证据面板的分析结论：结果表的格写列名，总数那一格的值按服务端的写法显示。①「这一处引用的是结果表第 1 行『share』」原来写的是列 id，现在服务端在 cited 上另给 column_label（结果表记的 labels，如「占变化的比重」），目录条目也带上 labels，证据面板、读屏的出处、导出页、可验证报告都照列名写；② 输入里「总数那一格 … = 159175.18」和同一节的闭合核对「159,175.18」两个写法并排，现在服务端另给 value_rendered（render_cell，千分位、NULL 写「—」，和闭合核对同一个写法），三处界面照原样显示。结果表数的换算（比率、空格）不在这次改动里

2026-11-15 00:01:58 +00:00 · 2026-10-21 15:53:27 +00:00
parent b919890b7b
commit c54196326f
13 changed files with 171 additions and 44 deletions
@@ -1067,7 +1067,10 @@ def _catalog_alias_of(catalog: dict[str, Any], artifact: Any) -> str | None:


 def _analysis_cited(cite: dict[str, Any], result: dict[str, Any] | None) -> dict[str, Any]:
-    """这一处引用的是结论的哪一项：判定语（phrase）、具名标量（value：键、原值、换算）、结果表的一格（cell：行、列、原值）。"""
+    """这一处引用的是结论的哪一项：判定语（phrase）、具名标量（value：键、原值、换算）、结果表的一格（cell：行、列、原值）。
+
+    结果表的列写给人看：column 是稳定的列 id（share、delta……，引用里就是这么写的），column_label 是结果表记的列名
+    （「占变化的比重」）；页面照原样显示 column_label，不把列 id 当叫法。"""
    locator = cite.get("locator") or {}
    if cite.get("kind") == "cell":
        row, column = locator.get("row"), locator.get("column")
@@ -1076,7 +1079,9 @@ def _analysis_cited(cite: dict[str, Any], result: dict[str, Any] | None) -> dict
        if table is not None:
            with contextlib.suppress(Exception):
                raw, column = locate_cell(table, row, column)
-        return {"what": "cell", "row": row, "column": column, "raw": raw}
+        labels = (table or {}).get("labels") if isinstance((table or {}).get("labels"), dict) else {}
+        return {"what": "cell", "row": row, "column": column,
+                "column_label": str(labels.get(column) or column) if column is not None else None, "raw": raw}
    if locator.get("key"):
        values = (result or {}).get("values") if isinstance((result or {}).get("values"), dict) else {}
        return {"what": "value", "key": locator["key"], "raw": values.get(locator["key"]),
@@ -1108,12 +1113,13 @@ async def _analysis_chain(seg: dict[str, Any], cite: dict[str, Any], entry: dict
    """分析结论的出处链（00-总方案 §3.3）：判定 → 方法 → 计算 → 输入 → 复算 → 封存。

    - 判定（analysis）：判定语、判定标准、判定码和点名的段、附带说明、算不了的原因，这一处引用的是哪一项（判定语、具名标量、
-      结果表的一格）；「在验证什么」另列，谁写的照实标：Agent 现场做的写「Agent 写的假设」
+      结果表的一格，列另写结果表记的列名 column_label）；「在验证什么」另列，谁写的照实标：Agent 现场做的写「Agent 写的假设」
      （HYPOTHESIS_NOTE），分析卡做的写「分析卡上写的『在验证什么』」（CARD_HYPOTHESIS_NOTE，模板作者写的）
    - 方法（analysis_method）：方法与版本、规范化后的参数、实际用的阈值（只列这个算子用到的）、现查的表范围（probes）
    - 计算（analysis_calc）：结果表（前 table_top_k 段，加上被引用的那一行附近）、具名的数、关键数的代入式
      （evidence.analysis_formulas）
-    - 输入（analysis_input，每份一步）：落到报告目录的哪个 Q、用了哪些列和行、headline 那一格和结论里总数的闭合核对、
+    - 输入（analysis_input，每份一步）：落到报告目录的哪个 Q、用了哪些列和行、headline 那一格（值另由服务端渲染成 value_rendered，
+      和闭合核对同一个写法）和结论里总数的闭合核对、
      下钻的上一层是哪个 A；后面接上每份输入快照的查询步骤（只认封存范围内交回过的，遮罩照查询步骤的办法）
    - 复算（analysis_recompute）：按 sha256 取回封存的输入、用同一个函数复算，和封存的逐字段比（evidence.analysis_recheck）；
      方法升过版、复算不了都照实写
@@ -1244,6 +1250,10 @@ async def _analysis_inputs(alias: str, result: dict[str, Any], catalog: dict[str
            "role_label": roles.get(role, role or "输入"), "artifact": artifact or None,
            **{k: item[k] for k in ("columns", "rows", "row", "column", "value") if k in item},
        }
+        # 那一格的值由服务端渲染（render_cell：千分位、NULL 写「—」，和闭合核对里的字同一个写法），页面照原样显示
+        if "value" in item:
+            with contextlib.suppress(RenderError):
+                step["value_rendered"] = render_cell(item.get("value"))
        if role == "parent":
            step["alias"] = _catalog_alias_of(catalog, artifact)
            parent, ok = _analysis_of(artifact)
@@ -1461,6 +1461,7 @@
    const loc = isObj(cite.locator) ? cite.locator : {}
    if (cite.kind === 'metric') return T.srcMetric(own(entry, 'name') || loc.metric || cite.ref || '')
    if (cite.kind === 'cell' && own(entry, 'kind') === 'analysis' && Number.isInteger(loc.row)) {
+      // 结果表的格写列名：核对到结论之后换成结论记的列名（checkAnalysis 的 info.source），这里先照引用里的列 id 说
      return T.srcAnalysisCell(cite.alias, loc.row, String(loc.column ?? ''))
    }
    if (cite.kind === 'cell') {
@@ -1573,6 +1574,9 @@
        const [raw, column, key] = locateCell(table, loc.row, loc.column)
        info.row = loc.row
        info.column = column
+        // 出处面板写结果表记的列名（labels），不把列 id 当叫法；目录条目和文档引用里带的那份先认，没有再查这份结论
+        info.columnLabel = analysisColumnLabel(table, column)
+        info.source = T.srcAnalysisCell(cite.alias, loc.row, info.columnLabel)
        info.truth = cellValue(raw, columnKind(table, column))
        info.truthToken = tokenAt(tokensOf(art), ['table', 'rows', loc.row, key])
        if (typeof raw === 'string' && typeof info.truth === 'number') info.truthText = pyStrip(raw)
@@ -1600,6 +1604,14 @@
    return checkShown(info, seg, () => renderCell(result.phrase, null, null, 'text', 1))
  }

+  /** 分析结论结果表的一格写给人看的列名：结论结果表记的 labels（「占变化的比重」），没起名字的列才写列 id。
+   *  文档里的引用只记列 id（引用就这么写），列名照随文件附带的结论认 */
+  function analysisColumnLabel(table, column) {
+    const id = String(column ?? '')
+    const labels = isObj(table.labels) ? table.labels : {}
+    return own(labels, id) ? String(labels[id]) : id
+  }
+
  /** 口径卡的单位：服务端是 str(unit or "")，不是文字的本页不认 */
  function unitOf(unit) {
    if (!unit) return ''
@@ -1030,7 +1030,8 @@ _ANALYSIS_KEYS = ("node_id", "exec", "artifact", "call_id", "item", "op", "mode"
 def _analysis_entry(entry: dict[str, Any], alias: str, snaps: _Snapshots) -> dict[str, Any]:
    """台账里的一条分析结论 → 写作目录的 A 条目：照抄台账的键，带上从工件取回、复验过哈希的整份结论（result，
    analysis.contract.catalog_result 读它）。取不回、哈希不符、形状不对的不带 result，记 unreadable（以编号开头的一句，引用
-    它时照实说）。rows 是结果表的行数：依据 [[see:A1.r3]] 按它核对，存进报告文档的摘要里也留着。"""
+    它时照实说）。rows 是结果表的行数：依据 [[see:A1.r3]] 按它核对，存进报告文档的摘要里也留着；labels 是结果表记的列名
+    （列 id → 给人看的名字）：证据面板、读屏的出处写「结果表第 1 行「占变化的比重」」，不把列 id 当叫法。"""
    out: dict[str, Any] = {"alias": alias, "kind": LEDGER_KIND, "eid": make_eid(LEDGER_KIND, entry.get("artifact"), {}),
                           "locator": {}, **{k: entry[k] for k in _ANALYSIS_KEYS if k in entry}}
    result, why = snaps.analysis(alias, {"artifact": entry.get("artifact")})
@@ -1039,8 +1040,11 @@ def _analysis_entry(entry: dict[str, Any], alias: str, snaps: _Snapshots) -> dic
    else:
        out["result"] = result
        table = result.get("table")
-        if isinstance(table, dict) and isinstance(table.get("rows"), list):
-            out["rows"] = len(table["rows"])
+        if isinstance(table, dict):
+            if isinstance(table.get("rows"), list):
+                out["rows"] = len(table["rows"])
+            if isinstance(table.get("labels"), dict) and table["labels"]:
+                out["labels"] = {str(k): str(v) for k, v in table["labels"].items()}
    phrase = _clean(str((result or {}).get("phrase") or ""))
    name = method_label(out.get("op"), out.get("mode"))
    out["label"] = f"{name} · {phrase}" if phrase else name
@@ -468,7 +468,8 @@
    const verdict = isObj(step.verdict) ? step.verdict : null
    const cited = isObj(step.cited) ? step.cited : {}
    const citedText = cited.what === 'value' ? T.citedValue(cited.key) : cited.what === 'cell'
-      ? T.citedCell(Number(cited.row) || 0, String(cited.column ?? '')) : T.citedPhrase
+      // 结果表的格写列名（服务端照结果表记的 labels 给），不把列 id 当叫法；老接口没给列名的照旧写 id
+      ? T.citedCell(Number(cited.row) || 0, String(cited.column_label || cited.column || '')) : T.citedPhrase
    body.push(kv([
      [T.dPhrase, String(step.phrase || NONE)],
      step.criterion ? [T.dCriterion, String(step.criterion)] : null,
@@ -561,7 +562,7 @@
      step.node_id ? [T.dNode, nodeLabel(step.node_id)] : null,
      Array.isArray(step.columns) && step.columns.length ? [T.dColumnsUsed, step.columns.join('、')] : null,
      rows ? [T.dRowsUsed, rows] : null,
-      'value' in step ? [ratio ? T.dHeadlineRatio : T.dHeadline, `${step.column ? `第 ${Number(step.row) + 1} 行「${step.column}」 = ` : ''}${valueText(step.value)}`] : null,
+      'value' in step ? [ratio ? T.dHeadlineRatio : T.dHeadline, `${step.column ? `第 ${Number(step.row) + 1} 行「${step.column}」 = ` : ''}${step.value_rendered ?? valueText(step.value)}`] : null,
      closure ? [T.dClosure, closed(closure.headline, closure.total)] : null,
      step.phrase ? [T.dPhrase, String(step.phrase)] : null,
    ])]
@@ -330,6 +330,8 @@ def test_the_doc_keeps_a_summary_and_reloads_the_result_from_the_artifact():
    doc = compose_doc("[[a:A1]]，华东占 [[a:A1.top_share|pct]]。", catalog, loader=store.load)
    stored = doc["catalog"]["A1"]
    assert "result" not in stored and stored["artifact"] and stored["rows"] == 4 and stored["verdict"] == "driver"
+    # 结果表记的列名（列 id → 给人看的名字）跟着摘要走：证据面板、读屏的出处写「结果表第 1 行「变化」」
+    assert stored["labels"]["delta"] == "变化"
    # 文档里的摘要照样解析得了：结论按工件取回
    assert resolve_ref("a:A1", doc["catalog"], loader=store.load)["status"] == "resolved"
    assert verify_doc(doc, doc["catalog"], loader=store.load)["ok"]
@@ -176,7 +176,8 @@ async def test_named_values_and_result_cells_point_at_what_they_cite(client):

    cell = (await client.get(
        f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'v:A1.r0.delta')}")).json()
-    assert cell["chain"][0]["cited"] == {"what": "cell", "row": 0, "column": "delta", "raw": "-120"}
+    assert cell["chain"][0]["cited"] == {"what": "cell", "row": 0, "column": "delta", "column_label": "变化",
+                                         "raw": "-120"}
    assert cell["chain"][2]["table"]["highlight"] == {"cells": [[0, "delta"]]}
    assert cell["note"] == "该值由系统从分析结论的结果表中取出、按固定规则显示，未经模型转写"

@@ -219,7 +220,9 @@ async def test_a_result_cell_rerenders_the_way_it_was_composed(client, monkeypat
    body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{segment['id']}")).json()
    judged = body["chain"][0]
    assert judged["rendered"] == shown and judged["render_ok"] is True
-    assert judged["cited"] == {"what": "cell", "row": 0, "column": by, "raw": snapshot["rows"][0][0]}
+    # column_label 是结果表记的列名：维度列照输入快照里的列名（store、slot），不另起叫法
+    assert judged["cited"] == {"what": "cell", "row": 0, "column": by, "column_label": by,
+                               "raw": snapshot["rows"][0][0]}
    # 导出页的数据块是同一条链
    data = data_block((await client.get(f"/api/runs/{made['run']}/report.html")).text)
    chain = data["chains"][data["refs"][segment["id"]]]
@@ -335,6 +338,9 @@ async def test_headline_inputs_show_the_closure(client):
    assert inputs[1]["closure"] == {"mode": "additive", "total_key": "target_total", "total": "2,026",
                                    "headline": "2,026", "closed": True}
    assert [s["role_label"] for s in inputs[:2]] == ["基期总数", "本期总数"]
+    # 那一格的值服务端渲染好了（render_cell：千分位），和闭合核对里的字同一个写法；页面照原样显示
+    assert inputs[0]["value"] == "2220" and inputs[0]["value_rendered"] == "2,220"
+    assert inputs[1]["value"] == "2026" and inputs[1]["value_rendered"] == "2,026"
    assert inputs[1]["query"] == "Q2" and inputs[1]["sealed"] is True and inputs[2]["query"] == "Q1"
    assert next(s for s in body["chain"] if s["step"] == "analysis_recompute")["status"] == "same"

@@ -20,6 +20,8 @@ import app.bundle
 import app.report_export
 from app.main import app as asgi
 from tests.test_analysis_bundle import bundle_of, card_bundle
+from app.core import artifact_store
+from tests.test_analysis_evidence import SNAPSHOT, analyzed
 from tests.test_analysis_evidence_api import (
    LAST_WEEK, SCOPE, TEXT, THIS_WEEK, comparability_card, ratio_run, sealed_run, seg_id,
 )
@@ -154,7 +156,8 @@ async def test_the_export_page_opens_the_analysis_steps_and_the_scope_numbers():
                 "按运行当天现算的界 date('now', '-7 days') = 2026-10-14（现查 · 2026-10-21T09:00:00）",
                 "orders.refund_date：这一列一个值都没有（现查", "orders.order_date：2025-01-01 至 2026-10-20（剖析"):
        assert text in body, text
-    assert "结果表第 1 行「delta」" in out[cell]["body"]
+    # 结果表的格写结果表记的列名（labels 的「变化」），不把列 id 当叫法
+    assert "结果表第 1 行「变化」" in out[cell]["body"]
    assert out["scope:A1"]["steps"] == opened["steps"]


@@ -169,12 +172,36 @@ async def test_the_export_page_speaks_of_the_overall_ratio_for_a_ratio_decomposi
    out = run_node(RUN_RUNTIME, RUNTIME, {"data": data, "segs": [
        {"seg": phrase, "kind": "analysis", "state": "deterministic", "text": made["result"]["phrase"]}]})
    body = out[phrase]["body"]
-    for text in ("基期整体比率", "整体比率那一格", "整体比率 0.035 与各段分子之和 ÷ 分母之和 0.035 闭合",
+    for text in ("基期整体比率", "整体比率那一格", "第 1 行「rate」 = 0.035",
+                 "整体比率 0.035 与各段分子之和 ÷ 分母之和 0.035 闭合",
                 "整体比率 0.053 与各段分子之和 ÷ 分母之和 0.053 闭合"):
        assert text in body, text
    assert "各段之和 0.0" not in body


+async def test_the_export_page_writes_the_headline_cell_the_way_the_server_rendered_it():
+    """总数那一格的值由服务端渲染（render_cell：千分位），页面照原样显示：以前照原值写「2220」，同一节里闭合核对写
+    「2,220」，一处引用里两个写法并排。"""
+    totals = {"columns": ["gmv_base", "gmv"], "column_types": {"gmv_base": "number", "gmv": "number"},
+              "rows": [[2220, 2026]], "sql": "SELECT 2220 AS gmv_base, 2026 AS gmv", "source": "shop",
+              "truncated": False}
+    t = artifact_store.content_hash(artifact_store.canonical_json(totals))
+    q = artifact_store.content_hash(artifact_store.canonical_json(SNAPSHOT))
+    result = await analyzed({"Q2": (SNAPSHOT, q), "Q5": (totals, t)}, headline=["Q5.r0.gmv_base", "Q5.r0.gmv"])
+    made = await sealed_run("[[a:A1]]。", result=result, queries=[totals])
+    async with AsyncClient(transport=ASGITransport(app=asgi), base_url="http://test") as client:
+        html = (await client.get(f"/api/runs/{made['run']}/report.html")).text
+    data = re.search(r'<script type="application/json" id="report-data">(.*?)</script>', html, re.S).group(1)
+    phrase = seg_id(made["doc"], "a:A1")
+    out = run_node(RUN_RUNTIME, RUNTIME, {"data": data, "segs": [
+        {"seg": phrase, "kind": "analysis", "state": "deterministic", "text": made["result"]["phrase"]}]})
+    body = out[phrase]["body"]
+    for text in ("总数那一格第 1 行「gmv_base」 = 2,220", "总数那一格第 1 行「gmv」 = 2,026",
+                 "总数 2,220 与各段之和 2,220 闭合"):
+        assert text in body, text
+    assert "= 2220" not in body and "= 2026" not in body
+
+
 async def test_the_bundle_page_shows_the_analysis_detail_and_the_scope():
    bundle, _, _ = bundle_of(doc_extra=SCOPE)
    out = run_node(RUN_VERIFIER, VERIFIER, {"data": json.dumps(bundle), "seal": bundle["meta"]["manifest_hash"]})
@@ -262,10 +262,13 @@ export function sourceOf(seg: EvidenceSegment, doc?: Pick<EvidenceDocData, 'cata
  if (!cite) return ''
  const entry = cite.alias ? doc?.catalog?.[cite.alias] : undefined
  if (cite.kind === 'analysis' || entry?.kind === 'analysis') {
-    // 分析结论：判定语（[[a:A1]]）、具名的数（[[a:A1.top_share]]）、结果表的一格（[[v:A1.r0.delta]]，cite 的种类是 cell）
+    // 分析结论：判定语（[[a:A1]]）、具名的数（[[a:A1.top_share]]）、结果表的一格（[[v:A1.r0.delta]]，cite 的种类是 cell）。
+    // 结果表的格写结果表记的列名（目录条目上的 labels），不把列 id 当叫法
    const E = ANALYSIS_TEXT.evidence
    const loc = cite.locator ?? {}
-    const what = cite.kind === 'cell' ? E.citedCell(Number(loc.row) || 0, String(loc.column ?? ''))
+    const column = String(loc.column ?? '')
+    const labels = entry && typeof entry.labels === 'object' && entry.labels ? entry.labels as Record<string, string> : {}
+    const what = cite.kind === 'cell' ? E.citedCell(Number(loc.row) || 0, labels[column] || column)
      : loc.key ? E.citedValue(loc.key) : E.citedPhrase
    return [`${E.title} ${cite.alias ?? ''}`.trim(), what].join(' · ')
  }
@@ -97,7 +97,8 @@ function JudgedPart({ step, recompute, seal, overall }: {
  const verdict = step.verdict && typeof step.verdict === 'object' ? step.verdict : null
  const cited = step.cited ?? {}
  const citedText = cited.what === 'value' ? E.citedValue(String(cited.key ?? ''))
-    : cited.what === 'cell' ? E.citedCell(Number(cited.row) || 0, String(cited.column ?? '')) : E.citedPhrase
+    // 结果表的格写列名（服务端照结果表记的 labels 给），不把列 id 当叫法；老接口没给列名的照旧写 id
+    : cited.what === 'cell' ? E.citedCell(Number(cited.row) || 0, String(cited.column_label || cited.column || '')) : E.citedPhrase
  const refused = step.status === 'refused'
  const trusted = !!overall?.sealed && overall.ok !== false
  const problems = [
@@ -267,7 +268,9 @@ function InputPart({ step, onQuery, trusted }: { step: EvidenceAnalysisStep; onQ
        rows ? [E.rowsUsed, rows] : null,
        'value' in step && !parent ? [ratio ? E.headlineRatio : E.headline, (
          <span key="v" className="tnum">
-            {step.column != null && typeof step.row === 'number' ? `${E.cellAt(step.row, step.column)} = ` : ''}{valueText(step.value)}
+            {step.column != null && typeof step.row === 'number' ? `${E.cellAt(step.row, step.column)} = ` : ''}
+            {/* 那一格的值服务端渲染好了（千分位、NULL 写「—」，和闭合核对同一个写法），照原样显示；老接口没给的按值本身写 */}
+            {step.value_rendered ?? valueText(step.value)}
          </span>
        )] : null,
        closure && closed ? [E.closure, (
@@ -423,7 +423,16 @@
     "op": "decompose",
     "rows": 4,
     "status": "ok",
-     "verdict": "driver"
+     "verdict": "driver",
+     "labels": {
+      "base": "基期",
+      "delta": "变化",
+      "flag": "说明",
+      "pct": "变化率",
+      "region": "地区",
+      "share": "占变化的比重",
+      "target": "本期"
+     }
    },
    "A2": {
     "alias": "A2",
@@ -460,7 +469,14 @@
     "op": "comparable",
     "rows": 5,
     "status": "ok",
-     "verdict": "comparable"
+     "verdict": "comparable",
+     "labels": {
+      "base": "基期",
+      "check": "检查项",
+      "note": "说明",
+      "status": "结果",
+      "target": "本期"
+     }
    },
    "Q1": {
     "alias": "Q1",
@@ -898,7 +914,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -919,7 +936,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -1301,7 +1319,8 @@
       "what": "cell",
       "row": 0,
       "column": "delta",
-       "raw": "-300"
+       "raw": "-300",
+       "column_label": "变化"
      },
      "rendered": "-300",
      "render_ok": true,
@@ -1557,7 +1576,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -1578,7 +1598,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -2212,7 +2233,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -2233,7 +2255,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -3508,7 +3531,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -3529,7 +3553,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -5034,7 +5059,16 @@
     "op": "decompose",
     "rows": 4,
     "status": "ok",
-     "verdict": "driver"
+     "verdict": "driver",
+     "labels": {
+      "base": "基期",
+      "delta": "变化",
+      "flag": "说明",
+      "pct": "变化率",
+      "region": "地区",
+      "share": "占变化的比重",
+      "target": "本期"
+     }
    },
    "Q1": {
     "alias": "Q1",
@@ -5420,7 +5454,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -5441,7 +5476,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -6063,7 +6099,8 @@
       "what": "cell",
       "row": 0,
       "column": "delta",
-       "raw": "-300"
+       "raw": "-300",
+       "column_label": "变化"
      },
      "rendered": "-300",
      "render_ok": true,
@@ -6319,7 +6356,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -6340,7 +6378,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -6722,7 +6761,8 @@
       "what": "cell",
       "row": 0,
       "column": "delta",
-       "raw": "-300"
+       "raw": "-300",
+       "column_label": "变化"
      },
      "rendered": "-300",
      "render_ok": true,
@@ -6978,7 +7018,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -6999,7 +7040,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -8158,7 +8200,8 @@
       "total": "2,700",
       "headline": "2,700",
       "closed": true
-      }
+      },
+      "value_rendered": "2,700"
     },
     {
      "step": "analysis_input",
@@ -8179,7 +8222,8 @@
       "total": "2,360",
       "headline": "2,360",
       "closed": true
-      }
+      },
+      "value_rendered": "2,360"
     },
     {
      "step": "analysis_input",
@@ -2435,8 +2435,16 @@ export interface EvidenceAnalysisStep {
  artifact?: string | null
  eid?: string
  eid_ok?: boolean
-  /** 这一处引用的是结论的哪一项：判定语、具名的数（key）、结果表的一格（row、column） */
-  cited?: { what?: 'phrase' | 'value' | 'cell' | string; key?: string; row?: number; column?: string; raw?: unknown; conv?: string }
+  /** 这一处引用的是结论的哪一项：判定语、具名的数（key）、结果表的一格（row、column；column_label 是结果表的列名，给人看的） */
+  cited?: {
+    what?: 'phrase' | 'value' | 'cell' | string
+    key?: string
+    row?: number
+    column?: string
+    column_label?: string | null
+    raw?: unknown
+    conv?: string
+  }
  rendered?: string
  render_ok?: boolean
  hash_ok?: boolean | null
@@ -2481,6 +2489,8 @@ export interface EvidenceAnalysisStep {
  row?: number
  column?: string
  value?: string | null
+  /** 那一格的值按服务端的规则渲染出来的字（render_cell：千分位、NULL 写「—」），和闭合核对里的写法一致 */
+  value_rendered?: string
  /** 输入落到报告目录的哪个 Q */
  query?: string | null
  sealed?: boolean
@@ -940,7 +940,8 @@ await section('样本页：各种形状', async () => {
      && shareDetail.includes('与按显示规则渲染的结果一致'), shareDetail.replace(/\s+/g, ' ').slice(0, 300))
    const cell = page.locator('.doc p button.num').filter({ hasText: /^-120$/ })
    const cellDetail = await cell.count() ? await detailText(page, cell.first()) : '找不到「-120」'
-    check('analysis：结果表的格：出处写分析结论 A1 的结果表 · 第 1 行 · delta', cellDetail.includes('分析结论 A1 的结果表 · 第 1 行 · delta'),
+    // 列写结果表记的列名（「变化」），不把列 id（delta）当叫法
+    check('analysis：结果表的格：出处写分析结论 A1 的结果表 · 第 1 行 · 变化', cellDetail.includes('分析结论 A1 的结果表 · 第 1 行 · 变化'),
      cellDetail.replace(/\s+/g, ' ').slice(0, 300))
    check('analysis：结果表整表引用：表头是给人看的列名，数字列右对齐', (await page.locator('.doc table th').allTextContents()).join('、') === '地区、基期、本期、变化'
      && await page.locator('.doc td.n button.num').count() === 12, (await page.locator('.doc table th').allTextContents()).join('、'))
@@ -619,6 +619,9 @@ await section('analysis', '分析结论：判定语、具名的数、结果表
    === JSON.stringify(['headline_base', 'headline_target', 'data']))
  check('……总数那一格和各段之和的闭合核对（服务端算好的数）', (await inputs.nth(0).innerText()).includes('总数 2,700 与各段之和 2,700 闭合')
    && await inputs.nth(0).locator('[data-ev-analysis-closure="closed"]').count() === 1)
+  // 那一格的值也由服务端渲染（千分位），和闭合核对里的数同一个写法：原来照原值写「2700」，同一节里 2,700 和 2700 并排
+  check('……那一格的值按服务端的写法显示（千分位，和闭合核对一致）', (await inputs.nth(0).innerText()).includes('第 1 行「gmv」 = 2,700')
+    && !(await inputs.nth(0).innerText()).includes('= 2700'), await inputs.nth(0).innerText())
  check('……数据：用到的列、全部行', (await inputs.nth(2).innerText()).includes('region、week、gmv') && (await inputs.nth(2).innerText()).includes('全部行'))
  check('输入快照的查询步骤接在后面，写明是 A1 的输入', await p.locator('[data-ev-input-of="A1"]').count() === 2)
  await inputs.nth(2).locator('[data-ev-goto-query="Q2"]').click()
@@ -634,9 +637,10 @@ await section('analysis', '分析结论：判定语、具名的数、结果表
    && await p.locator('[data-ev-analysis-value="top_share"][data-ev-cited]').count() === 1)
  await box.locator('[data-seg="s7"]').click()
  await p.locator('[data-ev-analysis-cited="cell"]').waitFor({ timeout: 8000 })
-  check('结果表的格（-300）：引用的是结果表第 1 行「delta」，那一格高亮', (await p.locator('[data-ev-analysis-cited]').innerText()) === '结果表第 1 行「delta」'
+  // 列写结果表记的列名（「变化」），不把列 id（delta）当叫法
+  check('结果表的格（-300）：引用的是结果表第 1 行「变化」，那一格高亮', (await p.locator('[data-ev-analysis-cited]').innerText()) === '结果表第 1 行「变化」'
    && await p.locator('[data-ev-analysis-table] td[data-highlight="cell"]').count() === 1)
-  check('……读屏写出处：分析结论 A1 · 结果表第 1 行「delta」', (await box.locator('[data-seg="s7"]').getAttribute('aria-label') ?? '').includes('分析结论 A1 · 结果表第 1 行「delta」'))
+  check('……读屏写出处：分析结论 A1 · 结果表第 1 行「变化」', (await box.locator('[data-seg="s7"]').getAttribute('aria-label') ?? '').includes('分析结论 A1 · 结果表第 1 行「变化」'))
  check('片段接口按片段 id 取（s3、s9、s7）', ['seg:s3', 'seg:s9', 'seg:s7'].every((h) => hits.includes(h)), hits.join(','))

  // 分析卡的结论：谁做的写分析卡（模板定好的分析）；表格里引用结果表的格