Desktop workbench overhaul: toolbar options, input/constants/result redesign, design-token pass

[yilibinbin]

Summary

Desktop GUI overhaul for DataLab's workbench: toolbar-driven options, a redesigned
input/constants/result layout, on-demand LaTeX, a full design-review token pass, plus a
batch of UX polish and bug fixes accumulated on this branch. One Python core, two frontends
— desktop changes only; the web app and compute layers are untouched.

Highlights

Layout / workbench

• Two-pane workbench (config+data | result) with a result-heavy ~1:3 default ratio.
• Input data + constants merged into 输入数据 / 常数 sheet tabs; each tab self-contained.
• Input-area expand/collapse toggle with a smooth 220ms splitter animation.
• Left column unified into [输入数据] + [one config card] per mode; mode selector on the toolbar.
• Result overview → compact clickable toolbar status chip; history moved to a toolbar popup.

Input / constants UX

• Removed the 使用数据文件 checkbox — a non-empty file path now takes precedence over the
  manual editor (which greys/hides live when a path is entered). Same for constants.
• Constants card gained a 输入常数 title, an N 行 row-count, and a ? help button (matching the
  data card); both input cards now carry a consistent border + padding.
• Excel-like block copy (Ctrl/Cmd+C → TSV) in the data/constants tables.

LaTeX / result

• On-demand 生成 TeX / 预览 PDF (no generate-checkbox); tex-rebuild inputs persist in the
  workspace so it works after reopen without recomputing.
• 不确定度位数 moved to the result panel with live re-render.
• Fitting model line honours the digit / scientific toggles; long compile-error dialogs no longer
  push OK off-screen; result font-size control live-updates.

Feature removal

• Removed the 单位标注 (units) feature end-to-end on the UI side (not general enough); the
  display-only backend is left as harmless dead code and old workspaces still open.

Design-review token pass (theme.py)

• Converged 3–4 near-duplicate hexes per role into semantic tokens
  (text_primary/text_muted/border/card_bg/region_bg/surface_raised/surface_hover).
• Radius collapsed to a 2-tier scale (RADIUS_CARD=8 / RADIUS_CONTROL=6); one canonical
  CARD_PADDING; formula-preview border fixed; fixed-white tutorial card text pinned dark.

Precision / workspace correctness

• Lossless mpf workspace serialization (was two-sided precision loss on reopen).
• Guarded workspace capture against missing data/constants files.

Review gates already passed on this branch

• Three-model serial adversarial review (Claude → Codex → Gemini) across multiple rounds;
  all confirmed findings fixed (mpf precision, theme-toggle stale styles, animation-fight,
  tutorial contrast, file-precedence live feedback, …). Gemini's agy channel timed out on the
  last two rounds — its lens was covered by direct code inspection and noted as such.
• /simplify 4-agent cleanup pass (reuse / simplification / efficiency / altitude): radius
  convergence finished, per-keystroke constants refresh gated, dead widget removed.
• Iterative user acceptance on the live GUI throughout.

Test plan

• QT_QPA_PLATFORM=offscreen pytest -q — full suite green (a couple of files run separately due
  to an offscreen Qt-teardown hang unrelated to these changes).
• Per-round targeted suites (theme tokens/spacing, two-pane layout, input/constants tabs,
  constants editor, formula preview, workspace controller, option reachability) all pass.
• Regression tests added for: file-precedence dimming, expand/collapse restore, formula-preview
  pixmap inset, constants row-count + control-row placement, theme-toggle restyle, mpf round-trip.

Requesting

CodeRabbit review before merge to main.

Summary by CodeRabbit

• New Features
  • On-demand LaTeX generation with a non-modal preview dialog (TeX source + compiled PDF).
  • History panel now collapses by default, with a toolbar popup; added a compact result overview popover and an always-visible result status strip.
  • Desktop toolbar option controls now open real non-modal dialogs.
• Bug Fixes
  • Excel-like block copy (Ctrl/Cmd+C → TSV) for table selections; improved formula preview border visibility and refreshed embedded constants/editor styling.
  • Web: fixed docs TOC heading anchors; added rate-limiting for heavy compute POST requests.
• Refactor
  • Merged two-pane workbench layout with improved resize/height behavior.
  • LaTeX numeric digit grouping supports native vs app-side grouping (engine-adaptive where available); removed TeX/PDF result tabs.

[coderabbitai]

Actionable comments posted: 6

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (3)

app_desktop/history_panel.py (1)

118-166: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Reapply the collapse state after language changes _apply_language() rewrites title_label from the registered bare title, so the ▸/▾ prefix is lost on translation switches. Call _apply_history_collapsed() from the language-change path, or keep the arrow out of the registered title text.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/history_panel.py` around lines 118 - 166, The collapse indicator
is being overwritten when `_apply_language()` refreshes `title_label`, so the
▸/▾ state is lost after a language switch. Update the language-change flow in
`HistoryPanel` to re-run `_apply_history_collapsed()` after
`_register_texts()`/title rewrites, or change `title_label` registration so it
does not include the arrow prefix; use `toggle_history_collapsed()`,
`set_history_collapsed()`, and `_apply_history_collapsed()` as the hooks to keep
the collapsed state and label in sync.

statistics_utils.py (2)

172-185: 🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Forward native_group_width in the statistics worker path.

generate_statistics_latex() defaults this to True, so app_desktop/workers_core.py will emit digit-group-size whenever grouped, non-dcolumn LaTeX is generated. The UI path already passes self._engine_supports_group_width(), but the worker path does not, so bundled Tectonic can still hit the unsupported key in background runs.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@statistics_utils.py` around lines 172 - 185, The statistics worker path is
not forwarding native_group_width into generate_statistics_latex, so grouped
non-dcolumn LaTeX can still emit digit-group-size for bundled Tectonic. Update
the worker-side call in workers_core.py to pass the engine capability check
result, using the same _engine_supports_group_width logic already used by the UI
path, so generate_statistics_latex only enables native grouping when supported.

---

496-519: 🎯 Functional Correctness | 🟠 Major | 🏗️ Heavy lift

Propagate the grouping fallback to the remaining statistics writers

generate_statistics_bootstrap_latex, generate_statistics_time_series_latex, and generate_statistics_hypothesis_latex still rely on siunitx for digit grouping; when native_group_width=False and latex_group_size != 3, they fall back to siunitx’s default 3-digit grouping instead of the requested width. Mirror the app-side fallback used by generate_statistics_latex/generate_statistics_latex_batches, and thread native_group_width through the worker callsite too if it still defaults to True.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@statistics_utils.py` around lines 496 - 519, The remaining statistics LaTeX
writers still ignore the requested grouping width when native_group_width is
disabled. Update generate_statistics_bootstrap_latex,
generate_statistics_time_series_latex, and generate_statistics_hypothesis_latex
to use the same app-side grouping fallback as generate_statistics_latex and
generate_statistics_latex_batches instead of relying on siunitx’s default
3-digit grouping, and make sure the worker callsite passes native_group_width
through where these generators are invoked. Use the existing symbols
_statistics_latex_preamble, latex_group_size, and native_group_width to keep the
behavior consistent.

🧹 Nitpick comments (15)

app_desktop/result_overview_popover.py (1)

90-90: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider using CARD_PADDING for layout consistency.

workbench_results.py was just refactored to use CARD_PADDING from theme.py, but this new popover uses hardcoded 12, 10, 12, 10. If the design-token pass is meant to be consistent across result-overview surfaces, consider importing and using CARD_PADDING here too (or a popover-specific token if the different padding is intentional).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/result_overview_popover.py` at line 90, The popover layout is
still using hardcoded contents margins instead of the shared design token, which
breaks consistency with the recent padding refactor. Update the result-overview
popover setup in the relevant layout initialization to use CARD_PADDING from
theme.py, or a dedicated popover padding token if the spacing is intentionally
different, and make sure the import and margin assignment stay aligned with the
existing card-style surfaces.

app_desktop/window_statistics_mixin.py (2)

288-349: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Duplicate single/multi-batch LaTeX dispatch logic between run-time and on-demand paths.

generate_statistics_latex_on_demand (Lines 319-346) re-implements the same single-vs-multi-batch generate_statistics_latex/generate_statistics_latex_batches argument-building already in _run_statistics_mode (Lines 506-533). Both blocks were updated together for native_group_width this time, but future option additions risk drifting between the two copies.

Consider extracting a shared helper, e.g. _write_statistics_latex(display_batches, rows, sigma_rows, digits, output_path, use_dcolumn, group_size, native, value_col_joined, caption), used by both call sites.

Also applies to: 504-535

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/window_statistics_mixin.py` around lines 288 - 349, The
single/multi-batch LaTeX dispatch is duplicated between
generate_statistics_latex_on_demand and the runtime path in
_run_statistics_mode, so future argument changes can drift. Extract the shared
single-vs-batch decision and generate_statistics_latex /
generate_statistics_latex_batches argument assembly into one helper, then call
it from both generate_statistics_latex_on_demand and _run_statistics_mode,
keeping native_group_width, caption, group_size, and value_col_joined handling
in one place.

---

318-318: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Repeated engine-capability lookup expression across 8+ sites.

self._engine_supports_group_width() if hasattr(self, "_engine_supports_group_width") else True is copy-pasted at every LaTeX generation call site in this file. Extracting a small helper (e.g. self._native_group_width_flag()) would centralize this so future changes to the fallback/probing logic don't need to be replicated everywhere.

♻️ Sketch

+    def _native_group_width_flag(self) -> bool:
+        return self._engine_supports_group_width() if hasattr(self, "_engine_supports_group_width") else True
+
     def generate_statistics_latex_on_demand(self) -> str | None:
         ...
-        native = self._engine_supports_group_width() if hasattr(self, "_engine_supports_group_width") else True
+        native = self._native_group_width_flag()

Also applies to: 519-519, 532-532, 641-641, 743-743, 1058-1058, 1289-1289, 1424-1424

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/window_statistics_mixin.py` at line 318, Repeated
engine-capability probing for the group-width flag is duplicated across many
LaTeX generation call sites in window_statistics_mixin.py. Add a small helper on
the same class, such as a private method like _native_group_width_flag(), that
encapsulates the hasattr(self, "_engine_supports_group_width") fallback logic
and returns the computed boolean. Then update the call sites that currently
inline self._engine_supports_group_width() if hasattr(self,
"_engine_supports_group_width") else True to use the new helper instead, keeping
the behavior identical while centralizing the logic.

app_desktop/window_extrapolation_mixin.py (1)

827-831: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Cross-module import of private helpers.

_input_units_for_headers/_result_unit_from_units are underscore-prefixed (private) helpers from .workers_core, imported directly into this UI mixin. Consider exposing them as public helpers (or moving to a shared units-utils module) to avoid coupling UI code to workers_core's private surface.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/window_extrapolation_mixin.py` around lines 827 - 831, The UI
mixin is importing underscore-prefixed private helpers from workers_core,
creating tight cross-module coupling; update window_extrapolation_mixin to stop
relying on _input_units_for_headers and _result_unit_from_units directly. Move
the shared unit conversion/header logic into a public helper location or a
dedicated units-utils module, then import those public symbols here and keep
workers_core's private surface internal.

app_desktop/result_status_strip.py (1)

17-17: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider exposing _overview_state/_status_badge as public API.

Importing underscore-prefixed private functions from workbench_results creates a fragile cross-module coupling. If these are stable internal contracts shared within the package, consider dropping the underscore prefix or providing public wrapper functions.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/result_status_strip.py` at line 17, The import in
result_status_strip is reaching into underscore-prefixed private helpers from
workbench_results, which creates a fragile internal dependency. Update
workbench_results to expose _overview_state and _status_badge through a stable
public API, either by renaming them to public symbols or adding public wrapper
functions, and then switch result_status_strip to import and use those public
names instead.

datalab_latex/latex_formatting.py (1)

636-679: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Consider extracting the shared digit-grouping loop.

The int-part (reversed) and frac-part (forward) loops both re-implement "insert sep every group_size digits" — same pattern already present in add_spacing_to_number above. A small shared helper (_group_digits(digits, group_size, sep, reverse)) would cut duplication across both call sites.

♻️ Suggested consolidation

+def _group_digit_string(digits: str, group_size: int, sep: str, *, from_left: bool) -> str:
+    seq = digits if from_left else digits[::-1]
+    chars: list[str] = []
+    for i, ch in enumerate(seq):
+        if i > 0 and i % group_size == 0:
+            chars.append(sep)
+        chars.append(ch)
+    grouped = "".join(chars)
+    return grouped if from_left else grouped[::-1]
+
 def group_digits_both_sides(number_str: str, group_size: int, sep: str = "\\,") -> str:
     ...
-    grouped_int_chars: list[str] = []
-    for i, ch in enumerate(reversed(int_part)):
-        if i > 0 and i % group_size == 0:
-            grouped_int_chars.append(sep)
-        grouped_int_chars.append(ch)
-    grouped_int = "".join(reversed(grouped_int_chars))
+    grouped_int = _group_digit_string(int_part, group_size, sep, from_left=False)
     result = (sign or "") + grouped_int
     if frac_part is not None:
-        grouped_frac_chars: list[str] = []
-        for i, ch in enumerate(frac_part):
-            if i > 0 and i % group_size == 0:
-                grouped_frac_chars.append(sep)
-            grouped_frac_chars.append(ch)
-        result += "." + "".join(grouped_frac_chars)
+        result += "." + _group_digit_string(frac_part, group_size, sep, from_left=True)
     return result + (tail or "")

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@datalab_latex/latex_formatting.py` around lines 636 - 679, The digit-grouping
logic in group_digits_both_sides duplicates the same “insert sep every
group_size digits” pattern used elsewhere, so extract that shared behavior into
a small helper (for example a private digit-grouping function) and reuse it for
both the reversed integer-part loop and the forward fractional-part loop. Keep
group_digits_both_sides focused on parsing sign/int/frac/tail and delegating to
the helper, and align the new helper with the existing add_spacing_to_number
implementation so the grouping behavior stays consistent.

app_desktop/theme.py (2)

258-287: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Formula preview dark background still hardcodes #1f2328 in two places.

formula_preview_surface_style and formula_preview_source_edit_style both special-case "#1f2328" for the dark background rather than resolving it through _tok(), duplicating the exact kind of per-function hex literal this token pass was introduced to eliminate. Consider adding a token role (e.g. formula_bg) so the two functions can't drift from each other.

♻️ Suggested fix

 _TOKENS: dict[str, tuple[str, str]] = {
     ...
     "surface_hover": ("`#eef2f7`", "`#303746`"),  # button/tab hover
+    "formula_bg": ("`#ffffff`", "`#1f2328`"),      # formula preview / source edit background
 }

 def formula_preview_surface_style(*, dark: bool | None = None) -> str:
     dark = is_dark_theme() if dark is None else bool(dark)
-    background = "`#1f2328`" if dark else _tok("card_bg", dark)
+    background = _tok("formula_bg", dark)

 def formula_preview_source_edit_style(*, dark: bool | None = None) -> str:
     dark = is_dark_theme() if dark is None else bool(dark)
-    background = "`#1f2328`" if dark else _tok("card_bg", dark)
+    background = _tok("formula_bg", dark)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/theme.py` around lines 258 - 287, The formula preview styles
still duplicate a hardcoded dark background value, so update
`formula_preview_surface_style` and `formula_preview_source_edit_style` to
resolve the background through `_tok()` instead of embedding `#1f2328`. Add or
reuse a dedicated token role such as `formula_bg` in the theme token lookup so
both helpers share the same source of truth and cannot drift apart.

---

441-452: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Tab-corner radius hardcodes 6px instead of RADIUS_CONTROL.

Same function uses {RADIUS_CARD}px for the pane border a few lines up (line 437); the tab corner radii at lines 449-450 are literal 6px even though RADIUS_CONTROL = 6 exists for exactly this purpose.

♻️ Suggested fix

-    border-top-left-radius: 6px;
-    border-top-right-radius: 6px;
+    border-top-left-radius: {RADIUS_CONTROL}px;
+    border-top-right-radius: {RADIUS_CONTROL}px;

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/theme.py` around lines 441 - 452, The tab corner radius in the
input data tabs style is hardcoded instead of using the shared control radius
constant. Update the `build_theme`/theme stylesheet section that defines
`QTabWidget#input_data_tabs QTabBar::tab` to replace the literal corner radius
values with `RADIUS_CONTROL` (consistent with the nearby `RADIUS_CARD` usage),
so the tab corners follow the same theme token as other control elements.

docs/SWARM_REVIEW_2026.md (1)

262-269: 📐 Maintainability & Code Quality | 🔵 Trivial

Self-flagged truncation should be resolved before merge.

The doc explicitly notes the __main__/SocketIO finding's recommendation text was truncated in the source JSON and only "reasonably completed" by the report author, asking to "核对原始 finding" before relying on it.

Want me to help track this down as a follow-up so the recommendation text is verified against the original finding before this report is treated as final?

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/SWARM_REVIEW_2026.md` around lines 262 - 269, The issue is that the
recommendation text for the `__main__` SocketIO finding was truncated and only
reconstructed from context, so it should not be treated as final. Verify the
original finding source and replace the placeholder wording in this SWARM review
entry with the exact recommendation text, keeping the `__main__`/SocketIO note
consistent with the rest of the report.

statistics_utils.py (1)

184-257: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Duplicated app_group/_maybe_group helper.

The same 6-line closure (app_group = ...; _maybe_group) is copy-pasted verbatim between generate_statistics_latex (196-201) and generate_statistics_latex_batches (341-346). Given this PR's own stated rationale elsewhere for centralizing near-duplicate siunitx logic ("5 near-duplicate inline copies... removing them all keeps the format drift-free"), the same argument applies here.

♻️ Extract a shared module-level helper

+def _maybe_group_cell(cell: str, app_group: bool, group_size: int) -> str:
+    from datalab_latex.latex_formatting import group_digits_both_sides
+    if app_group and "\\multicolumn" not in cell and "\\text" not in cell:
+        return "\\text{" + group_digits_both_sides(cell, group_size) + "}"
+    return cell

Then in both generate_statistics_latex and generate_statistics_latex_batches, replace the local closure with functools.partial(_maybe_group_cell, app_group=app_group, group_size=group_size) (or just call _maybe_group_cell(cell, app_group, group_size) directly at each call site).

Also applies to: 334-399

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@statistics_utils.py` around lines 184 - 257, The duplicated
app_group/_maybe_group closure should be removed and centralized so the siunitx
grouping behavior stays consistent across both statistics table generators.
Extract the shared grouping logic into a module-level helper, then update
generate_statistics_latex and generate_statistics_latex_batches to use that
helper instead of defining the same local closure twice; keep the existing
behavior for app_group, group_size, and _maybe_group call sites unchanged.

datalab_latex/sisetup_block.py (1)

26-53: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Docstring not updated for the new emit_digit_group_size parameter.

The Parameters section still only documents group_size and include_dcolumn; the new tri-state emit_digit_group_size (probe result vs. legacy heuristic) isn't described, even though it's a public function used across five call sites.

📝 Add a Parameters entry

     include_dcolumn:
         When ``True`` the document uses dcolumn for alignment; siunitx
         must NOT emit grouping characters or the alignment breaks. The
         block emits the minimal "no grouping" body in that case.
+    emit_digit_group_size:
+        ``True`` -> engine was probed and honours the key; emit unguarded.
+        ``False`` -> probed as unsupported; never emit.
+        ``None`` (default) -> no probe result; fall back to the legacy
+        ``\`@ifpackagelater``` date heuristic.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@datalab_latex/sisetup_block.py` around lines 26 - 53, Update the docstring
for build_sisetup_block to document the new emit_digit_group_size parameter
alongside group_size and include_dcolumn; add a Parameters entry explaining the
tri-state behavior (True emits digit-group-size, False omits it, None falls back
to the legacy heuristic) so the public API matches the implementation used by
the call sites.

app_desktop/table_copy.py (1)

35-42: 🎯 Functional Correctness | 🔵 Trivial | 💤 Low value

No escaping for embedded tabs/newlines in cell text.

If any cell's text() contains a literal tab or newline, the TSV grid produced here will misalign columns/rows on paste. Low risk for numeric data but worth a defensive check if free-text columns are ever copyable through this path.

🧹 Optional: sanitize cell text before joining

         for col in range(left, right + 1):
             item = table.item(row, col)
-            cells.append(item.text() if item is not None else "")
+            text = item.text() if item is not None else ""
+            cells.append(text.replace("\t", " ").replace("\n", " "))

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/table_copy.py` around lines 35 - 42, The TSV export in the table
copy path does not escape embedded tabs or newlines in cell text, so pasted data
can misalign rows and columns. Update the copy routine that builds lines from
table.item(...).text() so each cell value is sanitized before joining, replacing
or escaping literal tab/newline characters consistently. Keep the fix localized
to the table copy helper that appends cells and writes to
QApplication.clipboard().setText.

app_desktop/workbench_variable_panel.py (2)

259-270: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Clean up dead "constants" role from _variable_summary_text.

Since _mounts_in_panel_order no longer includes constants, no section will ever have role "constants". The "constants" key in counts and its entry in the iteration tuple are dead code.

♻️ Proposed cleanup

-    counts: dict[str, int] = {"parameters": 0, "constants": 0, "unknowns": 0}
+    counts: dict[str, int] = {"parameters": 0, "unknowns": 0}

-    for role in ("parameters", "unknowns", "constants"):
+    for role in ("parameters", "unknowns"):

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/workbench_variable_panel.py` around lines 259 - 270, The
`_variable_summary_text` helper still carries dead handling for the
`"constants"` role even though `_mounts_in_panel_order` no longer produces it.
Remove the unused `"constants"` entry from the `counts` map and drop
`"constants"` from the role iteration in `_variable_summary_text`, keeping only
the active roles so the summary logic matches the current section roles.

---

218-218: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Remove stale spec.constants from has_variables check.

Constants are no longer mounted in the variable panel (per _mounts_in_panel_order), so checking spec.constants here is semantically misleading. If a mode has only constants, has_variables is True but no sections are visible — the panel is hidden anyway after unnecessary iteration.

♻️ Proposed cleanup

-    has_variables = bool(spec and (spec.parameters or spec.tables or spec.constants))
+    has_variables = bool(spec and (spec.parameters or spec.tables))

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/workbench_variable_panel.py` at line 218, Update the
`has_variables` check in `workbench_variable_panel.py` to stop considering
`spec.constants`, since constants are no longer mounted by
`_mounts_in_panel_order`. Keep the visibility logic aligned with the actual
mounted sections by checking only the active variable sources used by the panel
(for example, parameters and tables), so `has_variables` reflects what can
really be shown and avoids the extra empty iteration path.

app_desktop/constants_editor.py (1)

146-168: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Deduplicate refresh_theme_style and set_embedded_in_workbench.

Both methods end with the identical setStyleSheet(constants_editor_style(embedded=...)) + unpolish/polish sequence. Have set_embedded_in_workbench delegate to refresh_theme_style instead of repeating it.

♻️ Proposed dedup

     def set_embedded_in_workbench(self, embedded: bool) -> None:
         embedded = bool(embedded)
         self.setProperty("datalab_constants_embedded", embedded)
         layout = self.layout()
         if layout is not None:
             # Embedded card now has its own border (like the data card) → pad content off it with
             # the shared CARD_PADDING; standalone keeps its tighter 8px inset.
             if embedded:
                 layout.setContentsMargins(*CARD_PADDING)
             else:
                 layout.setContentsMargins(8, 8, 8, 8)
-        self.setStyleSheet(constants_editor_style(embedded=embedded))
-        self.style().unpolish(self)
-        self.style().polish(self)
+        self.refresh_theme_style()

     def refresh_theme_style(self) -> None:
         """Re-apply the (theme-dependent) editor style for the current embedded state. Its style is
         otherwise set once at construction/embedding, so a live light↔dark toggle would leave the
         button colors stale — the theme refresh calls this."""
         embedded = bool(self.property("datalab_constants_embedded"))
         self.setStyleSheet(constants_editor_style(embedded=embedded))
         self.style().unpolish(self)
         self.style().polish(self)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/constants_editor.py` around lines 146 - 168, The style refresh
logic is duplicated between set_embedded_in_workbench and refresh_theme_style in
ConstantsEditor. Refactor so set_embedded_in_workbench only updates the embedded
property and layout margins, then delegates to refresh_theme_style for the
shared setStyleSheet/unpolish/polish sequence. Keep refresh_theme_style as the
single place that reads datalab_constants_embedded and reapplies
constants_editor_style.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@app_desktop/fitting_latex_writer.py`:
- Around line 115-131: The fallback grouping in fitting_latex_writer.py is
wrapping grouped values in \text{...}, which breaks rendering when app_group
emits them into a plain r column. Update the _maybe_group helper inside the
formatting function so it returns the result of group_digits_both_sides(cell,
_group) directly, keeping the existing multicolumn/text guards but removing the
extra \text wrapper.

In `@app_desktop/window_extrapolation_mixin.py`:
- Around line 728-867: The on-demand LaTeX generator methods can raise directly
and bypass the existing warning flow. Add try/except handling around
generate_root_latex_on_demand, generate_extrapolation_latex_on_demand, and
generate_error_latex_on_demand (or in
generate_latex_for_current_result/open_latex_preview) so malformed stashes or
bad paths are caught and reported via the same QMessageBox.warning path used by
_write_root_latex_if_requested, using the existing method names to keep the fix
localized.

In `@app_desktop/window_fitting_residuals_mixin.py`:
- Around line 518-576: The on-demand LaTeX regeneration helpers are creating a
fresh temporary file on every call via latex_output_path_for_run(True), which
leaks temp files across repeated refreshes. Update
generate_fitting_latex_on_demand and the other generate_*_latex_on_demand
methods it references to reuse or delete the previous per-kind temp path before
creating a new one, storing the last on-demand path on the instance and
unlinking it safely before requesting a replacement. Keep the fix localized
around latex_output_path_for_run, _load_latex_into_editor, and the
generate_fitting_latex_on_demand-style methods so repeated live option changes
do not accumulate files until window close.

In `@app_desktop/window.py`:
- Around line 3092-3125: The dispatch in generate_latex_for_current_result does
not treat the active statistics result as the preferred stash because
_last_result_kind uses statistics_single/statistics_batches while
generate_statistics_latex_on_demand reads the statistics stash key. Update the
routing so statistics result kinds resolve to the statistics stash key (or track
the active stash key separately) before checking builders, and keep the fallback
order unchanged so the current result always wins over older retained stashes.

In `@app_desktop/workspace_controller.py`:
- Around line 259-280: The capture logic in workspace_controller should decide
file-backed sections from path_text, not the use_file checkbox, so save/restore
stays consistent after a restore. Update the source_kind selection and the
file-read branch in the workspace capture path to treat any non-empty path_text
as the file source, matching the restore behavior that preserves the path even
when use_file_checkbox is unset. Keep the existing fallback to
manual_text/manual_table only when path_text is empty, and ensure the file
attachment is still captured whenever a valid path exists.

In `@tests/test_latex_inputs_serialization.py`:
- Around line 120-158: Add import guards for the Qt-based roundtrip test so it
skips cleanly when dependencies are missing: in
test_workspace_roundtrip_lets_generate_tex_work_after_reopen, add
pytest.importorskip checks for pytestqt and PySide6 before using qtbot or
importing ExtrapolationWindow. Keep the guard placement consistent with the
other Qt-dependent tests in this file so the test is only collected when the
required packages are available.

---

Outside diff comments:
In `@app_desktop/history_panel.py`:
- Around line 118-166: The collapse indicator is being overwritten when
`_apply_language()` refreshes `title_label`, so the ▸/▾ state is lost after a
language switch. Update the language-change flow in `HistoryPanel` to re-run
`_apply_history_collapsed()` after `_register_texts()`/title rewrites, or change
`title_label` registration so it does not include the arrow prefix; use
`toggle_history_collapsed()`, `set_history_collapsed()`, and
`_apply_history_collapsed()` as the hooks to keep the collapsed state and label
in sync.

In `@statistics_utils.py`:
- Around line 172-185: The statistics worker path is not forwarding
native_group_width into generate_statistics_latex, so grouped non-dcolumn LaTeX
can still emit digit-group-size for bundled Tectonic. Update the worker-side
call in workers_core.py to pass the engine capability check result, using the
same _engine_supports_group_width logic already used by the UI path, so
generate_statistics_latex only enables native grouping when supported.
- Around line 496-519: The remaining statistics LaTeX writers still ignore the
requested grouping width when native_group_width is disabled. Update
generate_statistics_bootstrap_latex, generate_statistics_time_series_latex, and
generate_statistics_hypothesis_latex to use the same app-side grouping fallback
as generate_statistics_latex and generate_statistics_latex_batches instead of
relying on siunitx’s default 3-digit grouping, and make sure the worker callsite
passes native_group_width through where these generators are invoked. Use the
existing symbols _statistics_latex_preamble, latex_group_size, and
native_group_width to keep the behavior consistent.

---

Nitpick comments:
In `@app_desktop/constants_editor.py`:
- Around line 146-168: The style refresh logic is duplicated between
set_embedded_in_workbench and refresh_theme_style in ConstantsEditor. Refactor
so set_embedded_in_workbench only updates the embedded property and layout
margins, then delegates to refresh_theme_style for the shared
setStyleSheet/unpolish/polish sequence. Keep refresh_theme_style as the single
place that reads datalab_constants_embedded and reapplies
constants_editor_style.

In `@app_desktop/result_overview_popover.py`:
- Line 90: The popover layout is still using hardcoded contents margins instead
of the shared design token, which breaks consistency with the recent padding
refactor. Update the result-overview popover setup in the relevant layout
initialization to use CARD_PADDING from theme.py, or a dedicated popover padding
token if the spacing is intentionally different, and make sure the import and
margin assignment stay aligned with the existing card-style surfaces.

In `@app_desktop/result_status_strip.py`:
- Line 17: The import in result_status_strip is reaching into
underscore-prefixed private helpers from workbench_results, which creates a
fragile internal dependency. Update workbench_results to expose _overview_state
and _status_badge through a stable public API, either by renaming them to public
symbols or adding public wrapper functions, and then switch result_status_strip
to import and use those public names instead.

In `@app_desktop/table_copy.py`:
- Around line 35-42: The TSV export in the table copy path does not escape
embedded tabs or newlines in cell text, so pasted data can misalign rows and
columns. Update the copy routine that builds lines from table.item(...).text()
so each cell value is sanitized before joining, replacing or escaping literal
tab/newline characters consistently. Keep the fix localized to the table copy
helper that appends cells and writes to QApplication.clipboard().setText.

In `@app_desktop/theme.py`:
- Around line 258-287: The formula preview styles still duplicate a hardcoded
dark background value, so update `formula_preview_surface_style` and
`formula_preview_source_edit_style` to resolve the background through `_tok()`
instead of embedding `#1f2328`. Add or reuse a dedicated token role such as
`formula_bg` in the theme token lookup so both helpers share the same source of
truth and cannot drift apart.
- Around line 441-452: The tab corner radius in the input data tabs style is
hardcoded instead of using the shared control radius constant. Update the
`build_theme`/theme stylesheet section that defines `QTabWidget#input_data_tabs
QTabBar::tab` to replace the literal corner radius values with `RADIUS_CONTROL`
(consistent with the nearby `RADIUS_CARD` usage), so the tab corners follow the
same theme token as other control elements.

In `@app_desktop/window_extrapolation_mixin.py`:
- Around line 827-831: The UI mixin is importing underscore-prefixed private
helpers from workers_core, creating tight cross-module coupling; update
window_extrapolation_mixin to stop relying on _input_units_for_headers and
_result_unit_from_units directly. Move the shared unit conversion/header logic
into a public helper location or a dedicated units-utils module, then import
those public symbols here and keep workers_core's private surface internal.

In `@app_desktop/window_statistics_mixin.py`:
- Around line 288-349: The single/multi-batch LaTeX dispatch is duplicated
between generate_statistics_latex_on_demand and the runtime path in
_run_statistics_mode, so future argument changes can drift. Extract the shared
single-vs-batch decision and generate_statistics_latex /
generate_statistics_latex_batches argument assembly into one helper, then call
it from both generate_statistics_latex_on_demand and _run_statistics_mode,
keeping native_group_width, caption, group_size, and value_col_joined handling
in one place.
- Line 318: Repeated engine-capability probing for the group-width flag is
duplicated across many LaTeX generation call sites in
window_statistics_mixin.py. Add a small helper on the same class, such as a
private method like _native_group_width_flag(), that encapsulates the
hasattr(self, "_engine_supports_group_width") fallback logic and returns the
computed boolean. Then update the call sites that currently inline
self._engine_supports_group_width() if hasattr(self,
"_engine_supports_group_width") else True to use the new helper instead, keeping
the behavior identical while centralizing the logic.

In `@app_desktop/workbench_variable_panel.py`:
- Around line 259-270: The `_variable_summary_text` helper still carries dead
handling for the `"constants"` role even though `_mounts_in_panel_order` no
longer produces it. Remove the unused `"constants"` entry from the `counts` map
and drop `"constants"` from the role iteration in `_variable_summary_text`,
keeping only the active roles so the summary logic matches the current section
roles.
- Line 218: Update the `has_variables` check in `workbench_variable_panel.py` to
stop considering `spec.constants`, since constants are no longer mounted by
`_mounts_in_panel_order`. Keep the visibility logic aligned with the actual
mounted sections by checking only the active variable sources used by the panel
(for example, parameters and tables), so `has_variables` reflects what can
really be shown and avoids the extra empty iteration path.

In `@datalab_latex/latex_formatting.py`:
- Around line 636-679: The digit-grouping logic in group_digits_both_sides
duplicates the same “insert sep every group_size digits” pattern used elsewhere,
so extract that shared behavior into a small helper (for example a private
digit-grouping function) and reuse it for both the reversed integer-part loop
and the forward fractional-part loop. Keep group_digits_both_sides focused on
parsing sign/int/frac/tail and delegating to the helper, and align the new
helper with the existing add_spacing_to_number implementation so the grouping
behavior stays consistent.

In `@datalab_latex/sisetup_block.py`:
- Around line 26-53: Update the docstring for build_sisetup_block to document
the new emit_digit_group_size parameter alongside group_size and
include_dcolumn; add a Parameters entry explaining the tri-state behavior (True
emits digit-group-size, False omits it, None falls back to the legacy heuristic)
so the public API matches the implementation used by the call sites.

In `@docs/SWARM_REVIEW_2026.md`:
- Around line 262-269: The issue is that the recommendation text for the
`__main__` SocketIO finding was truncated and only reconstructed from context,
so it should not be treated as final. Verify the original finding source and
replace the placeholder wording in this SWARM review entry with the exact
recommendation text, keeping the `__main__`/SocketIO note consistent with the
rest of the report.

In `@statistics_utils.py`:
- Around line 184-257: The duplicated app_group/_maybe_group closure should be
removed and centralized so the siunitx grouping behavior stays consistent across
both statistics table generators. Extract the shared grouping logic into a
module-level helper, then update generate_statistics_latex and
generate_statistics_latex_batches to use that helper instead of defining the
same local closure twice; keep the existing behavior for app_group, group_size,
and _maybe_group call sites unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 8a1aab45-9fc4-4976-90e5-477f92c5b751

📥 Commits

Reviewing files that changed from the base of the PR and between 4aa10ea and 02adc4d.

📒 Files selected for processing (127)

• app_desktop/constants_editor.py
• app_desktop/current_page_stack.py
• app_desktop/fitting_latex_writer.py
• app_desktop/formula_preview.py
• app_desktop/history_panel.py
• app_desktop/history_popup.py
• app_desktop/latex_inputs_serialization.py
• app_desktop/latex_preview_dialog.py
• app_desktop/message_dialogs.py
• app_desktop/options_dialogs.py
• app_desktop/panels.py
• app_desktop/result_overview_popover.py
• app_desktop/result_status_strip.py
• app_desktop/root_latex_writer.py
• app_desktop/shell_layout.py
• app_desktop/table_copy.py
• app_desktop/theme.py
• app_desktop/views/error.py
• app_desktop/views/fitting.py
• app_desktop/views/helpers.py
• app_desktop/views/root_solving.py
• app_desktop/views/statistics.py
• app_desktop/window.py
• app_desktop/window_data_mixin.py
• app_desktop/window_extrapolation_mixin.py
• app_desktop/window_fitting_formatters_mixin.py
• app_desktop/window_fitting_models_mixin.py
• app_desktop/window_fitting_residuals_mixin.py
• app_desktop/window_i18n_mixin.py
• app_desktop/window_latex_compile_mixin.py
• app_desktop/window_statistics_mixin.py
• app_desktop/workbench_formula_panel.py
• app_desktop/workbench_layout.py
• app_desktop/workbench_results.py
• app_desktop/workbench_toolbar.py
• app_desktop/workbench_variable_panel.py
• app_desktop/workbench_visual_contract.py
• app_desktop/workers_core.py
• app_desktop/workspace_controller.py
• datalab_latex/latex_formatting.py
• datalab_latex/latex_tables_common.py
• datalab_latex/latex_tables_error_propagation.py
• datalab_latex/latex_tables_extrapolation.py
• datalab_latex/latex_tables_fitting.py
• datalab_latex/latex_tables_root.py
• datalab_latex/latex_tables_statistics_grouped.py
• datalab_latex/latex_tables_statistics_matrix.py
• datalab_latex/sisetup_block.py
• docs/DATALAB_WEB_GUIDE.en.md
• docs/DATALAB_WEB_GUIDE.md
• docs/SWARM_REVIEW_2026.md
• docs/superpowers/specs/2026-07-03-adaptive-workbench-section1-design.md
• docs/superpowers/specs/2026-07-03-adaptive-workbench-section2-plan.md
• docs/superpowers/specs/2026-07-04-iconified-menubar-design.md
• docs/superpowers/specs/2026-07-04-toolbar-options-popup-design.md
• docs/superpowers/specs/2026-07-05-latex-ondemand-4.2-integration-notes.md
• docs/superpowers/specs/2026-07-05-latex-ondemand-4.4b-result-tabs-removal.md
• docs/superpowers/specs/2026-07-05-latex-ondemand-4.4d-remove-generate-checkbox.md
• docs/superpowers/specs/2026-07-05-latex-ondemand-refactor-design.md
• docs/superpowers/specs/2026-07-05-latex-pdf-window-cleanup-design.md
• docs/superpowers/specs/2026-07-05-two-pane-layout-design.md
• docs/superpowers/specs/2026-07-06-engine-capability-grouping.md
• docs/superpowers/specs/2026-07-08-left-config-card-unify-design.md
• docs/web/deploy.en.md
• docs/web/deploy.zh.md
• gunicorn.conf.py
• shared/latex_engine.py
• shared/ui_specs.py
• statistics_utils.py
• test_probe_widget_tree.py
• tests/test_constants_editor.py
• tests/test_deploy_docs_wsgi_targets.py
• tests/test_desktop_about_dialog.py
• tests/test_desktop_bilingual_inventory.py
• tests/test_desktop_custom_fit_ui.py
• tests/test_desktop_engine_group_width.py
• tests/test_desktop_error_propagation_ui.py
• tests/test_desktop_example_workspace_menu.py
• tests/test_desktop_global_options_ui.py
• tests/test_desktop_gui_redesign_scan.py
• tests/test_desktop_gui_schema_scan.py
• tests/test_desktop_gui_workflows.py
• tests/test_desktop_history_collapse.py
• tests/test_desktop_history_toolbar_popup.py
• tests/test_desktop_input_constants_tabs.py
• tests/test_desktop_latex_compile_ui.py
• tests/test_desktop_latex_ondemand_dispatch.py
• tests/test_desktop_latex_ondemand_error.py
• tests/test_desktop_latex_ondemand_extrapolation.py
• tests/test_desktop_latex_ondemand_fitting.py
• tests/test_desktop_latex_ondemand_root.py
• tests/test_desktop_latex_ondemand_statistics.py
• tests/test_desktop_latex_preview_dialog.py
• tests/test_desktop_message_dialogs.py
• tests/test_desktop_mode_selector_on_toolbar.py
• tests/test_desktop_mode_stack.py
• tests/test_desktop_option_reachability.py
• tests/test_desktop_options_dialogs.py
• tests/test_desktop_result_overview_popover.py
• tests/test_desktop_result_schema_ui.py
• tests/test_desktop_result_status_strip.py
• tests/test_desktop_root_solving_ui.py
• tests/test_desktop_shell_layout.py
• tests/test_desktop_statistics_ui.py
• tests/test_desktop_table_cell_copy.py
• tests/test_desktop_theme_tokens.py
• tests/test_desktop_toolbar_status_overview.py
• tests/test_desktop_two_pane_layout.py
• tests/test_desktop_workbench_data_area.py
• tests/test_desktop_workbench_editor_canvas.py
• tests/test_desktop_workbench_layout.py
• tests/test_desktop_workbench_results.py
• tests/test_desktop_workbench_state_ownership.py
• tests/test_desktop_workbench_variable_panel.py
• tests/test_desktop_workbench_visual_contract.py
• tests/test_desktop_workbench_visual_screenshots.py
• tests/test_file_size_ratchet.py
• tests/test_fitting_latex_writer.py
• tests/test_formula_preview_rendering.py
• tests/test_latex_engine_adaptive_review_fixes.py
• tests/test_latex_engine_capability.py
• tests/test_latex_group_size_zero.py
• tests/test_latex_inputs_serialization.py
• tests/test_sisetup_block.py
• tests/test_splitter_persistence.py
• tests/test_workspace_controller.py
• tools/scan_desktop_gui_schema.py

💤 Files with no reviewable changes (8)

• app_desktop/views/statistics.py
• tests/test_desktop_workbench_state_ownership.py
• app_desktop/views/fitting.py
• app_desktop/views/root_solving.py
• tests/test_desktop_custom_fit_ui.py
• tests/test_desktop_statistics_ui.py
• app_desktop/views/helpers.py
• app_desktop/views/error.py

[coderabbitai]

🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Capture code must derive file source from path_text, not use_file_checkbox, to match file-precedence restore behavior.

The restore path (line 1926) unconditionally sets use_file_checkbox.setChecked(False) and keeps the file path only if the file still exists. But the capture path still gates on use_file (the checkbox): source_kind = "file" if (use_file and path_text) else ... and if use_file and path_text: for the file read.

After a restore where the file exists, the checkbox is False and the path is set. On the next save, the capture sees use_file=False, so it:

1. Tags source_kind as "manual_table"/"manual_text" instead of "file".
2. Does not read the file — no attachment is saved.

On a subsequent restore where the file has since been moved/deleted, the path is cleared and the workspace falls back to the manual editor's data (possibly empty or stale), with no file attachment. The restore comment claims "its CONTENTS were captured as an attachment on save → the workspace stays self-contained," but this only holds for the initial save — after a restore→save cycle, the workspace is no longer self-contained.

The fix is to gate the file source on path_text instead of use_file:

🔒 Proposed fix

     _text_view = stack is not None and stack.currentIndex() == 1
     source_kind = (
-        "file" if (use_file and path_text) else ("manual_text" if _text_view else "manual_table")
+        "file" if path_text else ("manual_text" if _text_view else "manual_table")
     )
     if constants and editor is not None and not use_file:
         source_kind = "manual_text" if editor.using_text_view() else "manual_table"
     canonical: dict[str, Any]
-    if use_file and path_text:
+    if path_text:
         # Guard the file read (review P-A): if the file was moved/deleted/unreadable since the
         # user picked it, saving the workspace must NOT crash. Keep the source as "file" and the
         # path (so the user can re-point it), but attach empty content rather than raising.

This aligns the capture with the restore's file-precedence model: the file path is the sole determinant of whether the section is file-backed, not the legacy checkbox.

A regression test for the second save-restore cycle (restore → save → delete file → restore) should be added to confirm the workspace remains self-contained.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@app_desktop/workspace_controller.py` around lines 259 - 280, The capture
logic in workspace_controller should decide file-backed sections from path_text,
not the use_file checkbox, so save/restore stays consistent after a restore.
Update the source_kind selection and the file-read branch in the workspace
capture path to treat any non-empty path_text as the file source, matching the
restore behavior that preserves the path even when use_file_checkbox is unset.
Keep the existing fallback to manual_text/manual_table only when path_text is
empty, and ensure the file attachment is still captured whenever a valid path
exists.

[yilibinbin]

Addressed the CodeRabbit review in 1d8496b. Disposition of each finding:

#	File	Verdict	Action
1	fitting_latex_writer.py:131 — \text{} wrapper (Critical)	Confirmed	Return the grouped string directly (removed the wrapper).
2	window_extrapolation_mixin.py:867 — unguarded on-demand LaTeX (Major)	Confirmed	Wrapped generate_latex_for_current_result() in open_latex_preview with try/except → warning dialog.
3	window_fitting_residuals_mixin.py:576 — temp-file leak (Minor)	Confirmed	Added reuse=True to latex_output_path_for_run so on-demand regen reuses one temp .tex per session.
4	window.py:3125 — statistics stash key (Major)	Confirmed	Collapse any statistics* result kind to the statistics builder key in dispatch.
5	workspace_controller.py:280 — capture gates on checkbox (Major)	Refuted	The use_file_checkbox is a _FilePathChecked shim whose isChecked() == (a file path is present). So after a restore-where-file-exists → re-save, use_file is True, source_kind="file", and the attachment IS saved. Reproduced: the re-saved workspace still contains the file attachment (stays self-contained). CodeRabbit's premise (use_file=False after restore) reflects the old real-checkbox semantics.
6	test_latex_inputs_serialization.py:158 — missing importorskip (Minor)	Confirmed	Added importorskip("pytestqt")/("PySide6") inside the one qtbot test.

49 latex / serialization / on-demand-dispatch tests pass locally.

[coderabbitai]

🧹 Nitpick comments (1)

tests/test_app_web_docs_baseline.py (1)

88-104: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider making the heading-ID and TOC-anchor regexes more robust.

The heading-ID regex <h[123]\s+id="([^"]+)" requires id to be the first attribute after the tag name. If the template ever emits <h2 class="..." id="slug">, the regex won't capture it, potentially causing false failures in the toc_targets <= heading_ids assertion. Similarly, href="#([^"]+)" captures all in-page anchors, not just TOC links — any non-TOC anchor (footnote, nav) pointing to a non-heading element would fail the subset check.

If the current HTML output always puts id first and the page has no non-TOC anchors, the test is fine today. Consider tightening the patterns for long-term robustness.

♻️ Suggested more robust regex patterns

-    heading_ids = set(re.findall(r'<h[123]\s+id="([^"]+)"', html))
+    heading_ids = set(re.findall(r'<h[123]\s[^>]*\bid="([^"]+)"', html))
     assert heading_ids, "no heading id= attributes were emitted (TOC anchors would be dead)"
     # Every TOC anchor (href="`#slug`") must point at a heading id that exists on the page.
-    toc_targets = set(re.findall(r'href="#([^"]+)"', html))
+    toc_targets = set(re.findall(r'<a\s[^>]*\bhref="#([^"]+)"', html))

The \bid= pattern matches id regardless of attribute position, and restricting to <a tags narrows to actual anchor elements.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_app_web_docs_baseline.py` around lines 88 - 104, Make the regex
assertions in test_docs_headings_get_ids_so_toc_anchors_resolve robust to HTML
attribute ordering and anchor context: match heading id attributes using a word
boundary so id need not be first, and restrict href extraction to anchor tags
while preserving fragment matching. Update the patterns so valid headings with
other attributes and relevant TOC links are handled without false failures.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@tests/test_app_web_docs_baseline.py`:
- Around line 88-104: Make the regex assertions in
test_docs_headings_get_ids_so_toc_anchors_resolve robust to HTML attribute
ordering and anchor context: match heading id attributes using a word boundary
so id need not be first, and restrict href extraction to anchor tags while
preserving fragment matching. Update the patterns so valid headings with other
attributes and relevant TOC links are handled without false failures.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 9dfe179a-016a-492c-8ad4-f1e2ed109c22

📥 Commits

Reviewing files that changed from the base of the PR and between 0fe1aeb and c8cfac1.

📒 Files selected for processing (24)

• app_desktop/theme.py
• app_desktop/workspace_controller.py
• app_web/blueprints/docs.py
• app_web/blueprints/pages.py
• app_web/logic/common.py
• app_web/logic/error_propagation.py
• app_web/logic/extrapolation.py
• app_web/logic/fitting.py
• app_web/logic/root_solving.py
• app_web/logic/statistics.py
• datalab_latex/latex_formatting.py
• extrapolation_methods/accelerators.py
• fitting/hp_fitter.py
• tests/test_app_web_compute_rate_limit.py
• tests/test_app_web_docs_baseline.py
• tests/test_app_web_precision_clamp.py
• tests/test_constants_text_view.py
• tests/test_desktop_input_constants_tabs.py
• tests/test_desktop_theme_tokens.py
• tests/test_extrapolation_accelerators.py
• tests/test_file_size_ratchet.py
• tests/test_fitting_linear_model_sanity.py
• tests/test_latex_formatting_precision_floor.py
• tests/test_workspace_controller.py

💤 Files with no reviewable changes (1)

• tests/test_file_size_ratchet.py

🚧 Files skipped from review as they are similar to previous changes (4)

• tests/test_workspace_controller.py
• tests/test_desktop_theme_tokens.py
• app_desktop/workspace_controller.py
• tests/test_desktop_input_constants_tabs.py

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@app_web/logic/fitting.py`:
- Around line 254-265: In _build_self_consistent_problem, keep JSON loading
inside the try block but move the params_cfg dictionary validation and
normalized_cfg construction after it, so the intentional ValueError for
non-object JSON is not caught and wrapped again. Preserve exception chaining and
bilingual parsing-error handling for JSON load failures.

In `@shared/ui_specs.py`:
- Around line 360-370: The help metadata is missing the newly supported
self_consistent fit model. Add a self_consistent entry to shared/help_specs.json
matching the model’s identifier and description defined near the self_consistent
UI specification, keeping the desktop and web help content consistent.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: d33f42b5-4ca1-492d-ad5f-19e9f73ca94f

📥 Commits

Reviewing files that changed from the base of the PR and between c8cfac1 and caa4e27.

📒 Files selected for processing (9)

• app_web/logic/fitting.py
• app_web/static/js/i18n.js
• app_web/templates/fit.html
• datalab_latex/latex_formatting.py
• shared/ui_specs.py
• tests/test_app_web_fitting_self_consistent.py
• tests/test_auto_fit_removed.py
• tests/test_desktop_extrapolation_ui.py
• tests/test_file_size_ratchet.py

✅ Files skipped from review due to trivial changes (1)

• tests/test_file_size_ratchet.py

🚧 Files skipped from review as they are similar to previous changes (1)

• datalab_latex/latex_formatting.py

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Verify help_specs.json was updated for new user-visible parameters
if [ -f shared/help_specs.json ]; then
  echo "=== Checking for 'quadratic' ==="
  rg -n 'quadratic' shared/help_specs.json || echo "No 'quadratic' entries found"
  echo "=== Checking for 'self_consistent' ==="
  rg -n 'self_consistent' shared/help_specs.json || echo "No 'self_consistent' entries found"
else
  echo "shared/help_specs.json not found"
fi

Repository: yilibinbin/DataLab

Length of output: 286

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo "== file list =="
git ls-files 'shared/ui_specs.py' 'shared/help_specs.json'

echo
echo "== ui_specs.py around cited lines =="
sed -n '340,430p' shared/ui_specs.py

echo
echo "== help_specs.json around quadratic =="
sed -n '100,150p' shared/help_specs.json

echo
echo "== search for self_consistent across repo =="
rg -n 'self_consistent|self-consistent|self consistent' shared . --glob '!**/.git/**'

Repository: yilibinbin/DataLab

Length of output: 50374

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo "== help_specs fitting-related keys =="
rg -n '"(fitting|fit|model|self_consistent|quadratic|custom)"\s*:' shared/help_specs.json

echo
echo "== help_specs around any fitting section =="
python3 - <<'PY'
import json
from pathlib import Path

data = json.loads(Path("shared/help_specs.json").read_text(encoding="utf-8"))
for k in data:
    if any(s in k.lower() for s in ("fit", "model", "quadratic", "custom", "self")):
        print(k)
PY

Repository: yilibinbin/DataLab

Length of output: 281

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo "== ui_specs.py around 440-560 =="
sed -n '440,560p' shared/ui_specs.py

echo
echo "== full help_specs top-level keys =="
python3 - <<'PY'
import json
from pathlib import Path
data = json.loads(Path("shared/help_specs.json").read_text(encoding="utf-8"))
print("\n".join(data.keys()))
PY

Repository: yilibinbin/DataLab

Length of output: 4168

---

Add the self_consistent help entry

shared/help_specs.json already includes quadratic, but it still has no entry for the new self_consistent fit model added in shared/ui_specs.py:540-544. Keep the desktop/web help data in sync by adding that model description here.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@shared/ui_specs.py` around lines 360 - 370, The help metadata is missing the
newly supported self_consistent fit model. Add a self_consistent entry to
shared/help_specs.json matching the model’s identifier and description defined
near the self_consistent UI specification, keeping the desktop and web help
content consistent.

Source: Coding guidelines