Download the current financial view as a nine-row CSV

README.md

@@ -59,6 +59,25 @@ Unknown, duplicate or malformed view parameters show a notice and the latest
 configured panel. With JavaScript disabled, all six demo panels remain readable.
 The report links to its method, GitHub project and author for further review.
 
+## Download one view for review
+
+Each panel has **Download these metrics · CSV**. It saves only that cutoff/year's
+nine metric rows, including the formula, unit, status, exact fraction, rounded
+display and missing-input names. Switching views or reopening a shared link
+selects the matching panel's download. **All metrics CSV** still contains all
+54 demo rows. Every panel's download also works with JavaScript disabled.
+
+For Excel, use **Data → From Text/CSV**, select UTF-8, and keep the `as_of`,
+`period` and `exact` columns as text to preserve dates, identities and fractions.
+The `display` column is rounded; `exact` retains the saved rational value.
+The CSV uses the same text markers for formula-shaped labels as the full export.
+
+Try the before/after-filing links above, downloading each view. FY2024 before
+filing has nine `missing_input` rows with empty values. After filing,
+`free_cash_flow` retains `108807000000/1`, unit `USD`, status `ok` and the
+`operating_cash_flow - capex` formula. This is a filtered review export;
+full source/lineage review and semantic verification use the complete report.
+
 ## Contracts
 
 | Concern | Behavior |

docs/INTERVIEW.md

@@ -12,6 +12,12 @@
 先指出前一日缺失不等于零，再展开后一日的自由现金流 **108,807 百万美元**及其输入与 SEC 来源。
 这里使用手工整理的公开年报摘录与 SEC 申报日期，不是原样 API 下载或盘中可得时间证明。
 
+再用两分钟练一次实际交付：在两种截止日下分别点 **Download these metrics · CSV**，
+核对每份只有九行，且 `as_of`、`period` 与页面一致。申报前的 `exact`、`display` 留空，
+状态是 `missing_input`；申报后 FCF 为 `108807000000/1` USD，仍保留公式。
+Excel 用“数据 → 从文本/CSV”按 UTF-8 导入，将 `as_of`、`period`、`exact` 设为文本，
+再解释精确值、显示值与缺失值的区别。指标摘录须连同完整报告的事实和血缘继续审阅。
+
 **验证**：独立 SQL 查询逐条对照。修改指标 CSV 并更新哈希，语义复算仍拒绝。测试覆盖未来信息、季度混入、单位不匹配、同日冲突、缺失值与零分母。
 
 追问要点：`fy` 不能代替事实本身的期间；缺失不等于零；增长没有按周数归一化；哈希验证一致性而不证明来源真实性；扩展到其他公司需要补充概念、单位与会计政策映射。

docs/METHODOLOGY.md

@@ -80,3 +80,28 @@ reports should be regenerated into a new directory from their saved facts and
 request before verification with this renderer.
 Pages deployment also waits for this browser job, alongside the Linux/Windows
 test and installed-wheel checks.
+
+## Current-panel metric downloads
+
+Each saved panel embeds a UTF-8 CSV data URL with exactly its metric rows and
+the same nine columns as the full `metrics.csv`. Both use the same CSV serializer
+and text-label contract. Full fractions, negative numbers, formulas, statuses
+and empty missing values remain unchanged; the browser does not scrape displayed
+numbers or recalculate financial values. Filenames contain the validated cutoff
+date; original period identities stay in the rows rather than the filename.
+
+The native download anchor belongs to its panel, so link/history restoration
+and invalid-link fallback expose the corresponding download. Without JavaScript,
+all panels retain their own links. CSV importers may infer cell types; select
+UTF-8 and import `as_of`, `period` and `exact` as text when using Excel.
+This excerpt alone is not a complete replay bundle: source facts, lineage and
+verification remain available from the report's full nine-file inventory.
+
+Unit checks compare every panel's decoded download with its full-export slice,
+including formula-shaped labels, controls/Unicode, quoted newlines, negative
+values and hand-checked before/after-filing results. Real Chrome saves actual
+files, parses their rows independently and compares every column with the full
+export, including after view changes/fallback and without JavaScript. Browser
+checks assert the actual `innerWidth` at desktop/narrow sizes. Only committed
+HTML and its manifest hash are refreshed; seven financial/input files retain
+their original bytes.

examples/demo/manifest.json

@@ -3,7 +3,7 @@
     "analysis.json": "ae9fe1aec6d4b683ca15ec9007ea14eca32d63a4c15f26e88cac5308b43aaac5",
     "decisions.csv": "cebbcc9cc7a7bd25735cedb88fad80151b76f57fffb1bcd687666a75d796bdf3",
     "facts.json": "93d636674792ae8f5cffdbae4c7cf6e27ffa4ac003991fb0976d3e0225c9c8f5",
-    "index.html": "e40a8b859316a9ee4d17744ce15821ff4578fea11051422324a0b22b0bc60701",
+    "index.html": "ae1f9ec959ae1d4f339589d4024e70adcd4fa2646a0ba22feb26dbaa2797642b",
     "lineage.csv": "bba7e396cb6159d1cc786d4c77e288a33e5f62d2b32bedf2c5dca203f852e74f",
     "metrics.csv": "a0d996134f4e68bade395f7cde6bd3e51a53b6bd8723b3bb67e9027413fc08a7",
     "request.json": "042942ccc50dbf51be6f593213f8b9f47276bcd36026a62735946a3f5d1fe3ab",

scripts/check_browser.py

@@ -4,6 +4,8 @@
 CHROMEWEBDRIVER directory supplied by GitHub's Ubuntu runner image.
 """
 
+import csv
+import io
 import json
 import os
 import shutil
@@ -26,6 +28,10 @@
     if (JSON.stringify(values) !== JSON.stringify(expected.metrics)) throw new Error('Financial metrics differ');
     const sources = Array.from(selected[0].querySelectorAll('table')[1].querySelectorAll('a'), a => a.href);
     if (JSON.stringify(sources) !== JSON.stringify(expected.sources)) throw new Error('Source links differ');
+    const download = selected[0].querySelector('.view-csv');
+    if (!download || download.download !== 'statementtrace-metrics-' + expected.as_of + '.csv' ||
+        !download.href.startsWith('data:text/csv;charset=utf-8;base64,'))
+        throw new Error('Missing current-view CSV download');
     const link = document.getElementById('view-link');
     const url = new URL(link.href);
     const params = new URLSearchParams(url.hash.slice(1));
@@ -94,6 +100,8 @@ def call(method, endpoint, payload=None):
         except urllib.error.HTTPError as error:
             raise RuntimeError(error.read().decode()) from error
 
+    downloads = root / "downloads"
+    downloads.mkdir()
     session = None
     process = subprocess.Popen([driver, f"--port={port}"], stdout=subprocess.DEVNULL, stderr=subprocess.STDOUT)
     try:
@@ -114,7 +122,13 @@ def call(method, endpoint, payload=None):
                 "capabilities": {
                     "alwaysMatch": {
                         "browserName": "chrome",
-                        "goog:chromeOptions": {"args": ["--headless=new", "--no-sandbox", "--disable-dev-shm-usage"]},
+                        "goog:chromeOptions": {
+                            "args": ["--headless=new", "--no-sandbox", "--disable-dev-shm-usage"],
+                            "prefs": {
+                                "download.default_directory": str(downloads),
+                                "download.prompt_for_download": False,
+                            },
+                        },
                     }
                 }
             },
@@ -133,19 +147,60 @@ def wait_view(view, *, exact=True, unavailable=False):
                 time.sleep(0.05)
             return execute(CHECK_VIEW, view, exact, unavailable)  # Preserve the actual assertion diagnostic.
 
-        counts = dict(visits=0, selections=0, metric_rows=0, round_trips=0, history_checks=0, invalid_links=0)
+        def download_view(view, selector="section:not([hidden]) .view-csv"):
+            path = downloads / f"statementtrace-metrics-{view['as_of']}.csv"
+            path.unlink(missing_ok=True)
+            element = call("POST", f"/session/{session}/element", {"using": "css selector", "value": selector})[
+                "element-6066-11e4-a52e-4f735466cecf"
+            ]
+            call("POST", f"/session/{session}/element/{element}/click", {})
+            for _ in range(100):
+                if path.is_file():
+                    break
+                time.sleep(0.05)
+            else:
+                raise AssertionError("Chrome did not save the view CSV")
+            raw = path.read_bytes()
+            rows = list(csv.DictReader(io.StringIO(raw.decode("utf-8"), newline=""), strict=True))
+            if rows != view["csv_rows"]:
+                raise AssertionError("Downloaded CSV differs from the selected full-export slice")
+            counts["downloads"] += 1
+            counts["downloaded_metric_rows"] += len(rows)
+
+        counts = dict(
+            visits=0,
+            selections=0,
+            metric_rows=0,
+            round_trips=0,
+            history_checks=0,
+            invalid_links=0,
+            downloads=0,
+            downloaded_metric_rows=0,
+        )
+        viewport_widths = []
         for width in (1280, 390):
             call("POST", f"/session/{session}/window/rect", {"width": width, "height": 1000})
-            for case in cases:
+            for case_index, case in enumerate(cases):
                 base = (root / case["file"]).as_uri() + "?review=1"
                 navigate(base)
+                actual_width = execute("return window.innerWidth;")
+                if actual_width != width:
+                    raise AssertionError(f"Requested width {width}, observed {actual_width}")
+                if case_index == 0:
+                    viewport_widths.append(actual_width)
                 counts["visits"] += 1
                 latest = case["views"][-1]
                 wait_view(latest, exact=False)
                 for index, view in enumerate(case["views"]):
                     execute(SELECT_VIEW, view, "cutoff" if index % 2 else "period")
                     counts["selections"] += 1
                     counts["metric_rows"] += len(view["metrics"])
+                download_view(latest)
+                before = next(
+                    v for v in case["views"] if v["as_of"] == "2024-10-31" and v["period"] == latest["period"]
+                )
+                execute(SELECT_VIEW, before, "cutoff")
+                download_view(before)
                 first = case["views"][0]
                 shared = execute(SELECT_VIEW, first, "period")
                 navigate(base)
@@ -177,6 +232,8 @@ def wait_view(view, *, exact=True, unavailable=False):
                 ]
                 call("POST", f"/session/{session}/element/{element}/click", {})
                 wait_view(latest)
+                if case_index == 0:
+                    download_view(latest)
 
         call(
             "POST",
@@ -188,6 +245,7 @@ def wait_view(view, *, exact=True, unavailable=False):
             visible = execute("return document.querySelectorAll('section:not([hidden])').length;")
             if visible != len(cases[0]["views"]) or not execute("return document.getElementById('share-view').hidden;"):
                 raise AssertionError("No-JavaScript view must retain every saved panel")
+            download_view(cases[0]["views"][0], "section .view-csv")
         finally:
             call(
                 "POST",
@@ -200,7 +258,7 @@ def wait_view(view, *, exact=True, unavailable=False):
                     "status": "passed",
                     "browser": "Chrome",
                     "reports": len(cases),
-                    "viewports": [1280, 390],
+                    "viewports": viewport_widths,
                     "no_js_panels": visible,
                     **counts,
                 }
@@ -220,13 +278,14 @@ def wait_view(view, *, exact=True, unavailable=False):
 def main():
     from statementtrace.cli import demo_inputs
     from statementtrace.engine import analyze
-    from statementtrace.report import build
+    from statementtrace.report import build, csv_label
 
     pairs = [
         ["2023", "2024"],
         ["sample\rone", "sample\none"],
         ["sample\x00one", "sample�one"],
         [' 年报 +&%#?/ " 🧾 ', "\tAnnual <2024>\n"],
+        ["=1+1", "text: =1+1"],
     ]
     with tempfile.TemporaryDirectory() as temporary:
         root = Path(temporary).resolve()
@@ -236,13 +295,22 @@ def main():
             for row, label in zip(request["periods"], labels, strict=True):
                 row["label"] = label
             filename = f"report-{index}.html"
-            (root / filename).write_bytes(build(data, request)["index.html"])
+            outputs = build(data, request)
+            (root / filename).write_bytes(outputs["index.html"])
+            all_metrics = list(
+                csv.DictReader(io.StringIO(outputs["metrics.csv"].decode("utf-8"), newline=""), strict=True)
+            )
             views = [
                 {
                     "as_of": panel["as_of"],
                     "period": panel["period"]["label"],
                     "metrics": [[m[key] for key in ("metric", "display", "unit", "status")] for m in panel["metrics"]],
                     "sources": [fact["source_url"] for fact in panel["facts"].values() if fact is not None],
+                    "csv_rows": [
+                        row
+                        for row in all_metrics
+                        if row["as_of"] == panel["as_of"] and row["period"] == csv_label(panel["period"]["label"])
+                    ],
                 }
                 for panel in analyze(data, request)["panels"]
             ]

src/statementtrace/report.py

@@ -1,5 +1,6 @@
 """Self-contained HTML, CSV, and semantic replay of a deterministic report."""
 
+import base64
 import csv
 import html
 import io
@@ -20,6 +21,8 @@
     "index.html",
 }
 
+METRIC_FIELDS = ("as_of", "period", "metric", "formula", "unit", "status", "exact", "display", "missing")
+
 VIEW_SCRIPT = r"""
 const cutoff = document.querySelector('#cutoff');
 const period = document.querySelector('#period');
@@ -108,6 +111,18 @@ def csv_label(value):
     return value
 
 
+def metric_rows(panel):
+    return [
+        {
+            "as_of": panel["as_of"],
+            "period": panel["period"]["label"],
+            **{key: metric[key] for key in ("metric", "formula", "unit", "status", "exact", "display")},
+            "missing": ";".join(metric["missing"]),
+        }
+        for metric in panel["metrics"]
+    ]
+
+
 def render(analysis):
     esc = html.escape
     panels = []
@@ -135,12 +150,17 @@ def render(analysis):
             )
         label, cutoff = esc(panel["period"]["label"]), esc(panel["as_of"])
         period_key = esc(json.dumps(panel["period"]["label"], ensure_ascii=True), quote=True)
+        metric_csv = base64.b64encode(csv_bytes(metric_rows(panel), METRIC_FIELDS)).decode("ascii")
         panels.append(
             f"<section data-period='{period_key}' data-cutoff='{cutoff}'>"
             f"<h2>FY {label} · as of {cutoff}</h2><p>"
             f"{esc(panel['period']['start'])} → {esc(panel['period']['end'])} · "
             f"{panel['duration_days']} inclusive days · amounts in USD</p>"
-            "<h3>Financial metrics / 财务指标</h3><div class='scroll'><table>"
+            "<h3>Financial metrics / 财务指标</h3>"
+            f"<p><a class='view-csv' href='data:text/csv;charset=utf-8;base64,{metric_csv}' "
+            f"download='statementtrace-metrics-{cutoff}.csv'>Download these metrics · CSV</a>"
+            " · Includes this cutoff/year, formulas, exact values and missing-input statuses.</p>"
+            "<div class='scroll'><table>"
             "<thead><tr><th>Metric</th><th>Value</th><th>Unit</th><th>Status</th>"
             "<th>Formula and lineage</th></tr></thead><tbody>" + "".join(metrics) + "</tbody></table></div>"
             "<h3>Selected facts / 原始数据</h3><div class='scroll'><table>"
@@ -191,7 +211,7 @@ def options(values, *, identities=False):
         "<p id='view-status' role='status'></p>"
         "<noscript><p>JavaScript is off; all saved cutoff/year panels are shown below.</p></noscript>"
         "<p><a href='selected_facts.csv'>Facts CSV</a> · "
-        "<a href='metrics.csv'>Metrics CSV</a> · <a href='lineage.csv'>Full lineage</a> · "
+        "<a href='metrics.csv'>All metrics CSV</a> · <a href='lineage.csv'>Full lineage</a> · "
         "<a href='decisions.csv'>Selection decisions</a> · "
         "<a href='manifest.json'>SHA-256 manifest</a></p></div>"
         + "".join(panels)
@@ -225,14 +245,8 @@ def build(data, request):
                     "source_url": fact["source_url"] if fact else "",
                 }
             )
+        metrics.extend(metric_rows(panel))
         for metric in panel["metrics"]:
-            metrics.append(
-                {
-                    **base,
-                    **{key: metric[key] for key in ("metric", "formula", "unit", "status", "exact", "display")},
-                    "missing": ";".join(metric["missing"]),
-                }
-            )
             lineage.extend({**base, "metric": metric["metric"], **item} for item in metric["inputs"])
     files = {
         "facts.json": canonical(data),