@willcgage/module-schematic 0.54.0 → 0.56.0

package/dist/index.d.cts

@@ -1035,28 +1035,35 @@ interface ModuleFeatures {
 /** N-scale track gauge, inches (9 mm). */
 declare const RAIL_GAUGE_INCHES = 0.354;
 /**
- * ⚠️ **REFUTED AS A GENERAL RULE — LAST RESORT ONLY.** Points→frog for a #1
- * frog, i.e. a turnout's LEAD if lead were proportional to frog number. From an
- * Atlas code 55 #7 measuring 3⅜″ (Steve Branton, #173) ⇒ 3.375 / 7.
+ * ⚠️⚠️ **DEAD — REFUTED *AND* BASELESS. NOTHING SHOULD REACH THIS.** Kept only
+ * so a library with no measured leads at all still returns a number, and as a
+ * record of how the mistake was made.
  *
- * It is NOT proportional, and the error CHANGES SIGN. Measured lead ÷ N:
- * **#5 = 0.600, #7 = 0.482, #10 = 0.419** — it falls monotonically. This rule
- * reads **20% SHORT at N=5** (2.41″ vs a measured 3.00″) and **13% LONG at
- * N=10** (4.82″ vs 4³⁄₁₆″). There is no safe direction to be wrong in.
+ * It was points→frog for a #1 frog: a turnout's LEAD if lead were proportional
+ * to frog number, from an Atlas #7 measuring 3⅜″ (Steve Branton, #173) ⇒
+ * 3.375 / 7.
  *
- * Prefer {@link leadInchesForSize}, which returns a real part's measurement when
- * one exists and only falls back here for sizes nothing in the library covers.
- * ⚠️ The remaining fallback user is the #2.5 wye at 1.21″; since per-frog RISES
- * as N falls, that is very likely SHORT. Treat it as a placeholder.
+ * TWO independent things are wrong with it:
+ * 1. **Lead is not proportional to N.** Measured lead ÷ N: #5 = 0.600,
+ *    #7 = 0.513, #10 = 0.494. It reads ~20% SHORT at N=5.
+ * 2. **Its founding measurement was superseded.** Will Gage re-measured the same
+ *    part at 3¹⁹⁄₃₂″, ⁷⁄₃₂″ longer, confirmed against his own points and frog
+ *    positions. So 0.482 is not even the right constant for the wrong model.
+ *
+ * {@link leadInchesForSize} now INTERPOLATES across the measured parts instead.
+ * A single measurement can only ever be scaled — which is this rule — so the
+ * lesson is that one data point cannot support a shape, only a value.
  *
  * A constant-switch-angle model was also proposed here and is likewise REFUTED:
- * inverting {@link turnoutClosure} gives α = 0.036 (#5), 0.067 (#7), 0.069
- * (#10). It was floated when only the #7 and #10 were known, which agreed to
- * 3%; the #5 killed it. Kept as a warning — two agreeing parts are not a law,
- * and this library has now watched three separate rules die that way.
+ * inverting {@link turnoutClosure} gives α = 0.036 (#5), 0.067 (#7), 0.043
+ * (#10). Kept as a warning — it was floated when the #7 and #10 agreed to 3%,
+ * on a #10 figure that later turned out to be measured to the wrong landmark.
+ * TWO AGREEING PARTS ARE NOT A LAW, and neither is a coincidence you haven't
+ * checked the provenance of.
  *
- * The pattern that DOES fit both fully-measured parts is on {@link
- * ATLAS_CODE55_N}: the frog sits at a fixed 4.75″ and the points move.
+ * The pattern that currently fits is on {@link ATLAS_CODE55_N}: the moulding
+ * runs past the frog far enough to gain a constant 0.25″ of separation. Note it
+ * predicts the FROG POSITION, not the lead — this constant has no successor.
  */
 declare const TURNOUT_LEAD_INCHES_PER_FROG = 0.482;
 /** Where a dimension came from. `derived` = we computed it from another part or
@@ -1105,6 +1112,13 @@ interface TrackPart {
      * inside its moulding; the rest of that end is plain approach track. Needed to
      * fit a part into a space, and by any renderer drawing the real outline. */
     pointsOffset?: PartDimension;
+    /** End of the tie strip → the APEX OF THE FROG V, measured from the same end
+     * as {@link pointsOffset}. ⚠️ The apex, NOT the end of the frog casting — on a
+     * shallow #10 those are ¾″ apart, which is how this library recorded a wrong
+     * #10 lead through two releases. `lead` is the difference of the two, so
+     * prefer these when both are known: they're what someone can actually put a
+     * rule against. */
+    frogOffset?: PartDimension;
     /** End-to-end length of the part. */
     overallLength?: PartDimension;
     /** Diverging route radius (straight turnouts). */
@@ -1125,32 +1139,47 @@ declare const CODE55_RAIL_HEIGHT_INCHES = 0.055;
  * (3⅜″ points→frog, #173) and Will Gage's #5/#7/#10 overall lengths plus the
  * #10's lead and points offset. The #5's lead and the wye's are still derived.
  *
- * ⭐⭐ **THE FROG SITS AT 4.75″ FROM THE TIE END — ON BOTH MEASURED PARTS.**
+ * MEASURED GEOMETRY (all Will Gage's, off physical parts, frog taken at the V):
  *
- *     part   points    FROG     overall    lead
- *     #5     1¾″       4.75″    6.00″      3.0000″
- *     #10    ⁹⁄₁₆″     4.75″    8.00″      4.1875″
+ *     part   points   frog V    overall   lead       past frog   ÷N
+ *     #5     1¾″      4.75″     6.00″     3.00000″   1.25″       0.2500
+ *     #7     ⁵⁄₈″     4⁷⁄₃₂″    6.00″     3.59375″   1.78125″    0.2545
+ *     #10    ⁹⁄₁₆″    5.50″     8.00″     4.93750″   2.50″       0.2500
  *
- * Atlas appear to FIX where the frog casting sits in the moulding and move the
- * POINTS to change the frog number. That single fact explains every failed rule
- * below at once: lead was never a property of N — it is just
- * `4.75 − wherever the points landed`.
+ * Every one of these is Will Gage's, off a physical part, frog taken at the V.
+ * ⚠️ The #7 lead SUPERSEDES Steve Branton's 3⅜″ (#173) — the library's founding
+ * measurement, ⁷⁄₃₂″ short. Will's single-span reading and his two positions
+ * agree exactly, which is why it wins. Steve's number was also the sole basis of
+ * {@link TURNOUT_LEAD_INCHES_PER_FROG}; that constant is now baseless as well as
+ * refuted, and {@link leadInchesForSize} interpolates across this table instead.
  *
- * ⚠️ TESTABLE, AND CHEAP: Steve's #7 lead of 3⅜″ was measured independently, by
- * someone else, on a different part. If the frog really is fixed at 4.75″, that
- * part's POINTS must sit at 4.75 − 3.375 = **1⅜″**. Measure a physical 2052 and
- * this model is confirmed or dead. Nobody has measured the #7's points offset.
+ * ⭐ **THE PART RUNS PAST THE FROG FAR ENOUGH TO GAIN 0.25″ OF SEPARATION.**
+ * Past-frog run ÷ N is 0.2500 / 0.2545 / 0.2500 — the #5 and #10 EXACT, the #7
+ * within ¹⁄₃₂″ of the tie end (a 4¼″ frog would make it exact too, one tape
+ * division away). Provenance note: the measurer was told that prediction BEFORE
+ * re-reading and did not return 4¼″, so this is not a confirmation artefact. At the frog
+ * the routes are one gauge apart; at the end of the moulding they are
+ * 0.354 + 0.25 ≈ 0.6″ apart. So Atlas size the moulding for a constant
+ * CLEARANCE, not a constant length — which is what a manufacturer would
+ * sensibly do, and why overall length looks arbitrary until you divide by N.
+ * ⚠️ It predicts the FROG POSITION from the overall length. It does NOT predict
+ * the lead — the points offset is independent and must still be measured.
+ * ⚠️ HYPOTHESIS. It is the FOURTH rule fitted to this data; the first three all
+ * died. Two exact hits and a near miss is not a law.
  *
  * NOTHING HERE IS PROPORTIONAL TO FROG NUMBER. Three rules assumed otherwise;
  * all three were tested against a physical part and FAILED:
- * - ❌ **Lead ∝ N.** Measured lead ÷ N: #5 = 0.600, #7 = 0.482, #10 = 0.419 —
- *   falls monotonically. {@link TURNOUT_LEAD_INCHES_PER_FROG} reads 13% LONG at
- *   N=10 and 20% SHORT at N=5, so its error even changes sign.
+ * - ❌ **Lead ∝ N.** Measured lead ÷ N: #5 = 0.600, #7 = 0.513, #10 = 0.494.
+ *   {@link TURNOUT_LEAD_INCHES_PER_FROG} reads ~20% SHORT at N=5 — and its
+ *   founding measurement has since been superseded, so it is baseless too.
  * - ❌ **Constant switch angle.** Inverting {@link turnoutClosure} gives α =
- *   0.036 (#5), 0.067 (#7), 0.069 (#10). The #7 and #10 agreed to 3% and the #5
- *   killed it. Two agreeing parts are not a law.
+ *   0.036 (#5), 0.055 (#7), 0.043 (#10). Not constant.
  * - ❌ **Overall length ∝ N.** #5 = 6.00″ and #7 = 6.00″ — same moulding, two
  *   frog numbers.
+ * - ❌ **Frog fixed at 4.75″** (shipped in 0.54.0, retracted in 0.55.0). It
+ *   rested on the #5 and #10 both reading 4.75″, but the #10's was the END OF
+ *   THE FROG CASTING, not the V. Measured consistently the frogs are at 4.75″,
+ *   4³⁄₁₆″ and 5.5″ — all different.
  *
  * The definition is the reason: #N fixes the DIVERGENCE RATE AT THE FROG (1
  * across per N along) and says NOTHING about how much tie strip the manufacturer
@@ -1165,9 +1194,19 @@ declare function trackPart(id: string, library?: TrackPart[]): TrackPart | null;
 /** The closest built-in turnout for a frog number — what a bare `size` maps to
  * when a turnout names no part. Exact match wins; otherwise the nearest frog. */
 declare function turnoutPartForSize(size: number, library?: TrackPart[]): TrackPart | null;
-/** The lead to draw a turnout of this size with — a real part's measurement when
- * we have one, else the per-frog rule. Keeps `frogLegOf` honest without it
- * needing to know the library exists. */
+/**
+ * The lead to draw a turnout of this size with. Keeps `frogLegOf` honest without
+ * it needing to know the library exists.
+ *
+ * 1. An exact part match wins — a #7's measurement says nothing about a #4.
+ * 2. Otherwise INTERPOLATE piecewise-linearly across the measured parts. This
+ *    replaced `size × TURNOUT_LEAD_INCHES_PER_FROG`, which was refuted (its
+ *    error changes sign across the measured range) and is now baseless besides,
+ *    its founding #7 measurement having been superseded.
+ * 3. ⚠️ Outside the measured range the end segments EXTRAPOLATE. Measured leads
+ *    span N = 5…10, so a #4 or #12 is a projection, not a reading, and the #2.5
+ *    wye is a long way out. Treat those as placeholders until a part is measured.
+ */
 declare function leadInchesForSize(size: number, library?: TrackPart[]): number;
 /** One drawn piece of a part's geometry, in the part's own local frame. */
 type PartSegment = {

package/dist/index.d.ts

@@ -1035,28 +1035,35 @@ interface ModuleFeatures {
 /** N-scale track gauge, inches (9 mm). */
 declare const RAIL_GAUGE_INCHES = 0.354;
 /**
- * ⚠️ **REFUTED AS A GENERAL RULE — LAST RESORT ONLY.** Points→frog for a #1
- * frog, i.e. a turnout's LEAD if lead were proportional to frog number. From an
- * Atlas code 55 #7 measuring 3⅜″ (Steve Branton, #173) ⇒ 3.375 / 7.
+ * ⚠️⚠️ **DEAD — REFUTED *AND* BASELESS. NOTHING SHOULD REACH THIS.** Kept only
+ * so a library with no measured leads at all still returns a number, and as a
+ * record of how the mistake was made.
  *
- * It is NOT proportional, and the error CHANGES SIGN. Measured lead ÷ N:
- * **#5 = 0.600, #7 = 0.482, #10 = 0.419** — it falls monotonically. This rule
- * reads **20% SHORT at N=5** (2.41″ vs a measured 3.00″) and **13% LONG at
- * N=10** (4.82″ vs 4³⁄₁₆″). There is no safe direction to be wrong in.
+ * It was points→frog for a #1 frog: a turnout's LEAD if lead were proportional
+ * to frog number, from an Atlas #7 measuring 3⅜″ (Steve Branton, #173) ⇒
+ * 3.375 / 7.
  *
- * Prefer {@link leadInchesForSize}, which returns a real part's measurement when
- * one exists and only falls back here for sizes nothing in the library covers.
- * ⚠️ The remaining fallback user is the #2.5 wye at 1.21″; since per-frog RISES
- * as N falls, that is very likely SHORT. Treat it as a placeholder.
+ * TWO independent things are wrong with it:
+ * 1. **Lead is not proportional to N.** Measured lead ÷ N: #5 = 0.600,
+ *    #7 = 0.513, #10 = 0.494. It reads ~20% SHORT at N=5.
+ * 2. **Its founding measurement was superseded.** Will Gage re-measured the same
+ *    part at 3¹⁹⁄₃₂″, ⁷⁄₃₂″ longer, confirmed against his own points and frog
+ *    positions. So 0.482 is not even the right constant for the wrong model.
+ *
+ * {@link leadInchesForSize} now INTERPOLATES across the measured parts instead.
+ * A single measurement can only ever be scaled — which is this rule — so the
+ * lesson is that one data point cannot support a shape, only a value.
  *
  * A constant-switch-angle model was also proposed here and is likewise REFUTED:
- * inverting {@link turnoutClosure} gives α = 0.036 (#5), 0.067 (#7), 0.069
- * (#10). It was floated when only the #7 and #10 were known, which agreed to
- * 3%; the #5 killed it. Kept as a warning — two agreeing parts are not a law,
- * and this library has now watched three separate rules die that way.
+ * inverting {@link turnoutClosure} gives α = 0.036 (#5), 0.067 (#7), 0.043
+ * (#10). Kept as a warning — it was floated when the #7 and #10 agreed to 3%,
+ * on a #10 figure that later turned out to be measured to the wrong landmark.
+ * TWO AGREEING PARTS ARE NOT A LAW, and neither is a coincidence you haven't
+ * checked the provenance of.
  *
- * The pattern that DOES fit both fully-measured parts is on {@link
- * ATLAS_CODE55_N}: the frog sits at a fixed 4.75″ and the points move.
+ * The pattern that currently fits is on {@link ATLAS_CODE55_N}: the moulding
+ * runs past the frog far enough to gain a constant 0.25″ of separation. Note it
+ * predicts the FROG POSITION, not the lead — this constant has no successor.
  */
 declare const TURNOUT_LEAD_INCHES_PER_FROG = 0.482;
 /** Where a dimension came from. `derived` = we computed it from another part or
@@ -1105,6 +1112,13 @@ interface TrackPart {
      * inside its moulding; the rest of that end is plain approach track. Needed to
      * fit a part into a space, and by any renderer drawing the real outline. */
     pointsOffset?: PartDimension;
+    /** End of the tie strip → the APEX OF THE FROG V, measured from the same end
+     * as {@link pointsOffset}. ⚠️ The apex, NOT the end of the frog casting — on a
+     * shallow #10 those are ¾″ apart, which is how this library recorded a wrong
+     * #10 lead through two releases. `lead` is the difference of the two, so
+     * prefer these when both are known: they're what someone can actually put a
+     * rule against. */
+    frogOffset?: PartDimension;
     /** End-to-end length of the part. */
     overallLength?: PartDimension;
     /** Diverging route radius (straight turnouts). */
@@ -1125,32 +1139,47 @@ declare const CODE55_RAIL_HEIGHT_INCHES = 0.055;
  * (3⅜″ points→frog, #173) and Will Gage's #5/#7/#10 overall lengths plus the
  * #10's lead and points offset. The #5's lead and the wye's are still derived.
  *
- * ⭐⭐ **THE FROG SITS AT 4.75″ FROM THE TIE END — ON BOTH MEASURED PARTS.**
+ * MEASURED GEOMETRY (all Will Gage's, off physical parts, frog taken at the V):
  *
- *     part   points    FROG     overall    lead
- *     #5     1¾″       4.75″    6.00″      3.0000″
- *     #10    ⁹⁄₁₆″     4.75″    8.00″      4.1875″
+ *     part   points   frog V    overall   lead       past frog   ÷N
+ *     #5     1¾″      4.75″     6.00″     3.00000″   1.25″       0.2500
+ *     #7     ⁵⁄₈″     4⁷⁄₃₂″    6.00″     3.59375″   1.78125″    0.2545
+ *     #10    ⁹⁄₁₆″    5.50″     8.00″     4.93750″   2.50″       0.2500
  *
- * Atlas appear to FIX where the frog casting sits in the moulding and move the
- * POINTS to change the frog number. That single fact explains every failed rule
- * below at once: lead was never a property of N — it is just
- * `4.75 − wherever the points landed`.
+ * Every one of these is Will Gage's, off a physical part, frog taken at the V.
+ * ⚠️ The #7 lead SUPERSEDES Steve Branton's 3⅜″ (#173) — the library's founding
+ * measurement, ⁷⁄₃₂″ short. Will's single-span reading and his two positions
+ * agree exactly, which is why it wins. Steve's number was also the sole basis of
+ * {@link TURNOUT_LEAD_INCHES_PER_FROG}; that constant is now baseless as well as
+ * refuted, and {@link leadInchesForSize} interpolates across this table instead.
  *
- * ⚠️ TESTABLE, AND CHEAP: Steve's #7 lead of 3⅜″ was measured independently, by
- * someone else, on a different part. If the frog really is fixed at 4.75″, that
- * part's POINTS must sit at 4.75 − 3.375 = **1⅜″**. Measure a physical 2052 and
- * this model is confirmed or dead. Nobody has measured the #7's points offset.
+ * ⭐ **THE PART RUNS PAST THE FROG FAR ENOUGH TO GAIN 0.25″ OF SEPARATION.**
+ * Past-frog run ÷ N is 0.2500 / 0.2545 / 0.2500 — the #5 and #10 EXACT, the #7
+ * within ¹⁄₃₂″ of the tie end (a 4¼″ frog would make it exact too, one tape
+ * division away). Provenance note: the measurer was told that prediction BEFORE
+ * re-reading and did not return 4¼″, so this is not a confirmation artefact. At the frog
+ * the routes are one gauge apart; at the end of the moulding they are
+ * 0.354 + 0.25 ≈ 0.6″ apart. So Atlas size the moulding for a constant
+ * CLEARANCE, not a constant length — which is what a manufacturer would
+ * sensibly do, and why overall length looks arbitrary until you divide by N.
+ * ⚠️ It predicts the FROG POSITION from the overall length. It does NOT predict
+ * the lead — the points offset is independent and must still be measured.
+ * ⚠️ HYPOTHESIS. It is the FOURTH rule fitted to this data; the first three all
+ * died. Two exact hits and a near miss is not a law.
  *
  * NOTHING HERE IS PROPORTIONAL TO FROG NUMBER. Three rules assumed otherwise;
  * all three were tested against a physical part and FAILED:
- * - ❌ **Lead ∝ N.** Measured lead ÷ N: #5 = 0.600, #7 = 0.482, #10 = 0.419 —
- *   falls monotonically. {@link TURNOUT_LEAD_INCHES_PER_FROG} reads 13% LONG at
- *   N=10 and 20% SHORT at N=5, so its error even changes sign.
+ * - ❌ **Lead ∝ N.** Measured lead ÷ N: #5 = 0.600, #7 = 0.513, #10 = 0.494.
+ *   {@link TURNOUT_LEAD_INCHES_PER_FROG} reads ~20% SHORT at N=5 — and its
+ *   founding measurement has since been superseded, so it is baseless too.
  * - ❌ **Constant switch angle.** Inverting {@link turnoutClosure} gives α =
- *   0.036 (#5), 0.067 (#7), 0.069 (#10). The #7 and #10 agreed to 3% and the #5
- *   killed it. Two agreeing parts are not a law.
+ *   0.036 (#5), 0.055 (#7), 0.043 (#10). Not constant.
  * - ❌ **Overall length ∝ N.** #5 = 6.00″ and #7 = 6.00″ — same moulding, two
  *   frog numbers.
+ * - ❌ **Frog fixed at 4.75″** (shipped in 0.54.0, retracted in 0.55.0). It
+ *   rested on the #5 and #10 both reading 4.75″, but the #10's was the END OF
+ *   THE FROG CASTING, not the V. Measured consistently the frogs are at 4.75″,
+ *   4³⁄₁₆″ and 5.5″ — all different.
  *
  * The definition is the reason: #N fixes the DIVERGENCE RATE AT THE FROG (1
  * across per N along) and says NOTHING about how much tie strip the manufacturer
@@ -1165,9 +1194,19 @@ declare function trackPart(id: string, library?: TrackPart[]): TrackPart | null;
 /** The closest built-in turnout for a frog number — what a bare `size` maps to
  * when a turnout names no part. Exact match wins; otherwise the nearest frog. */
 declare function turnoutPartForSize(size: number, library?: TrackPart[]): TrackPart | null;
-/** The lead to draw a turnout of this size with — a real part's measurement when
- * we have one, else the per-frog rule. Keeps `frogLegOf` honest without it
- * needing to know the library exists. */
+/**
+ * The lead to draw a turnout of this size with. Keeps `frogLegOf` honest without
+ * it needing to know the library exists.
+ *
+ * 1. An exact part match wins — a #7's measurement says nothing about a #4.
+ * 2. Otherwise INTERPOLATE piecewise-linearly across the measured parts. This
+ *    replaced `size × TURNOUT_LEAD_INCHES_PER_FROG`, which was refuted (its
+ *    error changes sign across the measured range) and is now baseless besides,
+ *    its founding #7 measurement having been superseded.
+ * 3. ⚠️ Outside the measured range the end segments EXTRAPOLATE. Measured leads
+ *    span N = 5…10, so a #4 or #12 is a projection, not a reading, and the #2.5
+ *    wye is a long way out. Treat those as placeholders until a part is measured.
+ */
 declare function leadInchesForSize(size: number, library?: TrackPart[]): number;
 /** One drawn piece of a part's geometry, in the part's own local frame. */
 type PartSegment = {

package/dist/index.js

@@ -1111,6 +1111,11 @@ var ATLAS_CODE55_N = [
       source: "measured",
       note: "Will Gage, physical Atlas 2050 (#5 LH) \u2014 end of tie strip to point tips"
     },
+    frogOffset: {
+      inches: 4.75,
+      source: "measured",
+      note: "Will Gage, physical Atlas 2050 (#5 LH) \u2014 tie end to the apex of the V"
+    },
     overallLength: {
       inches: 6,
       source: "measured",
@@ -1132,9 +1137,19 @@ var ATLAS_CODE55_N = [
     partNumbers: { left: "2052", right: "2053" },
     frogNumber: 7,
     lead: {
-      inches: 3.375,
+      inches: 3.59375,
+      source: "measured",
+      note: "Will Gage, physical Atlas 2052 (#7), 3\xB9\u2079\u2044\u2083\u2082\u2033 points\u2192frog as a single span \u2014 and it matches his two positions exactly (4\u2077\u2044\u2083\u2082 \u2212 \u2075\u2044\u2088), so the reading is internally consistent. SUPERSEDES Steve Branton's 3\u215C\u2033 (#173), which was \u2077\u2044\u2083\u2082\u2033 short and was the library's founding measurement. \u26A0\uFE0F Steve's number was also the SOLE basis of TURNOUT_LEAD_INCHES_PER_FROG (3.375/7 = 0.482) \u2014 that constant is now baseless as well as refuted."
+    },
+    pointsOffset: {
+      inches: 0.625,
       source: "measured",
-      note: "Steve Branton, physical Atlas code 55 #7, 3\u215C\u2033 points\u2192frog (#173)"
+      note: "Will Gage, physical Atlas 2052 (#7) \u2014 \xB9\u2070\u2044\u2081\u2086\u2033 tie end to point tips"
+    },
+    frogOffset: {
+      inches: 4.21875,
+      source: "measured",
+      note: "Will Gage, physical Atlas 2052 (#7) \u2014 4\u2077\u2044\u2083\u2082\u2033, tie end to the apex of the V. Supersedes an initial 4\xB3\u2044\u2081\u2086\u2033. Worth noting for provenance: he was told beforehand that the clearance hypothesis predicted 4\xBC\u2033 and did NOT read 4\xBC\u2033 \u2014 so this number is not a confirmation artefact, and it sits \xB9\u2044\u2083\u2082\u2033 (one tape division) from the predicted value."
     },
     overallLength: {
       inches: 6,
@@ -1157,15 +1172,20 @@ var ATLAS_CODE55_N = [
     partNumbers: { left: "2054", right: "2055" },
     frogNumber: 10,
     lead: {
-      inches: 4.1875,
+      inches: 4.9375,
       source: "measured",
-      note: "Will Gage, physical Atlas 2054 (#10): points at 9/16\u2033 from the end, frog at 4.75\u2033 \u21D2 4\xB3\u2044\u2081\u2086\u2033. THIS IS THE MEASUREMENT THAT REFUTED the lead-per-frog rule \u2014 it predicted 4.82\u2033, 13% over."
+      note: "Will Gage, physical Atlas 2054 (#10): 5.5\u2033 frog \u2212 9/16\u2033 points = 4\xB9\u2075\u2044\u2081\u2086\u2033. \u26A0\uFE0F CORRECTED from 4\xB3\u2044\u2081\u2086\u2033 (0.53.0\u20130.54.0), which used a 4.75\u2033 frog reading taken before we'd agreed the landmark. That earlier figure was the END OF THE FROG CASTING; a #10's V is shallow, so the apex is \xBE\u2033 back from it."
     },
     pointsOffset: {
       inches: 0.5625,
       source: "measured",
       note: "Will Gage, physical Atlas 2054 (#10) \u2014 end of tie strip to point tips"
     },
+    frogOffset: {
+      inches: 5.5,
+      source: "measured",
+      note: "Will Gage, physical Atlas 2054 (#10) \u2014 tie end to the apex of the V"
+    },
     overallLength: {
       inches: 8,
       source: "measured",
@@ -1215,10 +1235,24 @@ function turnoutPartForSize(size, library = BUILT_IN_TRACK_PARTS) {
     (best, p) => Math.abs(p.frogNumber - size) < Math.abs(best.frogNumber - size) ? p : best
   );
 }
+function measuredLeadPoints(library) {
+  return library.filter(
+    (p) => p.kind === "turnout" && p.frogNumber != null && p.lead?.source === "measured"
+  ).map((p) => ({ n: p.frogNumber, lead: p.lead.inches })).sort((a, b) => a.n - b.n);
+}
 function leadInchesForSize(size, library = BUILT_IN_TRACK_PARTS) {
   const part = turnoutPartForSize(size, library);
   if (part?.lead && part.frogNumber === size) return part.lead.inches;
-  return size * TURNOUT_LEAD_INCHES_PER_FROG;
+  const pts = measuredLeadPoints(library);
+  if (!pts.length) return size * TURNOUT_LEAD_INCHES_PER_FROG;
+  if (pts.length === 1) return size / pts[0].n * pts[0].lead;
+  let i;
+  if (size >= pts[pts.length - 1].n) i = pts.length - 2;
+  else i = Math.max(0, pts.findIndex((p) => p.n >= size) - 1);
+  const lo = pts[i];
+  const hi = pts[i + 1];
+  const t = (size - lo.n) / (hi.n - lo.n);
+  return lo.lead + t * (hi.lead - lo.lead);
 }
 function parseXtpLibrary(text) {
   const parts = [];