label: stored.card.units.elevation_adjustment.clone(),
                windage_unit_label: stored.card.units.windage_adjustment.clone(),
                // The truncation the card was SAVED with. Nothing is re-derived from the
                // stored row set here — a reprint prints the card it was handed, and that
                // includes what the card already said about itself.
                truncation: stored.card.truncation.map(CardTruncationV1::from),
            }
        }
        None => {
            let (card, lead_adj, bc_for_solve) = range_table_rows(req, opts.target_speed)?;
            if card.rows.is_empty() {
                // `generate_dope_card_pdf` refuses an empty row set, and rightly: an empty
                // dope card is not a document. Name the cause the caller can act on instead.
                return Err(CardServiceError::InvalidRequest(format!(
                    "no card rows in {}..={} — the trajectory does not reach this range domain, \
                     or step is coarser than the samples",
                    req.start, req.end
                )));
            }
            RowsToPrint {
                rows: card
                    .rows
                    .iter()
                    .zip(&lead_adj)
                    .map(|(row, lead)| crate::card::CardRow {
                        range: row.range,
                        drop_linear: row.drop_linear,
                        drop_adj: row.drop_adj,
                        come_up: row.come_up,
                        wind_linear: row.wind_linear,
                        wind_adj: row.wind_adj,
                        velocity: row.velocity,
                        energy: row.energy,
                        time: row.time,
                        lead_adj: *lead,
                        wind_columns: row.wind_columns.clone(),
                    })
                    .collect(),
                source: PdfRowSource::Solve,
                bc: bc_for_solve,
                // No table version is knowable from a path, so none is claimed; the engine
                // version is this build's, which is what produced these rows.
                engine_version: env!("CARGO_PKG_VERSION").to_string(),
                table_version: String::new(),
                // The labels `card.range_table`'s `units` block reports for this request:
                // Drop in the elevation unit, Wind AND Lead in the (possibly different,
                // MBA-1410) windage unit.
                elevation_unit_label: card.units.elevation_adjustment.clone(),
                windage_unit_label: card.units.windage_adjustment.clone(),
                // The truncation of the very rows above, not a second determination of it:
                // `range_table_rows` IS `card.range_table`, so the paper and the screen
                // truncate identically or not at all.
                truncation: card.truncation,
            }
        }
    };

    // Refuse on the row/page count now that the rows are known — before a document is built.
    // The byte cap downstream can only measure a document it already paid for.
    let page_count = dope_card_page_count(to_print.rows.len(), font_scale);
    if to_print.rows.len() > MAX_PDF_ROWS || page_count > MAX_PDF_PAGES {
        return Err(CardServiceError::TooLarge(format!(
            "this card has too many rows to print: {} rows, {page_count} pages (the limits are \
             {MAX_PDF_ROWS} rows and {MAX_PDF_PAGES} pages)",
            to_print.rows.len()
        )));
    }

    // The request's own axes, for the header. Resolved here rather than taken from above
    // because neither branch hands a `Resolved` back — the reprint's is scoped to its arm and
    // the solve's lives inside `range_table_rows`. Resolving twice is safe precisely because
    // `Units::resolve` is a pure function of the request: the header cannot land on axes the
    // rows were not computed in.
    let u = Units::resolve(req);
    // The atmosphere defaults `resolve()` applies, restated for the header so the printed
    // conditions are the ones the solve used rather than blanks.
    let temperature = req.temperature.unwrap_or_else(|| u.default_temperature());
    let pressure = req.pressure.unwrap_or_else(|| u.default_pressure());
    // inHg <-> hPa via the CLI dope card's own factor (main.rs uses 33.8639 on both PDF
    // call sites), NOT pdf_dope_card::INHG_TO_HPA — matching the shipped header exactly.
    let pressure_inhg = u.pressure_to_inhg(pressure);
    let pressure_hpa = u.pressure_to_hpa(pressure);
    let temperature_f = u.temperature_to_fahrenheit(temperature);
    // NOTE: `CardRequestV1::altitude` defaults to METRES whatever the `units` preset says —
    // the one dimension whose fallback is not the preset's, because it never was (see
    // `CardRequestV1::altitude_unit`). The header reports the altitude the solve actually
    // used, so it converts out of whatever `altitude_unit` settled on.
    let altitude_ft = u.altitude_to_feet(req.altitude);

    let config = DopeCardConfig {
        rifle_name: opts.title.clone().unwrap_or_else(|| "Dope Card".to_string()),
        location: opts.location.clone().unwrap_or_default(),
        density_altitude_ft: calculate_density_altitude(altitude_ft, pressure_inhg, temperature_f),
        pressure_inhg,
        pressure_hpa,
        temperature_f,
        altitude_ft,
        wind_speed_mph: u.wind_to_mph(req.wind_speed),
        // Absent target speed prints 0 here (there is no lead to state) while the Lead
        // column itself stays em-dashed; an explicit 0.0 prints the same 0 and zeroes.
        target_speed_mph: match opts.target_speed {
            Some(speed) => u.wind_to_mph(speed),
            None => 0.0,
        },
        solver_mode: if cfg!(feature = "online") { "online".to_string() } else { "offline".to_string() },
        powder: opts.powder.clone().unwrap_or_default(),
        bullet: opts.bullet.clone().unwrap_or_default(),
        weight_gr: u.mass_to_grains_cli(req.mass),
        // The BC these rows came from: the stored card's own, or this solve's (which is the
        // muzzle-corrected value when a BC5D table applied one).
        bc: to_print.bc,
        drag_model: DragModel::from(req.drag_model).to_string(),
        velocity_fps: u.velocity_to_fps_pdf(req.muzzle_velocity),
        font_scale,
        bold_data: opts.bold_data,
        // The card's own axes: Drop in the elevation unit, Wind AND Lead in the (possibly
        // different, MBA-1410) windage unit — from the document that owns the rows.
        elevation_unit_label: to_print.elevation_unit_label.clone(),
        windage_unit_label: to_print.windage_unit_label.clone(),
        // Provenance on the paper, so a printed card and a screen can be reconciled later.
        engine_version: to_print.engine_version.clone(),
        table_version: to_print.table_version.clone(),
        // MBA-1477: and the truncation on the paper too, in the one wording every card
        // surface uses, denominated in the card's own distance unit (which
        // `validate_stored_card` has already forced to agree with the request's).
        truncation_note: to_print
            .truncation
            .map(|t| CardTruncation::from(t).printed_note(u.distance_label()))
            .unwrap_or_default(),
    };

    let range_unit = match u.distance {
        CardDistanceUnit::Yards => RangeUnit::Yards,
        CardDistanceUnit::Meters => RangeUnit::Meters,
    };
    // Read BEFORE the document is built, off the same string the header will draw, so the
    // report describes this card's own title rather than a truncation of it.
    let unprintable_title_chars = crate::pdf_dope_card::unprintable_chars(&config.rifle_name);
    let pdf_bytes = generate_dope_card_pdf(&config, &to_print.rows, range_unit)
        .map_err(|e| CardServiceError::Pdf(e.to_string()))?;

    Ok(PdfCardV1 {
        page_count,
        row_count: to_print.rows.len(),
        source: to_print.source,
        pdf_bytes,
        unprintable_title_chars,
        truncation: to_print.truncation,
    })
}