343 lines
18 KiB
Python
343 lines
18 KiB
Python
|
#!/usr/bin/env python3
|
|||
|
|
from __future__ import annotations
|
|||
|
|
|
|||
|
|
import csv
|
|||
|
|
import html
|
|||
|
|
import json
|
|||
|
|
from pathlib import Path
|
|||
|
|
|
|||
|
|
import matplotlib as mpl
|
|||
|
|
import matplotlib.pyplot as plt
|
|||
|
|
from matplotlib.patches import FancyBboxPatch, FancyArrowPatch, Rectangle
|
|||
|
|
import numpy as np
|
|||
|
|
|
|||
|
|
PROJECT_ROOT = Path(__file__).resolve().parents[1]
|
|||
|
|
ROOT = PROJECT_ROOT / "项目记录" / "拓展分析阶段_发表级完整实验_20260616"
|
|||
|
|
FIG_DIR = ROOT / "figures" / "nature_style"
|
|||
|
|
TABLE_DIR = ROOT / "tables" / "nature_style"
|
|||
|
|
REPORT = ROOT / "summary_report_publication.html"
|
|||
|
|
|
|||
|
|
mpl.rcParams.update({
|
|||
|
|
"font.family": "sans-serif",
|
|||
|
|
"font.sans-serif": ["Arial", "Helvetica", "DejaVu Sans", "sans-serif"],
|
|||
|
|
"svg.fonttype": "none",
|
|||
|
|
"pdf.fonttype": 42,
|
|||
|
|
"font.size": 7,
|
|||
|
|
"axes.spines.right": False,
|
|||
|
|
"axes.spines.top": False,
|
|||
|
|
"axes.linewidth": 0.7,
|
|||
|
|
"legend.frameon": False,
|
|||
|
|
})
|
|||
|
|
|
|||
|
|
PALETTE = {
|
|||
|
|
"ink": "#1b2430",
|
|||
|
|
"muted": "#687386",
|
|||
|
|
"line": "#d6dce6",
|
|||
|
|
"green": "#2f7d55",
|
|||
|
|
"green_soft": "#dcefe4",
|
|||
|
|
"red": "#b33a3a",
|
|||
|
|
"red_soft": "#f3d7d7",
|
|||
|
|
"blue": "#2d6f9f",
|
|||
|
|
"blue_soft": "#dce9f5",
|
|||
|
|
"amber": "#b5791f",
|
|||
|
|
"amber_soft": "#f4e3c5",
|
|||
|
|
"grey": "#f2f4f7",
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
|
|||
|
|
def load_json(rel: str):
|
|||
|
|
return json.loads((ROOT / rel).read_text(encoding="utf-8"))
|
|||
|
|
|
|||
|
|
|
|||
|
|
def short_name(name: str) -> str:
|
|||
|
|
return {
|
|||
|
|
"sequence_extraction": "Sequence\nextraction",
|
|||
|
|
"hmmer_domain": "HMMER\ndomain",
|
|||
|
|
"motif_meme": "MEME\nmotif",
|
|||
|
|
"phylogeny_tree": "Phylogeny\ntree",
|
|||
|
|
"reference_classification": "Reference\nclassification",
|
|||
|
|
"synteny_collinearity": "Synteny\ncollinearity",
|
|||
|
|
"kaks_pairwise": "Ka/Ks\npairwise",
|
|||
|
|
"pep_properties": "Protein\nproperties",
|
|||
|
|
}.get(name, name.replace("_", "\n"))
|
|||
|
|
|
|||
|
|
|
|||
|
|
def save_all(fig, stem: Path):
|
|||
|
|
stem.parent.mkdir(parents=True, exist_ok=True)
|
|||
|
|
fig.savefig(stem.with_suffix(".svg"), bbox_inches="tight")
|
|||
|
|
fig.savefig(stem.with_suffix(".pdf"), bbox_inches="tight")
|
|||
|
|
fig.savefig(stem.with_suffix(".png"), dpi=450, bbox_inches="tight")
|
|||
|
|
fig.savefig(stem.with_suffix(".tiff"), dpi=600, bbox_inches="tight")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def panel_label(ax, label):
|
|||
|
|
ax.text(-0.04, 1.04, label, transform=ax.transAxes, fontweight="bold", fontsize=10, va="top", ha="left", color=PALETTE["ink"])
|
|||
|
|
|
|||
|
|
|
|||
|
|
def draw_workflow(ax):
|
|||
|
|
ax.set_axis_off()
|
|||
|
|
ax.set_xlim(0, 1)
|
|||
|
|
ax.set_ylim(0, 1)
|
|||
|
|
steps = [
|
|||
|
|
("Executable\ncode + data", PALETTE["blue_soft"], 0.05),
|
|||
|
|
("Spec\nconversion", "#ffffff", 0.23),
|
|||
|
|
("Deterministic\nwrapper", PALETTE["green_soft"], 0.41),
|
|||
|
|
("Golden\ncomparison", PALETTE["amber_soft"], 0.59),
|
|||
|
|
("Template +\npublish gates", PALETTE["amber_soft"], 0.77),
|
|||
|
|
("Published\nextension", PALETTE["green_soft"], 0.91),
|
|||
|
|
]
|
|||
|
|
y = 0.62
|
|||
|
|
for i, (txt, face, x) in enumerate(steps):
|
|||
|
|
w = 0.13 if i < 5 else 0.11
|
|||
|
|
patch = FancyBboxPatch((x - w/2, y - 0.105), w, 0.21, boxstyle="round,pad=0.012,rounding_size=0.012", fc=face, ec=PALETTE["line"], lw=0.9)
|
|||
|
|
ax.add_patch(patch)
|
|||
|
|
ax.text(x, y, txt, ha="center", va="center", fontsize=7.2, color=PALETTE["ink"], fontweight="bold" if i in {2, 3, 4, 5} else "normal")
|
|||
|
|
if i < len(steps) - 1:
|
|||
|
|
x2 = steps[i + 1][2]
|
|||
|
|
ax.add_patch(FancyArrowPatch((x + w/2 + 0.005, y), (x2 - 0.065, y), arrowstyle="-|>", mutation_scale=8, lw=0.9, color=PALETTE["muted"]))
|
|||
|
|
ax.text(0.05, 0.88, "Validation-driven extension-stage construction", fontsize=10, fontweight="bold", color=PALETTE["ink"], ha="left")
|
|||
|
|
ax.text(0.05, 0.78, "API output is treated as a candidate; publication requires deterministic execution evidence.", fontsize=7.5, color=PALETTE["muted"], ha="left")
|
|||
|
|
ax.add_patch(FancyBboxPatch((0.53, 0.18), 0.36, 0.18, boxstyle="round,pad=0.012,rounding_size=0.012", fc=PALETTE["red_soft"], ec="#d8a5a5", lw=0.8))
|
|||
|
|
ax.text(0.71, 0.27, "Failed candidates are blocked\nbefore entering the published stage list", ha="center", va="center", fontsize=7.5, color=PALETTE["ink"])
|
|||
|
|
ax.add_patch(FancyArrowPatch((0.77, 0.52), (0.72, 0.36), arrowstyle="-|>", mutation_scale=8, lw=0.9, color=PALETTE["red"]))
|
|||
|
|
|
|||
|
|
|
|||
|
|
def draw_lifecycle_matrix(ax, det_rows, api_rows):
|
|||
|
|
phases = ["Build", "Trial", "Golden", "Template\ngate", "Publish"]
|
|||
|
|
cases = [r["archetype_id"] for r in det_rows]
|
|||
|
|
values = []
|
|||
|
|
for case in cases:
|
|||
|
|
d = next(r for r in det_rows if r["archetype_id"] == case)
|
|||
|
|
a = next((r for r in api_rows if r["archetype_id"] == case), {})
|
|||
|
|
values.append([bool(d["build_success"]), bool(d["trial_success"]), d["golden_status"] == "passed", d["template_gate_status"] == "passed", bool(d["publish_success"])])
|
|||
|
|
values.append([bool(a.get("build_success")), bool(a.get("trial_success")), a.get("golden_status") == "passed", a.get("template_gate_status") == "passed", bool(a.get("publish_success"))])
|
|||
|
|
ax.set_xlim(0, len(phases))
|
|||
|
|
ax.set_ylim(0, len(values))
|
|||
|
|
ax.invert_yaxis()
|
|||
|
|
for y, row in enumerate(values):
|
|||
|
|
for x, val in enumerate(row):
|
|||
|
|
ax.add_patch(Rectangle((x, y), 1, 1, fc=PALETTE["green_soft"] if val else PALETTE["red_soft"], ec="white", lw=1.2))
|
|||
|
|
ax.text(x + 0.5, y + 0.5, "✓" if val else "×", ha="center", va="center", fontsize=8, color=PALETTE["green"] if val else PALETTE["red"], fontweight="bold")
|
|||
|
|
for y in range(0, len(values), 2):
|
|||
|
|
ax.add_patch(Rectangle((0, y), len(phases), 2, fill=False, ec=PALETTE["line"], lw=0.6))
|
|||
|
|
ax.set_xticks(np.arange(len(phases)) + 0.5, phases, fontsize=7)
|
|||
|
|
ylabels = []
|
|||
|
|
for case in cases:
|
|||
|
|
ylabels.extend([short_name(case) + "\nD", short_name(case) + "\nAPI"])
|
|||
|
|
ax.set_yticks(np.arange(len(values)) + 0.5, ylabels, fontsize=5.8)
|
|||
|
|
ax.tick_params(length=0)
|
|||
|
|
for sp in ax.spines.values():
|
|||
|
|
sp.set_visible(False)
|
|||
|
|
ax.set_title("Lifecycle gate outcomes for deterministic and API/Agent conversion", loc="left", fontsize=8.5, fontweight="bold", pad=8)
|
|||
|
|
|
|||
|
|
|
|||
|
|
def draw_api_attempts(ax, api_rows):
|
|||
|
|
names = [short_name(r["archetype_id"]) for r in api_rows]
|
|||
|
|
attempts = [int(r.get("api_attempt_count") or 0) for r in api_rows]
|
|||
|
|
ok = [bool(r.get("publish_success")) for r in api_rows]
|
|||
|
|
colors = [PALETTE["blue"] if x else PALETTE["red"] for x in ok]
|
|||
|
|
y = np.arange(len(names))
|
|||
|
|
ax.barh(y, attempts, color=colors, height=0.62)
|
|||
|
|
for yi, val, success in zip(y, attempts, ok):
|
|||
|
|
ax.text(val + 0.2, yi, str(val), va="center", ha="left", fontsize=7, color=PALETTE["ink"])
|
|||
|
|
if not success:
|
|||
|
|
ax.text(max(val * 0.45, 0.2), yi, "blocked", va="center", ha="center", fontsize=6.5, color="white", fontweight="bold")
|
|||
|
|
ax.set_yticks(y, names, fontsize=6.6)
|
|||
|
|
ax.invert_yaxis()
|
|||
|
|
ax.set_xlabel("API attempts before final decision", fontsize=7)
|
|||
|
|
ax.set_xlim(0, max(attempts) + 2)
|
|||
|
|
ax.grid(axis="x", color=PALETTE["line"], lw=0.5)
|
|||
|
|
ax.set_title("API/Agent conversion required variable repair effort", loc="left", fontsize=8.5, fontweight="bold", pad=8)
|
|||
|
|
|
|||
|
|
|
|||
|
|
def draw_summary_bars(ax, det, api, neg):
|
|||
|
|
labels = ["Deterministic\nchain", "API/Agent\nconversion", "Negative\ncontrols"]
|
|||
|
|
numerator = [det["passed"], api["passed"], neg["blocked"]]
|
|||
|
|
denom = [det["case_count"], api["case_count"], neg["case_count"]]
|
|||
|
|
colors = [PALETTE["green"], PALETTE["blue"], PALETTE["amber"]]
|
|||
|
|
x = np.arange(3)
|
|||
|
|
ax.bar(x, [n / d for n, d in zip(numerator, denom)], color=colors, width=0.58)
|
|||
|
|
for xi, n, d in zip(x, numerator, denom):
|
|||
|
|
ax.text(xi, n / d + 0.045, f"{n}/{d}", ha="center", va="bottom", fontsize=8, fontweight="bold", color=PALETTE["ink"])
|
|||
|
|
ax.set_xticks(x, labels, fontsize=7)
|
|||
|
|
ax.set_ylim(0, 1.16)
|
|||
|
|
ax.set_ylabel("Fraction passing expected decision", fontsize=7)
|
|||
|
|
ax.set_yticks([0, 0.5, 1.0], ["0", "0.5", "1.0"])
|
|||
|
|
ax.grid(axis="y", color=PALETTE["line"], lw=0.5)
|
|||
|
|
ax.set_title("Three independent evidence layers support safe extensibility", loc="left", fontsize=8.5, fontweight="bold", pad=8)
|
|||
|
|
|
|||
|
|
|
|||
|
|
def draw_negative_controls(ax, rows):
|
|||
|
|
labels = [r["scenario"].replace("_", "\n") for r in rows]
|
|||
|
|
y = np.arange(len(rows))
|
|||
|
|
ax.barh(y, [1] * len(rows), color=PALETTE["red_soft"], height=0.6, edgecolor="#d8a5a5", lw=0.6)
|
|||
|
|
for yi, r in zip(y, rows):
|
|||
|
|
ax.text(0.5, yi, "blocked" if r["blocked"] and not r["published"] else "leaked", ha="center", va="center", fontsize=7, color=PALETTE["red"], fontweight="bold")
|
|||
|
|
ax.set_yticks(y, labels, fontsize=6.7)
|
|||
|
|
ax.set_xticks([])
|
|||
|
|
ax.set_xlim(0, 1)
|
|||
|
|
ax.invert_yaxis()
|
|||
|
|
for sp in ax.spines.values():
|
|||
|
|
sp.set_visible(False)
|
|||
|
|
ax.set_title("Failure-injection cases were blocked before publication", loc="left", fontsize=8.5, fontweight="bold", pad=8)
|
|||
|
|
|
|||
|
|
|
|||
|
|
def write_source_tables(det, api, neg):
|
|||
|
|
TABLE_DIR.mkdir(parents=True, exist_ok=True)
|
|||
|
|
with (TABLE_DIR / "source_data_main_figure.csv").open("w", encoding="utf-8", newline="") as f:
|
|||
|
|
fields = ["group", "archetype_id", "build", "trial", "golden", "template_gate", "publish", "api_attempts", "error"]
|
|||
|
|
w = csv.DictWriter(f, fieldnames=fields)
|
|||
|
|
w.writeheader()
|
|||
|
|
for group, rows in [("deterministic", det["rows"]), ("api_agent", api["rows"] )]:
|
|||
|
|
for r in rows:
|
|||
|
|
w.writerow({
|
|||
|
|
"group": group,
|
|||
|
|
"archetype_id": r["archetype_id"],
|
|||
|
|
"build": r.get("build_success"),
|
|||
|
|
"trial": r.get("trial_success"),
|
|||
|
|
"golden": r.get("golden_status"),
|
|||
|
|
"template_gate": r.get("template_gate_status"),
|
|||
|
|
"publish": r.get("publish_success"),
|
|||
|
|
"api_attempts": r.get("api_attempt_count", ""),
|
|||
|
|
"error": r.get("error", ""),
|
|||
|
|
})
|
|||
|
|
with (TABLE_DIR / "source_data_negative_controls.csv").open("w", encoding="utf-8", newline="") as f:
|
|||
|
|
fields = ["scenario", "blocked", "published", "error"]
|
|||
|
|
w = csv.DictWriter(f, fieldnames=fields)
|
|||
|
|
w.writeheader()
|
|||
|
|
for r in neg["rows"]:
|
|||
|
|
w.writerow({k: r.get(k, "") for k in fields})
|
|||
|
|
|
|||
|
|
|
|||
|
|
def make_main_figure(det, api, neg):
|
|||
|
|
fig = plt.figure(figsize=(7.2, 8.0), constrained_layout=False)
|
|||
|
|
gs = fig.add_gridspec(3, 2, height_ratios=[1.02, 1.6, 1.0], width_ratios=[1.18, 0.82], hspace=0.52, wspace=0.36)
|
|||
|
|
axA = fig.add_subplot(gs[0, :])
|
|||
|
|
axB = fig.add_subplot(gs[1, 0])
|
|||
|
|
axC = fig.add_subplot(gs[1, 1])
|
|||
|
|
axD = fig.add_subplot(gs[2, 0])
|
|||
|
|
axE = fig.add_subplot(gs[2, 1])
|
|||
|
|
draw_workflow(axA)
|
|||
|
|
panel_label(axA, "a")
|
|||
|
|
draw_lifecycle_matrix(axB, det["rows"], api["rows"])
|
|||
|
|
panel_label(axB, "b")
|
|||
|
|
draw_api_attempts(axC, api["rows"])
|
|||
|
|
panel_label(axC, "c")
|
|||
|
|
draw_summary_bars(axD, det, api, neg)
|
|||
|
|
panel_label(axD, "d")
|
|||
|
|
draw_negative_controls(axE, neg["rows"])
|
|||
|
|
panel_label(axE, "e")
|
|||
|
|
fig.text(0.012, 0.012, "Source data: real extension-stage publication experiment package; deterministic chain n=8, API/Agent n=8, negative controls n=4.", fontsize=6.5, color=PALETTE["muted"])
|
|||
|
|
save_all(fig, FIG_DIR / "figure_1_extension_stage_validation")
|
|||
|
|
plt.close(fig)
|
|||
|
|
|
|||
|
|
|
|||
|
|
def make_html(det, api, neg):
|
|||
|
|
failed = [r for r in api["rows"] if not r.get("publish_success")]
|
|||
|
|
failure_text = failed[0]["error"] if failed else ""
|
|||
|
|
rows_det = "".join(
|
|||
|
|
f"<tr><td>{html.escape(r['archetype_id'])}</td><td>{r['build_success']}</td><td>{r['trial_success']}</td><td>{html.escape(str(r['golden_status']))}</td><td>{html.escape(str(r['template_gate_status']))}</td><td>{r['publish_success']}</td><td>{r.get('output_count')}</td></tr>"
|
|||
|
|
for r in det["rows"]
|
|||
|
|
)
|
|||
|
|
rows_api = "".join(
|
|||
|
|
f"<tr><td>{html.escape(r['archetype_id'])}</td><td>{r.get('api_attempt_count')}</td><td>{html.escape(str(r.get('golden_status')))}</td><td>{html.escape(str(r.get('template_gate_status')))}</td><td>{r.get('publish_success')}</td><td>{html.escape(str(r.get('error') or '')[:220])}</td></tr>"
|
|||
|
|
for r in api["rows"]
|
|||
|
|
)
|
|||
|
|
rows_neg = "".join(
|
|||
|
|
f"<tr><td>{html.escape(r['scenario'])}</td><td>{r['blocked']}</td><td>{r['published']}</td><td>{html.escape(r['error'][:220])}</td></tr>"
|
|||
|
|
for r in neg["rows"]
|
|||
|
|
)
|
|||
|
|
doc = f"""<!doctype html>
|
|||
|
|
<html lang=\"zh-CN\">
|
|||
|
|
<head>
|
|||
|
|
<meta charset=\"utf-8\">
|
|||
|
|
<meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">
|
|||
|
|
<title>拓展分析阶段发表级实验结果</title>
|
|||
|
|
<style>
|
|||
|
|
:root {{ --ink:#1b2430; --muted:#687386; --line:#d6dce6; --blue:#2d6f9f; --green:#2f7d55; --red:#b33a3a; --amber:#b5791f; --panel:#f7f9fc; }}
|
|||
|
|
body {{ margin:0; font:14px/1.68 Arial, 'Microsoft YaHei', sans-serif; color:var(--ink); background:#fff; }}
|
|||
|
|
main {{ max-width:1180px; margin:0 auto; padding:30px 28px 60px; }}
|
|||
|
|
h1 {{ font-size:28px; margin:0 0 8px; }}
|
|||
|
|
h2 {{ margin:34px 0 12px; font-size:20px; border-left:5px solid var(--blue); padding-left:10px; }}
|
|||
|
|
h3 {{ margin:20px 0 8px; font-size:16px; }}
|
|||
|
|
.subtitle {{ color:var(--muted); max-width:920px; }}
|
|||
|
|
.grid {{ display:grid; grid-template-columns:repeat(4,minmax(0,1fr)); gap:12px; margin:18px 0; }}
|
|||
|
|
.metric {{ border:1px solid var(--line); background:var(--panel); padding:14px; border-radius:8px; }}
|
|||
|
|
.metric b {{ display:block; font-size:24px; color:var(--blue); }}
|
|||
|
|
.metric span {{ color:var(--muted); }}
|
|||
|
|
.note {{ border-left:4px solid var(--blue); background:#eaf2fa; padding:11px 13px; margin:12px 0; }}
|
|||
|
|
.warn {{ border-left-color:var(--amber); background:#f7ecd7; }}
|
|||
|
|
.good {{ color:var(--green); font-weight:700; }}
|
|||
|
|
.bad {{ color:var(--red); font-weight:700; }}
|
|||
|
|
table {{ width:100%; border-collapse:collapse; margin:12px 0 22px; table-layout:fixed; }}
|
|||
|
|
th,td {{ border:1px solid var(--line); padding:8px 9px; vertical-align:top; word-break:break-word; }}
|
|||
|
|
th {{ background:#edf2f8; text-align:left; }}
|
|||
|
|
.figure {{ border:1px solid var(--line); padding:12px; border-radius:8px; background:#fff; margin:14px 0 22px; }}
|
|||
|
|
.figure img {{ max-width:100%; display:block; margin:auto; }}
|
|||
|
|
.caption {{ color:var(--muted); font-size:13px; margin-top:8px; }}
|
|||
|
|
code {{ background:#f2f4f7; padding:1px 4px; border-radius:4px; font-size:12px; }}
|
|||
|
|
@media (max-width:850px) {{ .grid {{ grid-template-columns:1fr 1fr; }} main {{ padding:20px 14px 42px; }} table {{ font-size:12px; }} }}
|
|||
|
|
</style>
|
|||
|
|
</head>
|
|||
|
|
<body><main>
|
|||
|
|
<header>
|
|||
|
|
<h1>拓展分析阶段：发表级实验结果</h1>
|
|||
|
|
<p class=\"subtitle\">本文档将前期真实执行结果整理为论文结果写作状态。核心结论是：确定性拓展阶段构建链路在 8 类真实数据 archetype 上全部通过；API/Agent 自动转写在同一任务集上达到 7/8 发布成功，失败案例被 golden validation 与发布门禁拦截。</p>
|
|||
|
|
</header>
|
|||
|
|
<section>
|
|||
|
|
<h2>一、结果总览</h2>
|
|||
|
|
<div class=\"grid\">
|
|||
|
|
<div class=\"metric\"><b>{det['passed']}/{det['case_count']}</b><span>确定性构建链路通过</span></div>
|
|||
|
|
<div class=\"metric\"><b>{api['passed']}/{api['case_count']}</b><span>API/Agent 转写发布成功</span></div>
|
|||
|
|
<div class=\"metric\"><b>{neg['blocked']}/{neg['case_count']}</b><span>结构性负例被阻断</span></div>
|
|||
|
|
<div class=\"metric\"><b>60</b><span>拓展阶段相关单元测试通过</span></div>
|
|||
|
|
</div>
|
|||
|
|
<div class=\"note\">论文表述应分开写：<b>确定性主链路 8/8</b> 证明系统闭环稳定；<b>API/Agent 7/8</b> 是自动转写能力对照；<b>负例 4/4 blocked</b> 证明发布门禁有效。</div>
|
|||
|
|
</section>
|
|||
|
|
<section>
|
|||
|
|
<h2>二、主图</h2>
|
|||
|
|
<div class=\"figure\"><img src=\"figures/nature_style/figure_1_extension_stage_validation.svg\" alt=\"Extension-stage validation multi-panel figure\"><div class=\"caption\"><b>Figure 1 | Validation-driven construction of extension analysis stages.</b> a, Extension-stage construction and publication workflow. b, Lifecycle gate outcomes for deterministic and API/Agent conversion across eight archetypes. D = deterministic chain; API = DeepSeek/API Agent conversion. c, API attempts required before final decision. d, Summary of three evidence layers. e, Negative-control cases blocked before publication.</div></div>
|
|||
|
|
<p>导出文件：<code>figure_1_extension_stage_validation.svg</code>, <code>.pdf</code>, <code>.png</code>, <code>.tiff</code>。Source data 已导出到 <code>tables/nature_style/</code>。</p>
|
|||
|
|
</section>
|
|||
|
|
<section>
|
|||
|
|
<h2>三、确定性主链路</h2>
|
|||
|
|
<p>8 类真实数据 archetype 均通过 build、trial、golden comparison、template gate 和 publish gate。该结果可作为论文主结果。</p>
|
|||
|
|
<table><thead><tr><th>Archetype</th><th>Build</th><th>Trial</th><th>Golden</th><th>Template gate</th><th>Publish</th><th>Outputs</th></tr></thead><tbody>{rows_det}</tbody></table>
|
|||
|
|
</section>
|
|||
|
|
<section>
|
|||
|
|
<h2>四、API/Agent 对照</h2>
|
|||
|
|
<p>API/Agent 对照完整覆盖同一 8 类任务，结果为 <b>{api['passed']}/{api['case_count']}</b> 发布成功。失败的 <code>hmmer_domain</code> 并非系统执行链路失败，而是 API 生成的 semantic command template 未复现原始行为，被 golden/publish gate 阻断。</p>
|
|||
|
|
<div class=\"note warn\">失败原因摘要：{html.escape(failure_text[:420])}</div>
|
|||
|
|
<table><thead><tr><th>Archetype</th><th>API attempts</th><th>Golden</th><th>Template gate</th><th>Publish</th><th>Error</th></tr></thead><tbody>{rows_api}</tbody></table>
|
|||
|
|
</section>
|
|||
|
|
<section>
|
|||
|
|
<h2>五、失败拦截负例</h2>
|
|||
|
|
<p>4 个结构性负例全部被阻断，没有进入 published stage 列表。该实验应称为 <b>negative-control / failure-injection test</b>，不是严格意义的消融实验。</p>
|
|||
|
|
<table><thead><tr><th>Scenario</th><th>Blocked</th><th>Published</th><th>Blocking evidence</th></tr></thead><tbody>{rows_neg}</tbody></table>
|
|||
|
|
</section>
|
|||
|
|
<section>
|
|||
|
|
<h2>六、论文可用表述</h2>
|
|||
|
|
<p>推荐写法：</p>
|
|||
|
|
<div class=\"note\">We evaluated the extension-stage construction framework using eight real-data executable-code archetypes. The deterministic construction pipeline passed all eight cases through trial execution, original-vs-converted golden comparison, template-gate validation and publish-gate checking. In an API/Agent conversion benchmark on the same cases, seven of eight stages were successfully published, while one HMMER-domain case was blocked because the API-generated semantic command template failed to reproduce the original behavior. Four structural negative controls were all intercepted before publication.</div>
|
|||
|
|
<p>不应写成：<span class=\"bad\">API/LLM 自动转写 8/8 全部成功</span>。</p>
|
|||
|
|
</section>
|
|||
|
|
</main></body></html>
|
|||
|
|
"""
|
|||
|
|
REPORT.write_text(doc, encoding="utf-8")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def main():
|
|||
|
|
FIG_DIR.mkdir(parents=True, exist_ok=True)
|
|||
|
|
TABLE_DIR.mkdir(parents=True, exist_ok=True)
|
|||
|
|
det = load_json("results/archetype_spec/summary.json")
|
|||
|
|
api = load_json("results/deepseek_agent/summary.json")
|
|||
|
|
neg = load_json("negative_cases/negative_control_summary.json")
|
|||
|
|
write_source_tables(det, api, neg)
|
|||
|
|
make_main_figure(det, api, neg)
|
|||
|
|
make_html(det, api, neg)
|
|||
|
|
print(REPORT)
|
|||
|
|
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
main()
|