Audit round: stored degeneracy measurement, figure guards, 44 assertions

diag_maskdegen writes data/maskdegen.csv so the claim it supports is traceable; the legend guard inflates by the marker radius and refuses a legend that leaves the canvas; one legend size on every figure.
2026-08-19 15:08:38 +09:00
parent 2fb64bee3a
commit e243b29d29
11 changed files with 143 additions and 24 deletions
@@ -160,10 +160,10 @@ chk("ratio spans 1.32 to 1.54", round(min(rt), 2) == 1.32
    and round(max(rt), 2) == 1.54, "%.3f to %.3f" % (min(rt), max(rt)))

 # the three secrets named in the setup
-chk("secret sizes UL=256, perm 256, pad 16",
-    all(t in tex for t in ["$UL=256$ key entries",
-                           "one permutation of $256$ positions",
-                           "$16$ pad\nbits per user"]),
+chk("secret sizes: per-user direction, perm 256, pad 16",
+    all(t in tex for t in ["length-$64$ key direction per user",
+                           "one permutation of $256$",
+                           "$16$ pad bits per user"]),
    "searched tex", needs_tex=True)

 chk("no stale d=64 configuration in tex",
@@ -175,6 +175,47 @@ sc = rows("sec_sens_cmp.csv")
 dv = max(abs(float(r["ser_mask"]) - float(r["ser_perm"])) for r in sc)
 chk("permutation tracks mask in Fig. 5", dv < 0.06, "max gap %.3f" % dv)

+# --- the audit round's corrected quantities ---------------------------
+mf = {r["family"]: r for r in rows("sec_maskfam.csv")}
+fam_pct = (float(mf["random"]["legit_ser"])
+           / float(mf["hadamard"]["legit_ser"]) - 1) * 100
+chk("continuous family 29 percent worse", round(fam_pct) == 29,
+    "%.1f percent" % fam_pct)
+chk("no stale 2.5 factor in tex", "factor of $2.5$" not in tex,
+    "searched tex", needs_tex=True)
+
+sc2 = rows("sec_sens_cmp.csv")
+worst04 = min(min(float(r["ser_mask"]), float(r["ser_perm"]),
+                  float(r["ser_pad"])) for r in sc2
+              if float(r["frac"]) <= 0.4)
+chk("all three above 0.95 to 40 percent of key", worst04 > 0.95,
+    "min %.4f" % worst04)
+
+rf = rows("refresh.csv")
+res = max(1 - float(r["eve_invariant"]) for r in rf)
+chk("refresh residual below 2.4e-3", res < 2.4e-3, "max %.2e" % res)
+
+rk = rows("refresh_kpa.csv")
+nb = max(0.9999847412 - float(r["ser_next_block"]) for r in rk)
+chk("next block within 6e-4 of chance", nb < 6e-4, "max %.2e" % nb)
+
+bc = rows("sec_brute_cmp.csv")
+pm = min(float(r["ser_perm"]) for r in bc)
+chk("permutation floor 0.9996", pm > 0.9996, "min %.5f" % pm)
+
+md = {r["family"]: r for r in rows("maskdegen.csv")}
+ks = [int(x) for x in md["learned"]["support99_per_key"].split("/")]
+chk("learned keys degenerate: 5 to 8 of 64 entries",
+    min(ks) == 5 and max(ks) == 8 and int(md["learned"]["L"]) == 64,
+    md["learned"]["support99_per_key"])
+chk("learned support overlap 0.10",
+    round(float(md["learned"]["mean_overlap"]), 2) == 0.10,
+    md["learned"]["mean_overlap"])
+chk("degeneracy numbers in tex",
+    "$5$ to $8$ of the $64$ entries" in tex and "overlap of\n$0.10$" in tex
+    or "$5$ to $8$ of the $64$ entries" in tex and "overlap of $0.10$" in tex,
+    "searched tex", needs_tex=True)
+
 # --- abstract ---------------------------------------------------------
 a = (tex.split(r"\begin{abstract}")[1].split(r"\end{abstract}")[0].strip()
     if HAVE_TEX else "")
@@ -31,7 +31,7 @@ def support99(w):
    return set(order[:k].tolist()), k


-def describe(name, W):
+def describe(name, W, rows=None):
    L = W.shape[1]
    sups, ks = [], []
    for u in range(W.shape[0]):
@@ -43,17 +43,33 @@ def describe(name, W):
        for j in range(i + 1, len(sups)):
            ov.append(len(sups[i] & sups[j]) / max(1, min(len(sups[i]),
                                                          len(sups[j]))))
+    mo = sum(ov) / len(ov)
    print("%-14s L=%3d  99%%-energy entries per key: %s  "
-          "mean pairwise support overlap %.2f"
-          % (name, L, ks, sum(ov) / len(ov)))
+          "mean pairwise support overlap %.2f" % (name, L, ks, mo))
+    if rows is not None:
+        rows.append([name, L, "/".join(str(k) for k in ks), "%.4f" % mo])
+
+
+def write_rows(rows):
+    """Store the measurement so the manuscript sentence it justifies is
+    traceable to an artifact in data/ like every other quoted number."""
+    import csv
+    out = Path(__file__).resolve().parents[1] / "data" / "maskdegen.csv"
+    with open(out, "w", newline="") as f:
+        w = csv.writer(f)
+        w.writerow(["family", "L", "support99_per_key", "mean_overlap"])
+        w.writerows(rows)
+    print("[csv]", out)


 def main():
    print("main configuration d=%d" % MAIN_D)
+    rows = []
    m_free = get_model(iters=4000)            # keys learned, nothing frozen
-    describe("learned", m_free.masks().detach().cpu())
+    describe("learned", m_free.masks().detach().cpu(), rows)
    m_fix = main_model()
-    describe("Walsh-Hadamard", m_fix.masks().detach().cpu())
+    describe("Walsh-Hadamard", m_fix.masks().detach().cpu(), rows)
+    write_rows(rows)
    print()
    print("A degenerate key set shows few entries per key and near-zero")
    print("overlap; a dense one shows most entries and overlap near one.")
@@ -91,6 +91,17 @@ def col(rows, k, f=float):
    return [f(r[k]) for r in rows]


+def _inflate(box, fig):
+    """Grow a bounding box by the marker radius plus the line width, in
+    pixels, so a marker whose CENTER clears the box cannot still touch
+    its frame."""
+    pad = (plt.rcParams["lines.markersize"] / 2.0
+           + plt.rcParams["lines.linewidth"]) * fig.dpi / 72.0
+    from matplotlib.transforms import Bbox
+    return Bbox.from_extents(box.x0 - pad, box.y0 - pad,
+                             box.x1 + pad, box.y1 + pad)
+
+
 def save(fig, name, insets=()):
    """Write the figure and assert that no axis label is clipped.

@@ -118,7 +129,13 @@ def save(fig, name, insets=()):
        # same discipline as the clipping guard above.
        leg = ax.get_legend()
        if leg is not None:
-            lb = leg.get_window_extent()
+            raw = leg.get_window_extent()
+            if (raw.x0 < fbox.x0 or raw.y0 < fbox.y0
+                    or raw.x1 > fbox.x1 or raw.y1 > fbox.y1):
+                raise RuntimeError(
+                    f"{name}: the legend box leaves the canvas "
+                    f"({raw} outside {fbox}); narrow or move it")
+            lb = _inflate(raw, fig)
            for line in ax.get_lines():
                # full-span reference lines (axhline/axvline) carry axes-
                # fraction endpoints [0,1]; they are not data curves and,
@@ -171,6 +188,10 @@ def save(fig, name, insets=()):
    print("[OK]", name)


+PL_CHOSEN = []      # sizes the sweep settled on, one per figure
+PL_FORCED = None    # set by main() on its second pass
+
+
 def main_legit(snr_db="10"):
    """The legitimate SER of the main configuration, read from the curve
    the main configuration produced rather than looked up by key length."""
@@ -180,10 +201,10 @@ def main_legit(snr_db="10"):
    raise KeyError("no %s dB row in sec_snr.csv" % snr_db)


-def place_legend(ax, cands=("lower left", "center left", "center right",
-                           "lower center", "upper right", "upper center",
-                           "center", "lower right"),
-                 sizes=(7.6, 7.2, 6.8, 6.4, 6.0)):
+def place_legend(ax, cands=("lower left", "upper left", "center left",
+                           "center right", "lower center", "upper right",
+                           "upper center", "center", "lower right"),
+                 sizes=(7.0,), ncol=1):
    """Choose the location and font size whose box the fewest curve points
    fall inside, scored on rendered geometry rather than guessed from the
    data. The size sweep is what makes a long label set placeable: a
@@ -193,14 +214,22 @@ def place_legend(ax, cands=("lower left", "center left", "center right",
    The axes rectangle is applied first, because save() enforces the same
    test after applying it. Scoring the default layout and then checking
    a different one is how a placement that looked clear here failed
-    there."""
+    there.
+
+    When PL_FORCED is set, only that size is tried: the driver runs every
+    figure once to learn the smallest size any of them needs, then reruns
+    them all at that one size so the legends print uniformly."""
    ax.figure.subplots_adjust(**AXES_RECT)
+    if PL_FORCED is not None:
+        sizes = (PL_FORCED,)
    best = None
    for size in sizes:
        for loc in cands:
-            leg = ax.legend(loc=loc, prop={"size": size})
+            leg = ax.legend(loc=loc, prop={"size": size}, ncol=ncol,
+                            handlelength=1.4, columnspacing=0.9,
+                            handletextpad=0.5)
            ax.figure.canvas.draw()
-            lb = leg.get_window_extent()
+            lb = _inflate(leg.get_window_extent(), ax.figure)
            hits = 0
            for line in ax.get_lines():
                xy = line.get_xydata()
@@ -218,9 +247,14 @@ def place_legend(ax, cands=("lower left", "center left", "center right",
            if best is None or hits < best[2]:
                best = (loc, size, hits)
            if hits == 0:
-                ax.legend(loc=loc, prop={"size": size})
+                ax.legend(loc=loc, prop={"size": size}, ncol=ncol,
+                          handlelength=1.4, columnspacing=0.9,
+                          handletextpad=0.5)
+                PL_CHOSEN.append(size)
                return best
-    ax.legend(loc=best[0], prop={"size": best[1]})
+    ax.legend(loc=best[0], prop={"size": best[1]}, ncol=ncol,
+              handlelength=1.4, columnspacing=0.9, handletextpad=0.5)
+    PL_CHOSEN.append(best[1])
    return best


@@ -245,6 +279,9 @@ def fig_snr():
    ax.set_xlabel("SNR (dB)")
    ax.set_ylabel("SER")
    ax.set_xlim(min(x), max(x))
+    # most of a decade below the data leaves the lower-left genuinely
+    # empty, which is what gives the four-entry legend a clear berth
+    ax.set_ylim(bottom=8e-4)
    place_legend(ax)
    save(fig, "fig_sec_snr")

@@ -263,6 +300,7 @@ def fig_keylen():
                marker="^", ls=":", label=LBL["oma"])
    ax.semilogy(x, col(r, "eve_ser"), color=C_EVE, marker="s", ls="--",
                label=LBL["eve_key"])
+    ax.set_ylim(top=6.0)     # headroom above the flat eavesdropper curve
    ax.set_xlabel("Key length $L$")
    ax.set_ylabel("SER")
    ax.set_xscale("log", base=2)
@@ -292,8 +330,9 @@ def fig_jam():
    ax.plot(x, col(r, "perm_blind"), color=C_EVE, marker="s", ls="-.",
            markevery=(me // 2, me), label=LBL["perm"] + ", blind", **OVER)
    nojam = float(load("sec_jam.csv")[0]["nojam"])
-    ax.axhline(nojam, color=C_OMA, ls=(0, (1, 3)), lw=0.9,
-               label=LBL["nojam"])
+    # the unjammed reference is named in the caption rather than in the
+    # legend, which keeps the folded legend two rows tall
+    ax.axhline(nojam, color=C_OMA, ls=(0, (1, 3)), lw=0.9)
    ax.set_xlabel("JSR (dB)")
    ax.set_ylabel("SER")
    ax.set_xlim(min(x), max(x))
@@ -317,6 +356,9 @@ def fig_sens():
            markevery=(2, 3), lw=1.2, mfc="none", label=LBL["pad"])
    chance = 1.0 - (1.0 / 16.0) ** 4
    ax.axhline(chance, color=C_CH, ls=":", lw=0.9, label=LBL["chance"])
+    # the narration reads these curves against the legitimate rate
+    ax.axhline(main_legit(), color=C_OMA, ls=(0, (4, 2)), lw=0.9,
+               label=LBL["legit"])
    ax.set_xlabel("Fraction of the key recovered")
    ax.set_ylabel("Eavesdropper SER")
    ax.set_xlim(0, 1)
@@ -337,7 +379,8 @@ def fig_brute():
    ax.semilogx(x, col(r, "ser_mask"), color=C_LEGIT, marker="o", ls="-",
                label=LBL["mask"])
    legit = main_legit()
-    ax.axhline(legit, color=C_OMA, ls=":", lw=0.9, label=LBL["legit"])
+    ax.axhline(legit, color=C_OMA, ls=(0, (4, 2)), lw=0.9,
+               label=LBL["legit"])
    ax.set_xlabel("Number of key guesses $K$")
    ax.set_ylabel("Eavesdropper SER")
    ax.set_ylim(0.0, 1.05)      # keep the reference line off the spine
@@ -390,7 +433,8 @@ def fig_kpa():
    # in the main configuration, rather than the user-1 convention of the
    # scheme-comparison table
    legit = main_legit()
-    ax.axhline(legit, color=C_OMA, ls=":", lw=0.9, label=LBL["legit"])
+    ax.axhline(legit, color=C_OMA, ls=(0, (4, 2)), lw=0.9,
+               label=LBL["legit"])
    ax.set_xlabel("Known-plaintext frames $N$")
    ax.set_ylabel("Eavesdropper SER")
    ax.set_xscale("log", base=2)
@@ -401,7 +445,7 @@ def fig_kpa():
    save(fig, "fig_sec_kpa")


-def main():
+def run_all():
    fig_snr()
    fig_keylen()
    fig_jam()
@@ -418,6 +462,21 @@ def main():
        fig_kpa()
    except FileNotFoundError:
        print("[skip] known-plaintext CSV not present yet")
+
+
+def main():
+    """Two passes: the first learns the smallest legend size any figure
+    needs, the second forces that one size everywhere so the legends
+    print uniformly, which the figure standard requires."""
+    global PL_FORCED
+    PL_FORCED = None
+    PL_CHOSEN.clear()
+    run_all()
+    if PL_CHOSEN:
+        PL_FORCED = min(PL_CHOSEN)
+        print("[uniform] legend size %.1f pt on every figure" % PL_FORCED)
+        PL_CHOSEN.clear()
+        run_all()
    print("[done] figures in", FIG)


@@ -0,0 +1,3 @@
+family,L,support99_per_key,mean_overlap
+learned,64,5/5/8/7,0.0952
+Walsh-Hadamard,64,64/64/64/64,1.0000