Docs: Python cells as data (architecture, file formats, functions, roadmap)

architecture.md gains a Python cells section: the model type, the text
forms and the size limit, the xlsx Python part, calc laying out kept
results as graph nodes, and the limits on open. file-format-parity.md
gains a Python cells table (xlsx as [MS-XLSX] 2.3.10 specifies, the
formula naming its code in the Python part, with cached values and
spills; other formats; GridCraft JSON; the limits on open; and
rich-value object results as not kept). function-parity.md marks the
four Python errors done with their ERROR.TYPE numbers and adds a Python
in cells row (stored, not run). ROADMAP.md records the progress. The
return types are written as "0 or 1" so no table cell holds an
unescaped pipe. architecture.md's "Last updated" label loses the space
inside its bold markers.

ROADMAP.md

@@ -2,7 +2,7 @@
 
 **Stage: alpha** · next: beta, ~25 points of "ready for real work" and ~160–240 h away
 
-> **Last reviewed:** 2026-10-11 · **Last updated:** 2026-10-25 · **Change:** minor (AutoSum over a selected row puts its total to the right, and fills a selection's empty last row and column) · **Target:** Microsoft Excel (Microsoft 365)
+> **Last reviewed:** 2026-10-11 · **Last updated:** 2026-10-25 · **Change:** minor (Python cells stored and kept in files; earlier: AutoSum over a selected row puts its total to the right, and fills a selection's empty last row and column) · **Target:** Microsoft Excel (Microsoft 365)
 
 GridCraft aims at full Microsoft Excel parity, and at being better in places: faster, open (XLSX
 is the native format), fully scriptable by agents (MCP, control channel, CLI), and on the web.
@@ -135,6 +135,7 @@ Ranked; estimates in Opus 5.5 agent-hours. Detail and the full beta-gate list in
 
 | Date | What landed |
 |---|---|
+| 2026-10-25 | Python cells as data: `python.set` stores code in a cell, Python cells copy, move, fill and undo like formulas, and `.xlsx` reads and writes Python cells as [MS-XLSX] specifies (`_xlfn._xlws.PY(index,0 or 1)`, the code in the workbook's Python part) with cached values and spills, so such workbooks open with their results; nothing runs Python (new code shows `#BLOCKED!`), and opening a file limits code size, result size and spill ranges |
 | 2026-10-25 | AutoSum over a selection does what Excel does: a selected row gets one total to its right (in its last cell if that is empty), not a one-cell total under each cell; a block whose last column is empty gets each row's total there, one whose last row is empty each column's, and with both the corner gets the total of the block; otherwise totals go under each column as before |
 | 2026-10-24 | Go To lists the named ranges of every sheet (`Sheet2!Name` for a name scoped to another sheet; hidden names, constants and formulas left out) and the references entered before, a double-click goes to an entry, and the Reference box takes several names or references separated by commas (`Price, Type`, `B14:C22,F19:G30`) |
 | 2026-10-24 | Status bar: a right-click opens a menu that sets which selection statistics show (Average, Count and Sum by default; Numerical Count, Minimum and Maximum are new), kept with the preferences (`view.statusBar`) |
@@ -242,6 +243,7 @@ Ranked; estimates in Opus 5.5 agent-hours. Detail and the full beta-gate list in
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-10-25 | minor | Progress log: Python cells as data |
 | 2026-10-25 | minor | Progress log: AutoSum over a selected row, and a selection's empty last row and column |
 | 2026-10-23 | minor | Progress log: accounting negatives in number text |
 | 2026-10-23 | minor | Progress log: INDEX with one index on a 2-D reference is #REF! |

docs/architecture.md

@@ -1,6 +1,6 @@
 # Architecture
 
-> **Last reviewed:** 2026-10-10 · ** Last updated:** 2026-10-24 · **Change:** minor (engine building blocks for multi-command actions) · **Target:** Microsoft Excel (Microsoft 365)
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-25 · **Change:** minor (Python cells as data; earlier: engine building blocks for multi-command actions) · **Target:** Microsoft Excel (Microsoft 365)
 
 How GridCraft is built today. Rules for contributors are in [`AGENTS.md`](../AGENTS.md).
 
@@ -125,6 +125,25 @@ Actions made of many commands (macros, a debugger that runs one step per call) u
   under (a hidden workbook-scope name of either shape; a visible one is the person's own), and
   `formulas.createFromSelection` skips such labels, unless `force: true` is passed.
 
+## Python cells
+
+A Python cell is a `Cell` whose `py` holds a `model::PyCode`: the code, the return type (an Excel
+value or a Python object, the `0` or `1` of Excel's `PY(code, return_type)`), the last result and a
+stale flag. Its `value` is the cached result, so formulas, display and saving read it like any
+other cell. `model::python` has the text forms: `PY("…"&"…",n)` with literals of at most 255
+characters (the function's syntax, which some files hold), the `_xlfn._xlws.` prefix, the
+8,192-character limit measured on that formula (`PyCode::fits`), and the `xl("…")` literals in
+code, which structural edits rewrite like formula references.
+
+Calc treats a Python cell as a graph node that reads nothing: a pass takes its kept result instead
+of evaluating it, so an array result spills, is blocked and unblocked, and reaches its dependents
+through the same code as a formula's. The xlsx reader and writer map Python cells to and from
+`_xlfn._xlws.PY(index,n)` formulas with cached values and the workbook's Python part, which holds
+the code (`xlsx::python_part`, [MS-XLSX] 2.3.10); `io::adopt_python` and `io::enforce_python_limits`
+bound what a file can make the app hold (result size, spill ranges, total rebuilt cells, code
+size). This build runs no Python: `python.set` stores code, and new or changed code shows
+`#BLOCKED!`.
+
 ## Agent control
 
 - Desktop control channel: `gridcraft --control 7979`, JSON lines on localhost
@@ -150,6 +169,7 @@ rpm, tarball), FreeBSD, and web (WASM; WebGPU or WebGL2). Release workflows in
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-10-25 | minor | Python cells as data: `model::PyCode`, the file forms, calc laying out kept results, and the limits on open |
 | 2026-10-24 | minor | Engine building blocks for multi-command actions: undo groups, recording buffer, document trust, chunked hidden-name storage |
 | 2026-10-14 | minor | Document Undo's retained-cell-storage budget and latest-step guarantee |
 | 2026-10-10 | major | Created from the code on main: crates, layers, data flow, agent control, known gaps |

docs/file-format-parity.md

@@ -1,6 +1,6 @@
 # File-format parity
 
-> **Last reviewed:** 2026-10-11 · **Last updated:** 2026-10-21 · **Change:** minor (.xls dialog sheets are skipped instead of read as worksheets; `.xls` import reads each sheet's display options from WINDOW2: right-to-left, gridlines, headings, zeros, formulas; XLSX text boxes and shapes keep their paragraphs' right-to-left direction on save; CSV/TSV/TXT can open with every field kept as text via `file.open {csvAsText}`; formulas over Excel's 8,192 characters, and `%` or call chains past the nesting limit, are parse errors instead of stack overflows when a file is opened; XLSX PivotTables read and write `rowGrandTotals`/`colGrandTotals` the way Excel does, so a table saved with only one grand total opens with the right one; HTML export keeps right-to-left; CSV and text import detects likely Windows-1256 Arabic when invalid UTF-8 has Arabic byte runs; other invalid UTF-8 remains Windows-1252; HTML export places pictures and charts, as inline SVG, at their anchors) · **Target:** Microsoft Excel (Microsoft 365), Excel for Mac 16.113.4
+> **Last reviewed:** 2026-10-11 · **Last updated:** 2026-10-25 · **Change:** minor (Python cells in `.xlsx` as [MS-XLSX] specifies, `_xlfn._xlws.PY(…)` naming code in the Python part, with cached values, and in GridCraft JSON, with limits on open; earlier: .xls dialog sheets are skipped instead of read as worksheets; `.xls` import reads each sheet's display options from WINDOW2: right-to-left, gridlines, headings, zeros, formulas; XLSX text boxes and shapes keep their paragraphs' right-to-left direction on save; CSV/TSV/TXT can open with every field kept as text via `file.open {csvAsText}`; formulas over Excel's 8,192 characters, and `%` or call chains past the nesting limit, are parse errors instead of stack overflows when a file is opened; XLSX PivotTables read and write `rowGrandTotals`/`colGrandTotals` the way Excel does, so a table saved with only one grand total opens with the right one; HTML export keeps right-to-left; CSV and text import detects likely Windows-1256 Arabic when invalid UTF-8 has Arabic byte runs; other invalid UTF-8 remains Windows-1252; HTML export places pictures and charts, as inline SVG, at their anchors) · **Target:** Microsoft Excel (Microsoft 365), Excel for Mac 16.113.4
 
 Every format Excel opens or saves, with GridCraft's read and write support, fidelity and tests.
 Part of [target-app-parity.md](target-app-parity.md). **File formats: ~50% ready, 80–120 h**
@@ -93,6 +93,16 @@ Server, Access, ODBC, …), web pages, SharePoint and online services, through P
 GridCraft imports CSV/TSV, JSON and HTML tables only, with no query editor, no connections and no
 refresh. Counted under features (Get & Transform, ~10% ready), not here.
 
+## Python cells
+
+| Where | GridCraft | Notes |
+|---|---|---|
+| `.xlsx` | read and write | Written as [MS-XLSX] 2.3.10 specifies: the cell's formula is `_xlfn._xlws.PY(i,n)`, where `i` is the index of the cell's script in the workbook's Python part (`xl/python.xml`, `application/vnd.ms-excel.python+xml`) and `n` the return type, 0 for an Excel value or 1 for a Python object, with the cached value. The code keeps its `xl()` references as text, so the formula has no further arguments. A spilling result is written as a dynamic array with its spilled values cached. Read back with or without the `_xlfn.` and `_xlws.` prefixes: an index names a script of the Python part, whose `%Px%` references become the text of the formula's argument `x`, and the older form with the code in string literals, `PY("…"&"…",n)`, still reads. So workbooks laid out as the format specifies open with Python cells, their cached values and spills. Python errors are cached as `#VALUE!`, as Excel caches `#SPILL!` and `#CALC!` |
+| `.xls`, `.xlsb`, `.ods` | read | A `PY("…",0)` or `PY("…",1)` formula read from any format becomes a Python cell; the `.xlsb` and `.ods` imports keep values only, so Python cells there come in as their cached values |
+| GridCraft JSON | read and write | The whole Python cell: code, return type, the last result and the exact error |
+| Limits on open | — | Code that doesn't fit Excel's 8,192-character formula as `PY("…",n)` keeps only its cached value, with a warning, and so does a formula naming a script the Python part doesn't hold. The Python part is read up to 64 MB and 1,000,000 scripts, and the cells take at most 64 MB of code from it in all (several formulas can name one script). A kept result that is not a well-formed array of at most 1,000,000 cells is dropped, a spill range larger than that is ignored, and results are rebuilt for at most 4,000,000 cells over the workbook. Opening runs nothing |
+| Not kept | — | Python object results stored as rich values (Excel's `richData` parts) come in as their cached values. The Python part's environment definition (its id and initialization code) is not read, and a save writes none |
+
 ## Work (76–114 h)
 
 | Work | Hours |
@@ -110,6 +120,7 @@ refresh. Counted under features (Get & Transform, ~10% ready), not here.
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-10-25 | minor | Python cells: `.xlsx` reads and writes `_xlfn._xlws.PY(i,0 or 1)` with the code in the Python part (and reads the string-literal form) with cached values and spills, other formats adopt PY formulas, GridCraft JSON keeps the whole cell, and opening limits code size, result size and spill ranges |
 | 2026-10-21 | minor | HTML (Save as Web Page) writes pictures (`data:` URI `<img>`) and charts (inline SVG from the chart crate's drawing primitives, text included) at their anchors; the table uses the sheet's column widths and row heights when the sheet has either, and output for sheets without pictures or charts is unchanged |
 | 2026-10-20 | trivial | XLSX shared strings: Japanese phonetic runs (ふりがな, `<rPh sb="…" eb="…">` with `<phoneticPr>`) are read into the workbook and written back next to the string with the same text, so an open + save no longer deletes them; bad offsets and readings are dropped with a warning, a `<phoneticPr>` without runs is kept on its own, and when duplicate `<si>` items share a text but not a reading only the first reading is written (reported on open) (`phonetic_tests.rs`) |
 | 2026-10-20 | minor | A workbook theme's major/minor per-script typefaces (`<a:font script="Arab" typeface="…"/>`, also Hebr, Jpan and the rest) are read and written back in file order after `<a:cs>`; they were dropped before. A hostile file is held to 64 entries of 256 characters |

docs/function-parity.md

@@ -1,6 +1,6 @@
 # Function parity: worksheet functions and formulas
 
-> **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-25 · **Change:** minor (the LAMBDA row lists the recursion limit and its `#NUM!`, and the error a defined name holding an error gives when called; earlier: number text with accounting parentheses coerces like Excel 365; INDEX with one index on a 2-D reference is #REF! as in Excel; earlier: SUBTOTAL and AGGREGATE cap the cells they visit, #2175) · **Target:** Microsoft Excel (Microsoft 365), Excel for Mac 16.113.4
+> **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-25 · **Change:** minor (the Python errors `#PYTHON!`, `#BUSY!`, `#TIMEOUT!` and `#BLOCKED!`, and a Python in cells row; earlier: the LAMBDA row lists the recursion limit and its `#NUM!`, and the error a defined name holding an error gives when called; earlier: number text with accounting parentheses coerces like Excel 365; INDEX with one index on a 2-D reference is #REF! as in Excel; earlier: SUBTOTAL and AGGREGATE cap the cells they visit, #2175) · **Target:** Microsoft Excel (Microsoft 365), Excel for Mac 16.113.4
 
 Excel's worksheet functions, category by category, against GridCraft's function library. Part of
 [target-app-parity.md](target-app-parity.md); gaps feed [gaps.md](gaps.md).
@@ -93,7 +93,8 @@ This improves series preparation; it does not change the model-fitting or season
 | External workbook references (`[Book.xlsx]Sheet1!A1`) | parsed; links dropped on open, no update |
 | Localized function names and `;` separators | **missing** (#45); en-US only |
 | Long flat operator chains (`=A1+A2+…`, `-`, `*`, `/`, `^`, `&`, comparisons) | done: no operand limit below Excel's 8,192-character formula limit (Excel, observed: `=1+1+…` with 4,096 operands is exactly 8,192 characters with the `=` and calculates; one more character is refused). Evaluated (#484), printed (#628), stored, copied, adjusted, saved and reopened without recursing once per operand, on a 1 MiB stack (#916); written to XLSX as plain formulas, not dynamic arrays; Spanish entry accepts them |
-| Error values | the seven classic errors plus `#SPILL!`, `#CALC!`, `#GETTING_DATA` done; `#FIELD!`, `#BLOCKED!`, `#CONNECT!`, `#BUSY!`, `#UNKNOWN!`, `#PYTHON!` missing (minor: they come from data types, cloud and Python features) |
+| Error values | the seven classic errors plus `#SPILL!`, `#CALC!`, `#GETTING_DATA`, and the Python errors `#PYTHON!`, `#BUSY!`, `#TIMEOUT!`, `#BLOCKED!` done; `#FIELD!`, `#CONNECT!`, `#UNKNOWN!` missing (minor: they come from data types and cloud features). `ERROR.TYPE` gives 11 for `#BLOCKED!` as Excel does; Microsoft documents no number for `#PYTHON!`, `#BUSY!` or `#TIMEOUT!`, and GridCraft gives them 19, 20 and 21 |
+| Python in cells | partial: Python cells are stored, edited (`python.set`), copied, moved and saved, and Excel's `_xlfn._xlws.PY` cells open with their cached results; the code doesn't run, so new or changed code shows `#BLOCKED!`. Typing `=PY(` is refused with a pointer to `python.set`, so `PY` stays counted as missing above |
 
 ## Depth (estimated)
 
@@ -123,6 +124,7 @@ workbooks; results checked in as data, not Excel files) and fixing what it finds
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-10-25 | minor | Python in cells (stored, not run): `#PYTHON!`, `#BUSY!`, `#TIMEOUT!` and `#BLOCKED!` done, with their `ERROR.TYPE` numbers; `PY` still counted as missing |
 | 2026-10-25 | minor | LAMBDA row: a LAMBDA that calls itself more than 140 times, and a formula nested past the depth limit, give `#NUM!` (#1809; nesting gave `#CALC!` before). `=Foo(1)` where the defined name Foo holds `#N/A` gives `#N/A`, not `#NAME?` |
 | 2026-10-23 | minor | Number text coercion (VALUE, arithmetic, typed entry): a dollar sign before accounting parentheses (`$(1,234)`, `$ (5)`) and a percent sign after them (`(5)%`) read as negatives; `(-5)`, `((5))`, `($-5)`, `$5%`, `$(5)%` and dates or times in parentheses are `#VALUE!`, as observed in Excel 365 |
 | 2026-10-23 | minor | INDEX with only a row index on a reference of several rows and columns (`INDEX(A1:C3,2)`, `INDEX(A1:C3,0)`) is `#REF!`, as observed in Excel, so it also fails as a range endpoint and inside other functions; an empty column argument (`INDEX(A1:C3,2,)`) still gives the row. Checked that a column array with one index stays a 1×1 array (`TYPE` 64), which is what Excel returns |