fix(selectors): de-blind the diagnostics, and fix the two selectors hiding behind them

CHANGELOG.md

@@ -9,6 +9,32 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- **Flow's own diagnostics were blind on the migrated host, and two selectors were broken behind
+  it.** `gflow`'s incident-bundle DOM dump queried `i.google-symbols, span.google-symbols`, and
+  `diagnostics.py`'s `STRUCTURAL_DOM_JS` queried `i.google-symbols` — so on `flow.google.com`,
+  where every Material Symbols ligature rides a `<mat-icon>`, **both reported zero ligatures**.
+  The instrument used to diagnose selector drift was blind to the host the drift lives on. That
+  is why [#727](https://github.com/ffroliva/gflow-cli/issues/727) and
+  [#731](https://github.com/ffroliva/gflow-cli/issues/731) stayed invisible, and why an incident
+  bundle from a migrated user named no ligatures at all. Both queries are now class-only.
+
+  Two selectors were broken behind that blindness. `SUBMIT_BUTTON_SELECTORS` was **working by
+  luck**: its two tag-qualified entries matched nothing on the migrated host and the submit only
+  landed because a third `has-text` entry matches the `<mat-icon>`'s *text* — observed live as
+  `prompt_submitted via="button:has-text('arrow_forward')"`, after paying two misses per submit.
+  `IMAGE_MODEL_PICKER_TRIGGER` had neither twin nor fallback, so its miss was total and silent:
+  the picker is best-effort, so generation simply proceeded on whatever model tier the editor
+  opened at. Both now anchor on the `google-symbols` **class**, which sits on the labs `<i>` and
+  the migrated `<mat-icon>` alike — verified against a real CSS engine, offline, in
+  `tests/api/transports/test_ligature_carrier.py`.
+
+  `SUBMIT_BUTTON_SELECTORS` also moves from `:text` to **`:text-is`**. `:text` is a substring
+  match and accepts `arrow_forward_ios`, a real Material Symbol; the `<i>` qualifier had been
+  containing that, and dropping it for the class-only carrier made the over-match reachable on
+  the submit path. The two-carrier fixture caught it as `matched 3 of 2` before it ran live.
+  ([#730](https://github.com/ffroliva/gflow-cli/issues/730),
+  [spike](docs/superpowers/spikes/2026-09-07-ligature-carrier-and-name-drift.md))
+
 - **`character create --format-prompt` works again — both halves of it.** The flag had two
   independent faults, and fixing only the first would have left it just as useless. On the
   migrated `flow.google.com` host

docs/superpowers/spikes/2026-09-07-ligature-carrier-and-name-drift.md

@@ -0,0 +1,113 @@
+# The carrier is one character. The ligature NAME is the other half nobody counted
+
+**Date:** 2026-09-07 · **Issue:** [#730](https://github.com/ffroliva/gflow-cli/issues/730) ·
+**Account:** `denon82` (migrated `flow.google.com`) · **Cost:** `$0` — navigation, one free
+`createEntity` (deleted again), DOM reads. Nothing submitted, no credit, no image quota.
+
+**Script:** `scripts/dev/spike_ligature_carrier_sweep.py` ·
+**Capture:** `scripts/dev/_spike_out/ligature_carrier_sweep_denon82_20260907_*.json` (gitignored)
+
+---
+
+## 1. What was asked
+
+#727 and #731 both came from the labs/migrated **carrier split**: the same Material Symbols
+ligature rendered in `<i class="google-symbols">` on labs and `<mat-icon>` on the migrated
+host. A static sweep of `src/gflow_cli` found the real unit is the *constant*, not the literal:
+
+| | count |
+|---|---|
+| constants covering **both** carriers | 6 |
+| constants that are **`<i>`-only** | **24** |
+
+(#730 says "55 single-carrier literals". That number counts literals, and double-counts
+cascades that alternate carriers across entries. 24 constants is the actionable figure.)
+
+Static analysis cannot say which of the 24 *matter*. A selector whose surface does not exist
+on this host is not a bug; one whose miss is rescued by a fallback is a bug that is currently
+invisible. So the probe measured, per ligature, on two live surfaces.
+
+## 2. The carrier result — unambiguous
+
+Control: `prompt_boxes=1`, `ligature_hits=8`. The probe was standing on a real editor, so its
+zeros are about selectors, not arrival.
+
+```
+=== character editor ===
+ligature                i.gs   mat   .gs   any  carriers
+accessibility_new          0     1     1     1  mat-icon
+add                        0     2     2     2  mat-icon
+arrow_drop_down            0     1     1     1  mat-icon
+arrow_forward              0     1     1     1  mat-icon
+close                      0     1     1     1  mat-icon
+upload                     0     1     1     1  mat-icon
+voice_selection            0     1     1     1  mat-icon
+```
+
+**`i.google-symbols` matched zero, for every ligature, on both surfaces.** `mat-icon` matched,
+and **`.google-symbols` matched exactly the same count as `mat-icon`, every time**.
+
+The class is on both carriers. `.google-symbols:text-is(L)` therefore covers labs *and*
+migrated with no cascade entry at all — the fix for the 24 constants is **one character each**
+(`i.google-symbols` → `.google-symbols`), not 24 duplicated entries.
+
+Two selectors are confirmed broken-but-masked:
+
+- `SUBMIT_BUTTON_SELECTORS` — `arrow_forward` is `i.gs=0 / mat=1`. Both `<i>` entries miss;
+  only the third, `button:has-text('arrow_forward')`, fires, matching the `<mat-icon>`'s *text*
+  rather than its tag. Independently observed live the same day:
+  `prompt_submitted via="button:has-text('arrow_forward')"`. **It works by luck.**
+- `IMAGE_MODEL_PICKER_TRIGGER` — `arrow_drop_down` is `i.gs=0 / mat=1`, and the constant has no
+  twin and no text fallback.
+
+## 3. The finding that a carrier-only fix would have shipped past
+
+**The migrated host renders `add`. It does not render `add_2` anywhere.**
+
+```
+composer:  add = 1     add_2 = 0
+editor:    add = 2     add_2 = 0
+```
+
+That is a **ligature NAME change, not a carrier change.** Consequences:
+
+- `NEW_PROJECT_SELECTORS` (`add_2`) misses — which is why the first run of this very spike died
+  with `Could not find 'New project' CTA on Flow gallery` on `flow.google.com`.
+- `_CHARACTER_SLOT_ADD_SELECTOR` misses **even though it already covers both carriers**. It was
+  swept for the carrier in #703 and is still broken on this host.
+- `ADD_MEDIA_BUTTON` (`add_2`) is in the same position.
+
+Changing the carrier on those three would have produced a green sweep, a satisfied checklist,
+and three selectors that still match nothing. **The carrier was never the whole story; it was
+just the half that had a name.**
+
+## 4. What was NOT measured
+
+- **Ligatures absent from both surfaces are NOT proven absent from the product**:
+  `add_2`, `cancel`, `clear`, `edit_square`, `image`, `play_circle`, `tune`, `article_spark`,
+  `chrome_extension`, `crop_free` and most `crop_*` did not appear. They plausibly live behind
+  menus, dialogs or modes this probe never opened. The probe read two surfaces in their default
+  state, and that is all it can speak to. `add_2` is the exception only because a control that
+  *should* be on the surface it was probed on (`New project`, `slot add`) was independently
+  observed to fail.
+- **labs was not measured at all.** No unmoved account exists here. The claim that
+  `.google-symbols` also matches `<i class="google-symbols">` is CSS semantics, not an
+  observation — it should be verified on labs before anyone calls this closed.
+- **Whether broadening `i.google-symbols` → `.google-symbols` changes `.first` picks on labs.**
+  A wider match can select a different element where a cascade relies on document order.
+
+## 5. The rule this earns
+
+```
+SWEEPING FOR THE FAILURE YOU ALREADY NAMED FINDS ONLY THAT FAILURE.
+```
+
+The probe was built to compare two carriers. It found the name drift only because it also
+counted whether the ligature was present *at all* — the control column, added so a zero could
+not be misread as absence. That column, added for epistemic hygiene, is what turned up the
+second defect class.
+
+See also:
+[2026-09-07 — the Format button's anchor](2026-09-07-character-format-button-anchor.md),
+[2026-09-07 — a click is not an effect](2026-09-07-format-click-is-not-a-format.md),
+[`skills/spike/SKILL.md`](../../../skills/spike/SKILL.md).