[BUG] pptx: add/validate accept adjust-value <a:gd> on presets without adjust handles (rect/ellipse) — PowerPoint reports file as corrupt

[jasonwei1002]

Summary

add --type shape happily writes an adjust-value <a:gd> into a preset geometry that has no adjust handles (e.g. rect, ellipse), and validate reports no errors — but PowerPoint refuses to open the file, reporting it as corrupt. This is a silent data-corruption path: all officecli-side checks stay green.

Repro (v1.0.136, macOS)

officecli create repro.pptx
officecli add repro.pptx / --type slide
officecli add repro.pptx '/slide[1]' --type shape \
  --prop geometry=rect --prop x=1cm --prop y=1cm --prop width=4cm --prop height=2cm \
  --prop text="test" --prop adj="adj:val 14000"
officecli validate repro.pptx
# -> "Validation passed: no errors found."

Resulting slide XML:

<a:prstGeom prst="rect"><a:avLst><a:gd name="adj" fmla="val 14000"/></a:avLst></a:prstGeom>

Opening this file in desktop PowerPoint triggers the "repair/corrupt" dialog.

Expected

Two layers of defense, either of which would have caught this:

1. add/set should reject adjust values whose name is not legal for the target prst (e.g. rect/ellipse accept none; roundRect/chevron/homePlate/triangle accept adj; round2SameRect and the arrow family accept adj1/adj2, etc.).
2. validate should flag existing <a:gd> entries whose name is not in the preset's avLst as defined by ECMA-376 presetShapeDefinitions.xml.

The full legal-handle table is mechanical to derive from presetShapeDefinitions.xml (each preset's <avLst> lists its gd names), so this can be a static lookup table.

Impact

High — the file passes validate, unzips fine, and renders fine in view html/screenshot, yet is unopenable in PowerPoint. Agents driving officecli have no way to detect the problem short of regex-auditing the raw XML themselves.

[goworm]

Thanks for the detailed report. We could reproduce the officecli side (the <a:gd name="adj" fmla="val 14000"/> does get written into rect's <a:avLst>, and validate stays green), but not the corruption: the exact file produced by your repro steps opens and renders normally in desktop PowerPoint on Windows in our testing — no repair dialog. That matches ECMA-376 behavior, where adjust values in <a:avLst> that don't correspond to a handle of the preset are ignored rather than treated as invalid.

To move forward, could you provide:

1. Your PowerPoint version and platform (Windows or Mac, build number) that shows the repair/corrupt dialog;
2. The actual .pptx file that triggers the dialog — ideally the original file where you first hit this, not just one regenerated from the repro commands. It's possible the corruption in your real file comes from something else and the minimized repro doesn't capture it.

With a sample that reproducibly fails to open we can pinpoint the real cause. Adding a validate-level warning for adjust names not defined by the preset is still worth considering as a lint, but we'd like to confirm the corruption mechanism first before changing add/set behavior.

[jasonwei1002]

1.macOS，PowerPoint version 16.110.3 (26070318)
2.The version regenerated by reproducing the command can also replicate the corruption issue.
repro235.pptx

[argszero]

I'd like to work on this.

Re-measured on the released 1.0.151 today, and the officecli-side half of your report still stands exactly as you described it — that is the half that can be settled without PowerPoint:

$ officecli add repro.pptx /slide[1] --type shape --prop geometry=rect --prop adj='adj:val 14000' --prop text=t
Added shape at /slide[1]/shape[@id=100000]

$ unzip -p repro.pptx ppt/slides/slide1.xml | grep -o '<a:prstGeom.*</a:prstGeom>'
<a:prstGeom prst="rect"><a:avLst><a:gd name="adj" fmla="val 14000"/></a:avLst></a:prstGeom>

$ officecli validate repro.pptx
Validation passed: no errors found.

ellipse behaves the same, and so does the other spelling — adj1 on rect or roundRect is accepted too. So both layers of defence you asked for are still missing: add writes a guide the preset does not declare, and validate stays green on it.

What makes this worth closing rather than debating is that the codebase already acts on your premise. Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs carries two rules of exactly this shape:

• R18 — "PowerPoint validates each <a:gd name="…"> against the names the preset's own definition declares; an unknown name (e.g. adj1 on a single-handle preset whose guide is literally adj) makes real PowerPoint refuse the file (0x80070570) even though the OpenXML SDK considers it schema-valid" — so the supplied name is remapped to the one the preset declares (CanonicalAdjName).
• R19b — a multi-guide preset must carry the complete declared guide set or PowerPoint refuses it, so 5 star/hexagon presets get their full guide list emitted (MultiGuidePresetDefaults).

Handle-less presets are the remaining hole: CanonicalAdjName knows only Donut/NoSmoking (single guide literally named adj), MultiGuidePresetDefaults knows only the 5 multi-guide shapes, and everything else falls through to "Unknown / single-guide preset: keep prior behavior — emit exactly what the user supplied". That is the branch that writes <a:gd name="adj"> into rect.

Intended contribution — one atomic change, but I'd like your call on the shape before I write it:

1. Widen that mechanism into a preset → declared-guide-names table, with the handle-less presets recorded as declares none, and refuse an adj= spec whose name the target preset does not declare. Presets the repo already handles (donut/noSmoking, the 5 multi-guide shapes) keep their current behavior; this closes the "declares none / name not declared" case only.
2. A validate rule flagging an existing <a:gd> whose prstGeom preset declares no such handle. This is the layer that catches files authored elsewhere — which is where "passes validate, PowerPoint repairs it" actually bites, and it is also the layer your first bullet asks for.

The design question the thread turns on, and where I'd like your ruling: R18/R19b read an undeclared guide name as refused by PowerPoint (hence remap / complete-the-set), while your 2026-07-19 reply reads an adjust value on a handle-less preset as ignored rather than invalid. If the lenient reading is the one you want for the zero-handle case, the minimal version is "drop the spec on a handle-less preset and warn" plus the validate rule; on the strict reading the write should error. I'd default to error + validate, because the repo's own comments already act on the strict reading — but that is your call, and either way validate should flag existing occurrences.

Two caveats, stated up front: this host has no .NET SDK, so I cannot build or run the repo's suite — I verify by inspection plus a behavioural mirror against the released binary, and I will say so on the patch itself. And I cannot run PowerPoint here, so the repair-dialog verdict stays with your macOS PowerPoint 16.110.3 measurement (repro235.pptx could not be fetched from this sandbox — the user-attachments URL times out here); the officecli-side defect above stands on its own and is what the patch would fix. The preset → handles table is the bulk of the work: there is no local copy of presetShapeDefinitions.xml here, so I would build it from the ECMA-376 preset definitions and state its provenance in the patch.

[argszero]

Taking your 2026-07-19 ruling as the scope, so this is the lint half only:

Adding a validate-level warning for adjust names not defined by the preset is still worth considering as a lint, but we'd like to confirm the corruption mechanism first before changing add/set behavior.

So add/set are untouched here — the write path still accepts geometry=rect --prop adj=adj:val 14000. What the patch adds is the missing observability: view issues now reports the undeclared value instead of going green on it. That is the half of the report that stands without PowerPoint (and it also covers geometry=roundRect --prop adj=adj1:…, the case your own R18 comment describes — accepted today, and invisible to both validate and view issues).

First, a correction to my own earlier comment. The command sequence I quoted there was incomplete: create makes a 0-slide deck, so /slide[1] rejects a shape (Error: Slide 1 not found (total: 0)) until a slide exists. The reproducible three-step sequence is:

$ officecli create repro.pptx
$ officecli add repro.pptx / --type slide
$ officecli add repro.pptx /slide[1] --type shape --prop geometry=rect --prop adj='adj:val 14000'
Added shape at /slide[1]/shape[@id=100000]

$ officecli view repro.pptx issues --json          # today: {"count": 0}
$ officecli validate repro.pptx                    # today: Validation passed: no errors found.  (rc 0)

I re-ran it that way against 1.0.151 before writing this, and ppt/slides/slide1.xml carries <a:prstGeom prst="rect"><a:avLst><a:gd name="adj" fmla="val 14000" /></a:avLst></a:prstGeom> — rect declares no adjustment value, validate stays green, view issues returns count: 0. Sorry for the missing step; the defect it demonstrates is unchanged.

What the patch does

One atomic change — flag an adjustment value the preset does not declare — in three files, +300/−2, 5 hunks:

• Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs — a generated preset → declared adjustment values table (187 rows), plus FindUndeclaredAdjustValues(prstGeom, out preset). It reads only prstGeom + its avLst.
• Handlers/Pptx/PowerPointHandler.View.cs — one warning per offending shape inside the existing per-shape lint loop (Id = A…, IssueType.Format, Severity.Warning), with the offending <a:gd> elements in Context and the fix in Suggestion.
• Core/IssueSubtypes.cs — the undeclared_adjust_value subtype, registered in ValidSubtypes so --type undeclared_adjust_value validates and filters, and added to the per-format help prose.

The table is generated from the definitions themselves, not recalled

A preset's <a:avLst> in presetShapeDefinitions.xml is its list of adjustment values (adj, adj1…adj8, plus the star/hexagon hf/vf); the computed guides in its <gdLst> are not authorable through prstGeom, so they are deliberately excluded. I did not write the table from memory — it is generated from the ECMA-376 definitions LibreOffice vendors verbatim at oox/source/drawingml/customshapes/presetShapeDefinitions.xml (538972 bytes, sha256 eaff19f4405b3be6822428c96ef46cc685499217366fd58e5e63a5a24739f02f): 187 presets, of which 64 declare no adjustment value at all (rect, ellipse, cloud, …). Those 64 are listed with an empty array rather than omitted, so "declares none" stays distinguishable from "preset unknown to this table" — an unknown token is never judged, so the lint fails open.

It is the same table, not a second one: CanonicalAdjName hand-codes donut/noSmoking and MultiGuidePresetDefaults hand-codes five shapes, and this generalises both to all 187. The agreement is checked mechanically in the verification below (the generated rows must equal the hand-written rows, name for name and in order) — that is also the evidence that the table and the file's existing entries come from the same spec revision. Keyed by the prstGeom@prst token, not the SDK enum, so it covers presets officecli cannot itself write but PowerPoint files can carry.

Why view issues and not validate

You asked for a validate-level warning, and I want to be explicit rather than quietly pick something else:

1. SKILL.md draws the line itself — validate is "Validate against OpenXML schema" (line 99) while issues is "Formatting/content/structure problems" (line 108). validate's extra detectors (DetectOutOfGridSheetRefs, DetectMissingDefaultRelsContentType) exist for files that are schema-legal but refused by the app — and your 2026-07-19 message records the opposite here: "adjust values in <a:avLst> that don't correspond to a handle of the preset are ignored rather than treated as invalid."
2. validate has no non-fatal tier: its list count is the exit code and the delivery gate (CommandBuilder.Check.cs, the resident path). Putting a lint there would turn a deck that opens fine into rc=1 — the behaviour change you explicitly deferred.
3. view issues already carries exactly this species of finding with Severity.Warning and a filterable subtype — low_contrast is the closest precedent ("reads fine on a laptop and vanishes on projection").

If you would rather have it inside validate, that needs a non-fatal tier added to ValidationError (plus its three consumers and an exit-code decision) — a separate atomic change from this one. Say the word and I will send it that way instead; nothing in this patch would be wasted, since the table and the detector are shared.

What a reviewer should see

$ officecli view repro.pptx issues
Found 1 issue(s):

Format Issues (1):
  [A1] /slide[1]/shape[@id=100000]: Adjustment value "adj" is not declared by preset "rect" (preset declares none) — PowerPoint ignores an undeclared adjustment value, so the geometry the author asked for never takes effect.
       Context: "<a:prstGeom prst="rect"><a:avLst><a:gd name="adj" fmla="val 14000"/></a:avLst>"
       Suggestion: Drop the undeclared value from the avLst, or use a preset that declares it.

and in JSON the same record with "subtype": "undeclared_adjust_value", "severity": 1, plus the two optional fields the issue sets. --type format, --type f and --type undeclared_adjust_value all select it; --type structure does not. validate is unchanged in every respect, including rc.

Deliberate limits, stated rather than implied: slides and their top-level shapes only — the same scope as the neighbouring lints in that loop; a subset of a multi-guide preset's declared set is not flagged (that is R19b's completeness concern, not a name problem, and flagging it would fire on decks PowerPoint handles); name comparison is ordinal, matching the definitions' case sensitivity.

Verification — and what I could not do

This host has no .NET SDK, so the patch is not compiled and the suite is not run: I cannot claim a green build, only that the change is textual + apply-verified + behaviour-mirrored. Four tiers, 153 assertions, 0 failures:

tier	what it proves	asserts
generator self-check (r15_235_make_patch.py)	the emitted table equals the spec's data row for row; the detector is total (no path throws) and reads the avLst only; every edit region is reversible byte-for-byte; the base bytes are origin/main bytes	46
patch application (r15_235_verify_patch.py)	git apply --check and patch -p1 --dry-run accept it; applying reproduces the intended sources byte-for-byte; reverse-applying restores the base byte-for-byte; the host tree, HEAD and status are untouched	16
behaviour mirror (r15_235_behavior_mirror.py)	the rule re-implemented from the patched C# and run over ten decks written by the released 1.0.151 plus 15 synthetic avLst shapes; the generated rows match CanonicalAdjName / MultiGuidePresetDefaults exactly	58
outcome demo (r15_235_outcome_demo.py)	today's real output on this deck (issue count 0, validate rc 0) vs the record the patch emits, with the filter contract and the untouched-validate claim checked against the patch itself	33

The false-positive question is settled empirically, which is the part I would want checked hardest: the repo's own writer never emits an undeclared name except in exactly the two defect cases. Across the ten decks the mirror flags 3 — rect+adj, ellipse+adj (this issue) and roundRect+adj1 (R18's case) — and stays silent on the other seven, including star5 (adj, hf, vf) and hexagon (adj, vf), where MultiGuidePresetDefaults already completes the declared set. So the lint agrees with the writer rather than fighting it.

To confirm on a machine with the SDK: dotnet build, then officecli create repro.pptx, add … / --type slide, add /slide[1] --type shape --prop geometry=rect --prop adj='adj:val 14000', view repro.pptx issues → 1 warning; and view issues on a deck using geometry=roundRect --prop adj=adj1:val 10000 → 0. r15-235-undeclared-adjust-value.patch, 21304 bytes, sha256 7758f7d56ce48c32…, 300 added / 2 removed lines.

diff --git a/src/officecli/Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs b/src/officecli/Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs
index 26736c0b14ab..688d7d71bb7b 100644
--- a/src/officecli/Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs
+++ b/src/officecli/Handlers/Pptx/PowerPointHandler.Helpers.Geometry.cs
@@ -417,6 +417,248 @@
         };
 
     /// <summary>
+    /// The adjustment values each ECMA-376 preset shape definition declares,
+    /// keyed by the <c>prstGeom@prst</c> token. A preset's own
+    /// <c>&lt;avLst&gt;</c> in presetShapeDefinitions.xml <em>is</em> its list of
+    /// adjustment values (<c>adj</c>, <c>adj1</c>…<c>adj8</c>, plus the star /
+    /// hexagon <c>hf</c>/<c>vf</c>); the computed guides in its
+    /// <c>&lt;gdLst&gt;</c> are not authorable through <c>prstGeom</c> and are
+    /// deliberately excluded. The 64 definitions that declare none at all
+    /// (<c>rect</c>, <c>ellipse</c>, <c>cloud</c>, …) are listed with an empty
+    /// array rather than omitted, so "declares none" stays distinguishable from
+    /// "preset unknown to this table".
+    ///
+    /// Generated from the spec's own table rather than recalled. LibreOffice
+    /// vendors the ECMA-376 definitions verbatim at
+    /// <c>oox/source/drawingml/customshapes/presetShapeDefinitions.xml</c>
+    /// (sha256 <c>eaff19f4405b3be6822428c96ef46cc685499217366fd58e5e63a5a24739f02f</c>,
+    /// 538972 bytes); every row below is the <c>&lt;gd name&gt;</c> list of that
+    /// preset's own <c>&lt;avLst&gt;</c>. This is the same knowledge
+    /// <see cref="CanonicalAdjName"/> hand-codes for donut/noSmoking and
+    /// <see cref="MultiGuidePresetDefaults"/> hand-codes for five shapes,
+    /// completed for all 187.
+    /// </summary>
+    private static readonly IReadOnlyDictionary<string, string[]> PresetDeclaredAdjustValues =
+        new Dictionary<string, string[]>(StringComparer.Ordinal)
+        {
+            ["accentBorderCallout1"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["accentBorderCallout2"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6" },
+            ["accentBorderCallout3"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6", "adj7", "adj8" },
+            ["accentCallout1"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["accentCallout2"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6" },
+            ["accentCallout3"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6", "adj7", "adj8" },
+            ["actionButtonBackPrevious"] = Array.Empty<string>(),
+            ["actionButtonBeginning"] = Array.Empty<string>(),
+            ["actionButtonBlank"] = Array.Empty<string>(),
+            ["actionButtonDocument"] = Array.Empty<string>(),
+            ["actionButtonEnd"] = Array.Empty<string>(),
+            ["actionButtonForwardNext"] = Array.Empty<string>(),
+            ["actionButtonHelp"] = Array.Empty<string>(),
+            ["actionButtonHome"] = Array.Empty<string>(),
+            ["actionButtonInformation"] = Array.Empty<string>(),
+            ["actionButtonMovie"] = Array.Empty<string>(),
+            ["actionButtonReturn"] = Array.Empty<string>(),
+            ["actionButtonSound"] = Array.Empty<string>(),
+            ["arc"] = new[] { "adj1", "adj2" },
+            ["bentArrow"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["bentConnector2"] = Array.Empty<string>(),
+            ["bentConnector3"] = new[] { "adj1" },
+            ["bentConnector4"] = new[] { "adj1", "adj2" },
+            ["bentConnector5"] = new[] { "adj1", "adj2", "adj3" },
+            ["bentUpArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["bevel"] = new[] { "adj" },
+            ["blockArc"] = new[] { "adj1", "adj2", "adj3" },
+            ["borderCallout1"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["borderCallout2"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6" },
+            ["borderCallout3"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6", "adj7", "adj8" },
+            ["bracePair"] = new[] { "adj" },
+            ["bracketPair"] = new[] { "adj" },
+            ["callout1"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["callout2"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6" },
+            ["callout3"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5", "adj6", "adj7", "adj8" },
+            ["can"] = new[] { "adj" },
+            ["chartPlus"] = Array.Empty<string>(),
+            ["chartStar"] = Array.Empty<string>(),
+            ["chartX"] = Array.Empty<string>(),
+            ["chevron"] = new[] { "adj" },
+            ["chord"] = new[] { "adj1", "adj2" },
+            ["circularArrow"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5" },
+            ["cloud"] = Array.Empty<string>(),
+            ["cloudCallout"] = new[] { "adj1", "adj2" },
+            ["corner"] = new[] { "adj1", "adj2" },
+            ["cornerTabs"] = Array.Empty<string>(),
+            ["cube"] = new[] { "adj" },
+            ["curvedConnector2"] = Array.Empty<string>(),
+            ["curvedConnector3"] = new[] { "adj1" },
+            ["curvedConnector4"] = new[] { "adj1", "adj2" },
+            ["curvedConnector5"] = new[] { "adj1", "adj2", "adj3" },
+            ["curvedDownArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["curvedLeftArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["curvedRightArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["curvedUpArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["decagon"] = new[] { "vf" },
+            ["diagStripe"] = new[] { "adj" },
+            ["diamond"] = Array.Empty<string>(),
+            ["dodecagon"] = Array.Empty<string>(),
+            ["donut"] = new[] { "adj" },
+            ["doubleWave"] = new[] { "adj1", "adj2" },
+            ["downArrow"] = new[] { "adj1", "adj2" },
+            ["downArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["ellipse"] = Array.Empty<string>(),
+            ["ellipseRibbon"] = new[] { "adj1", "adj2", "adj3" },
+            ["ellipseRibbon2"] = new[] { "adj1", "adj2", "adj3" },
+            ["flowChartAlternateProcess"] = Array.Empty<string>(),
+            ["flowChartCollate"] = Array.Empty<string>(),
+            ["flowChartConnector"] = Array.Empty<string>(),
+            ["flowChartDecision"] = Array.Empty<string>(),
+            ["flowChartDelay"] = Array.Empty<string>(),
+            ["flowChartDisplay"] = Array.Empty<string>(),
+            ["flowChartDocument"] = Array.Empty<string>(),
+            ["flowChartExtract"] = Array.Empty<string>(),
+            ["flowChartInputOutput"] = Array.Empty<string>(),
+            ["flowChartInternalStorage"] = Array.Empty<string>(),
+            ["flowChartMagneticDisk"] = Array.Empty<string>(),
+            ["flowChartMagneticDrum"] = Array.Empty<string>(),
+            ["flowChartMagneticTape"] = Array.Empty<string>(),
+            ["flowChartManualInput"] = Array.Empty<string>(),
+            ["flowChartManualOperation"] = Array.Empty<string>(),
+            ["flowChartMerge"] = Array.Empty<string>(),
+            ["flowChartMultidocument"] = Array.Empty<string>(),
+            ["flowChartOfflineStorage"] = Array.Empty<string>(),
+            ["flowChartOffpageConnector"] = Array.Empty<string>(),
+            ["flowChartOnlineStorage"] = Array.Empty<string>(),
+            ["flowChartOr"] = Array.Empty<string>(),
+            ["flowChartPredefinedProcess"] = Array.Empty<string>(),
+            ["flowChartPreparation"] = Array.Empty<string>(),
+            ["flowChartProcess"] = Array.Empty<string>(),
+            ["flowChartPunchedCard"] = Array.Empty<string>(),
+            ["flowChartPunchedTape"] = Array.Empty<string>(),
+            ["flowChartSort"] = Array.Empty<string>(),
+            ["flowChartSummingJunction"] = Array.Empty<string>(),
+            ["flowChartTerminator"] = Array.Empty<string>(),
+            ["foldedCorner"] = new[] { "adj" },
+            ["frame"] = new[] { "adj1" },
+            ["funnel"] = Array.Empty<string>(),
+            ["gear6"] = new[] { "adj1", "adj2" },
+            ["gear9"] = new[] { "adj1", "adj2" },
+            ["halfFrame"] = new[] { "adj1", "adj2" },
+            ["heart"] = Array.Empty<string>(),
+            ["heptagon"] = new[] { "hf", "vf" },
+            ["hexagon"] = new[] { "adj", "vf" },
+            ["homePlate"] = new[] { "adj" },
+            ["horizontalScroll"] = new[] { "adj" },
+            ["irregularSeal1"] = Array.Empty<string>(),
+            ["irregularSeal2"] = Array.Empty<string>(),
+            ["leftArrow"] = new[] { "adj1", "adj2" },
+            ["leftArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["leftBrace"] = new[] { "adj1", "adj2" },
+            ["leftBracket"] = new[] { "adj" },
+            ["leftCircularArrow"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5" },
+            ["leftRightArrow"] = new[] { "adj1", "adj2" },
+            ["leftRightArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["leftRightCircularArrow"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5" },
+            ["leftRightRibbon"] = new[] { "adj1", "adj2", "adj3" },
+            ["leftRightUpArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["leftUpArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["lightningBolt"] = Array.Empty<string>(),
+            ["line"] = Array.Empty<string>(),
+            ["lineInv"] = Array.Empty<string>(),
+            ["mathDivide"] = new[] { "adj1", "adj2", "adj3" },
+            ["mathEqual"] = new[] { "adj1", "adj2" },
+            ["mathMinus"] = new[] { "adj1" },
+            ["mathMultiply"] = new[] { "adj1" },
+            ["mathNotEqual"] = new[] { "adj1", "adj2", "adj3" },
+            ["mathPlus"] = new[] { "adj1" },
+            ["moon"] = new[] { "adj" },
+            ["noSmoking"] = new[] { "adj" },
+            ["nonIsoscelesTrapezoid"] = new[] { "adj1", "adj2" },
+            ["notchedRightArrow"] = new[] { "adj1", "adj2" },
+            ["octagon"] = new[] { "adj" },
+            ["parallelogram"] = new[] { "adj" },
+            ["pentagon"] = new[] { "hf", "vf" },
+            ["pie"] = new[] { "adj1", "adj2" },
+            ["pieWedge"] = Array.Empty<string>(),
+            ["plaque"] = new[] { "adj" },
+            ["plaqueTabs"] = Array.Empty<string>(),
+            ["plus"] = new[] { "adj" },
+            ["quadArrow"] = new[] { "adj1", "adj2", "adj3" },
+            ["quadArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["rect"] = Array.Empty<string>(),
+            ["ribbon"] = new[] { "adj1", "adj2" },
+            ["ribbon2"] = new[] { "adj1", "adj2" },
+            ["rightArrow"] = new[] { "adj1", "adj2" },
+            ["rightArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["rightBrace"] = new[] { "adj1", "adj2" },
+            ["rightBracket"] = new[] { "adj" },
+            ["round1Rect"] = new[] { "adj" },
+            ["round2DiagRect"] = new[] { "adj1", "adj2" },
+            ["round2SameRect"] = new[] { "adj1", "adj2" },
+            ["roundRect"] = new[] { "adj" },
+            ["rtTriangle"] = Array.Empty<string>(),
+            ["smileyFace"] = new[] { "adj" },
+            ["snip1Rect"] = new[] { "adj" },
+            ["snip2DiagRect"] = new[] { "adj1", "adj2" },
+            ["snip2SameRect"] = new[] { "adj1", "adj2" },
+            ["snipRoundRect"] = new[] { "adj1", "adj2" },
+            ["squareTabs"] = Array.Empty<string>(),
+            ["star10"] = new[] { "adj", "hf" },
+            ["star12"] = new[] { "adj" },
+            ["star16"] = new[] { "adj" },
+            ["star24"] = new[] { "adj" },
+            ["star32"] = new[] { "adj" },
+            ["star4"] = new[] { "adj" },
+            ["star5"] = new[] { "adj", "hf", "vf" },
+            ["star6"] = new[] { "adj", "hf" },
+            ["star7"] = new[] { "adj", "hf", "vf" },
+            ["star8"] = new[] { "adj" },
+            ["straightConnector1"] = Array.Empty<string>(),
+            ["stripedRightArrow"] = new[] { "adj1", "adj2" },
+            ["sun"] = new[] { "adj" },
+            ["swooshArrow"] = new[] { "adj1", "adj2" },
+            ["teardrop"] = new[] { "adj" },
+            ["trapezoid"] = new[] { "adj" },
+            ["triangle"] = new[] { "adj" },
+            ["upArrow"] = new[] { "adj1", "adj2" },
+            ["upArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["upDownArrow"] = new[] { "adj1", "adj2" },
+            ["upDownArrowCallout"] = new[] { "adj1", "adj2", "adj3", "adj4" },
+            ["uturnArrow"] = new[] { "adj1", "adj2", "adj3", "adj4", "adj5" },
+            ["verticalScroll"] = new[] { "adj" },
+            ["wave"] = new[] { "adj1", "adj2" },
+            ["wedgeEllipseCallout"] = new[] { "adj1", "adj2" },
+            ["wedgeRectCallout"] = new[] { "adj1", "adj2" },
+            ["wedgeRoundRectCallout"] = new[] { "adj1", "adj2", "adj3" },
+        };
+
+    /// <summary>
+    /// Names in <paramref name="prstGeom"/>'s avLst that the preset does not
+    /// declare, in document order and without duplicates — the adjustment values
+    /// real PowerPoint accepts but ignores, because the preset's definition has
+    /// no such value for them to set. Empty when every name is declared, when
+    /// there is no avLst, or when the preset is absent from
+    /// <see cref="PresetDeclaredAdjustValues"/>: an unknown preset cannot be
+    /// judged from this table, and a lint must not guess. Comparison is ordinal —
+    /// the definitions are case-sensitive.
+    /// </summary>
+    internal static List<string> FindUndeclaredAdjustValues(
+        Drawing.PresetGeometry? prstGeom, out string presetToken)
+    {
+        presetToken = prstGeom?.Preset?.InnerText ?? "";
+        var undeclared = new List<string>();
+        if (prstGeom == null || presetToken.Length == 0) return undeclared;
+        if (!PresetDeclaredAdjustValues.TryGetValue(presetToken, out var declared)) return undeclared;
+        var avLst = prstGeom.GetFirstChild<Drawing.AdjustValueList>();
+        if (avLst == null) return undeclared;
+        foreach (var gd in avLst.Elements<Drawing.ShapeGuide>())
+        {
+            var name = gd.Name?.Value;
+            if (string.IsNullOrEmpty(name) || undeclared.Contains(name)) continue;
+            if (!declared.Contains(name)) undeclared.Add(name);
+        }
+        return undeclared;
+    }
+
+    /// <summary>
     /// Map the adjust-handle name at <paramref name="index"/> to the name the
     /// given <paramref name="preset"/> actually declares. Presets that define a
     /// single adjust handle name it <c>adj</c> (donut, noSmoking, …); writing the
diff --git a/src/officecli/Handlers/Pptx/PowerPointHandler.View.cs b/src/officecli/Handlers/Pptx/PowerPointHandler.View.cs
index 476385b4915c..b195be93cdb5 100644
--- a/src/officecli/Handlers/Pptx/PowerPointHandler.View.cs
+++ b/src/officecli/Handlers/Pptx/PowerPointHandler.View.cs
@@ -762,6 +762,51 @@
                         Severity = IssueSeverity.Warning,
                         Path = shapePath,
                         Message = overflow
+                    });
+                }
+
+                // Adjustment values the preset does not declare (#235). `add` /
+                // `set geometry=` copy an `adj=name:fmla` name straight into the
+                // avLst unless the preset is one CanonicalAdjName knows
+                // (donut/noSmoking) or one MultiGuidePresetDefaults covers, so
+                // `geometry=rect --prop adj=adj:val 14000` writes <a:gd
+                // name="adj"/> into a preset that declares no adjustment value at
+                // all, and `geometry=roundRect --prop adj=adj1:…` writes the wrong
+                // name for the single value roundRect does declare. ECMA-376
+                // leaves an undeclared value unused rather than invalid: the file
+                // still opens and the schema validator (so `validate`) stays
+                // green, while the geometry the author asked for silently never
+                // materialises. Warning, not Error — a lint, not a gate.
+                var prstGeom = shape.ShapeProperties?.GetFirstChild<Drawing.PresetGeometry>();
+                var undeclaredAdj = FindUndeclaredAdjustValues(prstGeom, out var adjPresetToken);
+                if (undeclaredAdj.Count > 0)
+                {
+                    PresetDeclaredAdjustValues.TryGetValue(adjPresetToken, out var declaredAdj);
+                    var declaredHint = declaredAdj is { Length: > 0 }
+                        ? $"preset declares {string.Join(", ", declaredAdj)}"
+                        : "preset declares none";
+                    // Context carries only the offending <a:gd> elements, not the
+                    // whole avLst, so the report points at the defect.
+                    var gdXml = string.Join("",
+                        (prstGeom?.GetFirstChild<Drawing.AdjustValueList>()?.Elements<Drawing.ShapeGuide>()
+                            ?? Enumerable.Empty<Drawing.ShapeGuide>())
+                        .Where(g => undeclaredAdj.Contains(g.Name?.Value ?? ""))
+                        .Select(g => $"<a:gd name=\"{g.Name?.Value}\" fmla=\"{g.Formula?.Value}\"/>"));
+                    issues.Add(new DocumentIssue
+                    {
+                        Id = $"A{++issueNum}",
+                        Type = IssueType.Format,
+                        Subtype = Core.IssueSubtypes.UndeclaredAdjustValue,
+                        Severity = IssueSeverity.Warning,
+                        Path = shapePath,
+                        Message = $"Adjustment value{(undeclaredAdj.Count > 1 ? "s" : "")} "
+                                + string.Join(", ", undeclaredAdj.Select(n => $"\"{n}\""))
+                                + $" {(undeclaredAdj.Count > 1 ? "are" : "is")} not declared by preset "
+                                + $"\"{adjPresetToken}\" ({declaredHint})"
+                                + " — PowerPoint ignores an undeclared adjustment value, so the"
+                                + " geometry the author asked for never takes effect.",
+                        Context = $"<a:prstGeom prst=\"{adjPresetToken}\"><a:avLst>{gdXml}</a:avLst>",
+                        Suggestion = "Drop the undeclared value from the avLst, or use a preset that declares it."
                     });
                 }
 
diff --git a/src/officecli/Core/IssueSubtypes.cs b/src/officecli/Core/IssueSubtypes.cs
index d3c39ec4776b..7a78fb75e190 100644
--- a/src/officecli/Core/IssueSubtypes.cs
+++ b/src/officecli/Core/IssueSubtypes.cs
@@ -55,6 +55,16 @@
     /// translucent runs, and colors carrying lumMod/shade transforms are
     /// skipped to keep false positives near zero. Format bucket, Warning.</summary>
     public const string LowContrast = "low_contrast";
+    /// <summary>pptx-only: a preset geometry's <c>&lt;a:avLst&gt;</c> names an
+    /// adjustment value the preset's own ECMA-376 definition does not declare —
+    /// an <c>&lt;a:gd&gt;</c> the preset has no value for. PowerPoint accepts and
+    /// then ignores it, so the geometry silently renders with the preset default
+    /// instead of the authored value, while the schema (and therefore
+    /// <c>validate</c>) stays green. Reachable from the CLI itself:
+    /// <c>add … --type shape --prop geometry=rect --prop adj=adj:val 14000</c>
+    /// writes an undeclared <c>adj</c> into <c>rect</c>, which declares no
+    /// adjustment value. Format bucket, Warning.</summary>
+    public const string UndeclaredAdjustValue = "undeclared_adjust_value";
 
     /// <summary>Broad IssueType bucket names — the canonical surface shown
     /// in error messages and help. Single-letter aliases (<see cref="BucketAliases"/>)
@@ -78,7 +88,7 @@
     {
         FormulaNotEvaluated, FormulaCacheStale, FormulaRefMissingSheet, FormulaEvalError,
         FieldNotEvaluated, FieldCacheStale,
-        SlideFieldNotEvaluated, NotesUnresolvedRid, LowContrast,
+        SlideFieldNotEvaluated, NotesUnresolvedRid, LowContrast, UndeclaredAdjustValue,
         ChartSeriesRefMissingSheet, ChartCacheStale,
         DefinedNameBroken, DefinedNameTargetMissing,
         BrokenPartRef, NumericOverflow, GeneralPrecisionLoss,
@@ -102,7 +112,8 @@
             + "Opt-in only (request by exact name; not included in --type content): "
             + string.Join(", ", OptInSubtypes) + ". "
             + "Subtypes are format-specific — formula_* / chart_* / definedname_* / numeric_overflow apply to xlsx, "
-            + "field_* to docx, slide_field_* / notes_unresolved_rid / broken_part_ref / low_contrast to pptx; requesting a subtype that does not apply to "
+            + "field_* to docx, slide_field_* / notes_unresolved_rid / broken_part_ref / low_contrast / "
+            + "undeclared_adjust_value to pptx; requesting a subtype that does not apply to "
             + "the queried file returns count=0 (not an error). "
             + "All values are case-insensitive and surrounding whitespace is trimmed.";
     }

If you want the write-side guard (add/set refusing or dropping an undeclared name) after you have the corruption mechanism pinned down, this table is the piece it needs — it would be a few lines on top of ApplyAdjustHandles. And the same split exists for xlsx, where ExcelHandler writes prstGeom too; happy to extend the lint there as a separate change if it is wanted.

[argszero]

The lint half is now implemented, built and pushed, so this no longer rests on my earlier description of it: #422 (fix/undeclared-adjust-value, 3 files, +302/−1, branched off current main).

Scope — unchanged from your 2026-07-19 ruling. add / set and validate behave exactly as before; what changes is that the defect is now visible.

$ officecli create repro.pptx
$ officecli add repro.pptx / --type slide
$ officecli add repro.pptx /slide[1] --type shape --prop geometry=rect --prop adj=adj:val 14000

$ officecli view repro.pptx issues          # before this change: "Found 0 issue(s)"
Found 1 issue(s):

Format Issues (1):
  [A1] /slide[1]/shape[@id=100000]: Adjustment value "adj" is not declared by preset "rect" (preset declares none) — PowerPoint ignores an undeclared adjustment value, so the geometry the author asked for never takes effect.
       Context: "<a:prstGeom prst="rect"><a:avLst><a:gd name="adj" fmla="val 14000"/></a:avLst>"
       Suggestion: Drop the undeclared value from the avLst, or use a preset that declares it.

$ officecli validate repro.pptx             # unchanged
Validation passed: no errors found.          (exit code 0)

The subtype is undeclared_adjust_value, Warning, Format bucket, so --type undeclared_adjust_value and --type format / f select it and --type structure does not. It is registered alongside low_contrast / picture_aspect_distorted, the same species of finding. The written file is byte-identical to what the current release writes, and validate's exit code is untouched.

Correction to my earlier comment. It said the patch was "textual + apply-verified + behaviour-mirrored" with "no .NET SDK", so I could not claim a green build. That no longer holds — .NET 10 is available on this host now, so this is compiled and run: dotnet build -c Release gives 0 errors and only the 2 pre-existing warnings. I built the same commit twice — unpatched and with the change — and diffed the two binaries' behaviour directly rather than inferring it.

Measured against those two builds:

• 15-scenario A/B (undeclared names, correct names, multi-value specs, empty avLst, multi-guide presets): the unpatched binary fails 5 scenarios — it writes the value and reports 0 issues — and the patched one passes all 15. In every scenario the add output and the on-disk <a:gd> set are identical between the two binaries, which is the evidence that only the lint was added.
• Sweep over all 63 examples/**/*.pptx (3621 prstGeom shapes): both binaries report the exact same 1976 findings. The corpus does contain 5 real roundRect adjustment values and neither binary flags them, so this is a non-vacuous zero-false-positive result rather than an empty corpus.
• The preset → declared values table is re-derived from presetShapeDefinitions.xml (538972 bytes, sha256 eaff19f4…) and checked to equal the rows in the PR exactly, in order; the rows for donut/noSmoking and the five MultiGuidePresetDefaults shapes must equal the names those functions already hand-code — same spec revision, not a second opinion. All 187 presets are covered, 64 of them declaring none; an unknown prst token is never judged (fail-open), and the 63-deck sweep confirms nothing fires on real files.

One implementation note versus my earlier comment: picture_aspect_distorted landed in between, so the new subtype is registered after it rather than after low_contrast; the table, the detector and the message are as described.