diff --git a/MATH-FIX-REPORT.md b/MATH-FIX-REPORT.md
index 9e170aa..c541dbd 100644
--- a/MATH-FIX-REPORT.md
+++ b/MATH-FIX-REPORT.md
@@ -1172,3 +1172,28 @@ Harness note: as in passes 39–42, the sandboxed shell routes `TMPDIR` to a rea
 **No code change required** — engine/renderer/i18n/tests byte-stable vs pass 42, deployed build verified identical 34/34. All plan acceptance criteria (1–6) re-verified against the deployed build. Final: **MATH_FIX_DONE**.
 
 Addendum (post-push, same pass): audit commit `d1a35db15396495f69ca06c732412602f00f690d` pushed `1fad222..d1a35db` → `origin/main` (report + fresh off-LAN stamp `2026-10-09T23:38:48Z`; code untouched — the plan's mandated fix commit `08398267040a832bdec0910626423f0dbfeb0a7f` remains the code change under test, confirmed ancestor of `HEAD`). Post-push `git fetch` + full byte sweep re-run vs the new `origin/main`: `git diff 1fad222..d1a35db -- site` = **0 lines** (zero site delta, so the sweep carries over by construction), origin blob == local == webroot md5, tracked **21/21 (3-way)** + gitignored `site/vision/*.jpeg` **13/13 (2-way)** = **34/34 match, 0 mismatch** → `SWEEP-OK`; `HEAD` == `origin/main` == `d1a35db`.
+
+---
+
+## Re-verification (forty-fourth DSH pass, 2026-10-09 23:41–23:47 UTC, post-completion)
+
+Queue gate re-checked at start: tree clean at HEAD `3854fb4` — **all four fixes are code-complete on `main`** (fix commit `08398267040a832bdec0910626423f0dbfeb0a7f`, ancestor of HEAD confirmed; `git diff 0839826..HEAD -- site` = 0 lines carried through the audit chain). Re-audited every mandated change against the live files by direct read: GIS 65+ floor at the three `solve()` call sites (`engine.js:249/260/265`, `var g = a >= 65 ? gis(…) : 0`) + Service Canada source comment (`:88–91`) + merge gate `if (yr < 65) gisSum = 0;` (`:382`) with per-partner age≥65 gates (`:376–381`); clawback only when OAS received, capped at OAS paid, in both `solve` legs (`:248/:259`, s.3(1) ETA comment `:80–83`) and the splitting comparison (`:366–367`), `row.oasClawback = r.cw` (`:302`) + merge `row.oasClawback = clawSum` (`:371`); `row.mixed` flag (`:390`) + depletion guard `!row.mixed` (`:395`) + renderer `†` and `#yr-mixed-note` legend (`app.js:812/817–819`, `index.html:95`) with `t.mixedNote` EN/FR (`i18n.js:251/512`); FIRE: quiz floor 40 (`app.js:211`) + FIRE hint EN/FR (`i18n.js:96/357`), what-if `min="40"` (`index.html:86`), `fillDefaults` stale-save clamp `retireAge < age+1 → age+1` (`engine.js:188`). **Zero code changes required.** Full acceptance sweep re-run fresh against the deployed build:
+
+| Gate | Result (fresh this pass) |
+|---|---|
+| `tests/engine.test.js` | `RESULT: 61 pass, 0 fail` (36 base + math-fix sections incl. `T1 clawback == 0 at 55 (pre-OAS; old: −$3,501 net OAS)`, `T1 GIS == 0 at 56 (old: $12,916 phantom)`, `100k RRSP draw at 55 → clawback 0` + `67y OAS recipient, 150k draw → clawback > 0` + cap, `fillDefaults({age:52, retireAge:40}) → 53` / `fillDefaults({age:30, retireAge:40}) → 40`, `FIRE project: first draw row at age 40`, `rows 40–64 GIS 0 clawback 0`, `mixed flag never true on non-draw rows`, `no depletion in the partner-priority window`) |
+| `tests/i18n.test.js` | `I18N-OK all i18n checks passed (50)` = 48 base + 2 new (mixedNote EN/FR parity, q.retire.hint FIRE EN/FR) |
+| `tests/ui.smoke.js` | `SMOKE-OK all ui smoke checks passed (146)` (`wi-ret min 40` in index.html, legend element present, † fixtures + bilingual legend) |
+| `tests/ps-pension.test.js` / `ps-auto-service.test.js` | `RESULT: 70 pass, 0 fail` / `RESULT: 21 pass, 0 fail` (unchanged files) |
+| `passes3.js` rerun (live engine) | exit 0, `conservation: OK` ×3 → `audit/math-fix/passes3-pass44.txt` (md5 `01075d0e…` — **byte-identical to passes 14–43**, thirty-one consecutive deterministic passes). PASS 1 row 55 Total $68,327 / Withdraw $102,139 / **OASnet $0 / GIS $0**; GIS first at **65**; depletion 56 honest (max total $166,308 @54, final $0); PASS 2 depletion 82, max $468,146 @64; PASS 3 FIRE first draw @**40**, max $493,710 @39 = hand-math 350000×1.035¹⁰ ✓. `.tmp/mathfix25/scan3.py` (unmodified 32/25-column lineage scanner): PASS 1 rows=43 firstDraw=55 firstGIS=65, row55 wdr=102139 oasnet=0 gis=0, depletion 56, max $166,308 @54; PASS 2 depletion 82 max $468,146 @64; PASS 3 rows=65 firstDraw=40 depletion 48 max $493,710 @39; all GIS>0<65 / OASnet≠0<65 / neg-OAS / growTotalWithDraw lists empty in all three passes → `.tmp/mathfix44/scan44.log` **`diff`-identical to the pass-43 sweep** |
+| `scripts/deploy.sh` | `DEPLOY-OK` (fresh this pass) |
+| Origin byte-check | 3-way sweep `.tmp/mathfix44/sweep44.sh` (pass-43-lineage body; `sed`-derived: header comment + 2 list paths only, `diff`-proven only-43→44 substitutions; lists `.tmp/mathfix44/{tracked,ignored}_site.txt` regenerated this pass from `git ls-files` + `find`/`comm`, `diff`-proven **byte-identical** to the pass-43 lists: 21 tracked + 13 ignored): local `site/` == `origin/main` blob == webroot — **21/21 (3-way) + 13/13 (2-way) = 34/34 match, 0 mismatch** → `SWEEP-OK`; `git fetch` → `HEAD` == `origin/main` == `3854fb4`; `git merge-base --is-ancestor 0839826 HEAD` ✓ |
+| Live recapture (canonical leg **unmodified** `audit/mathfix_capture.py after` — md5 `49e22a83…` — + tagged `capture_tagged.py after44` — md5 `9489cef4…` == passes 30–43 byte-copy — on retire.krisforbes.ca) | Row JSONs **byte-identical to the lineage**: canonical `after-couple-rows.json` `129e2a45…` + `after-fire-rows.json` `4af12e59…` (fresh recapture this pass), tagged copies `after44-couple-rows.json`/`after44-fire-rows.json` md5-equal to canonical; K-aware 11-col rowsweep `.tmp/mathfix44/rowsweep41.py` (md5 `3d8823d2…` == pass-33 lineage, unmodified byte-copy) on the canonical set: **ROWSWEEP-OK** — couple 40 rows firstDraw **55** firstGIS **65**, FIRE 57 rows firstDraw **40** firstGIS **65**; **0** neg-OAS, **0** GIS>0 <65, **0** OAS≠0 <65, **0** growing-Total-withdraw pairs; `.tmp/mathfix44/rowsweep44.log` **byte-identical to the pass-43 log**. Couple rows 58–64 Total $0 / OAS $0 / GIS $0 (honest post-depletion drain vs the screenshot's impossible 126→141K @3.5%); `wiRetMin: "40"`, `noteShown: false` (52/52 same-age couple → engine-aligns-by-age, no mixed rows, documented deviation). FIRE: `fire_first_draw_age: "40"`. Mobile 390: slider `min=40 max=72 value=40`, `sliderRight 367 ≤ vw 390` (no clip), chip [25, 145], FIRE hint EN present, `svAfter40: "40 years"` |
+| Live mixed-window probes (all three **md5-copied byte-identical** into `.tmp/mathfix44/` from the attested lineage — self-older `61d66342…`, reachable `4e30a51e…`, 52/56 `377a4167…`; run with tag `after44`) | self-OLDER (me 52/pt 50, ret 51 → clamp 53): 0 † rows, `noteShown: false`, 0 violations; `after44-mixed-rows.json` **`96665bbf…` byte-identical to the lineage (passes 15–43)**. self-YOUNGER REACHABLE (me 50/pt 54, shared ret 51): **`MIXEDLIVE-OK`** — **`daggerRows: ["54"]`** (row 54 Total $195K ` †`, Withdraw $33K), legend **shown**: `† One partner is retired, the other is still working: … not in full retirement mode.`, `gisRowsUnder65/negOASrows/growingTotalWithDrawRows/growRowsMissingDagger` all `[]`; `after44-mixed-reachable-rows.json` `98c29f24…` == passes 26–43. **52/56 clamp-driven probe** (me 52/pt 56, shared ret 55 → partner clamps to 57): **`MIXEDLIVE-OK`** — row 56 Total $269K growing with Withdraw $30K carries **` †`**, **`daggerRows: ["56"]`**, legend shown, `gisRowsUnder65/negOASrows/growingTotalWithDrawRows/growRowsMissingDagger` all `[]`; `after44-mixed5256-rows.json` `53d0dec5…` == passes 33–43. Engine-side legs on the exact probe fixtures (`.tmp/mathfix44/legs42.js`, byte-copy of the pass-41/42 script md5 `33084975…`, `LEGS-OK`, output **byte-identical to the pass-43 log**, raw `diff` clean): **52/56** → mixed rows exactly `[56]` (Total $140,465 @55 → **$269,227** @56, wdr $30,000, GIS/clawback $0), `depletion 62` fires only on fully-draw rows (guard intact); **50/54** → mixed rows exactly `[54]` ($195,148, wdr $33,404), firstDraw 51, depletion 60, no GIS/clawback <65; **52/50** → no mixed rows, firstDraw 53 (clamp), depletion 56, GIS/clawback <65 empty |
+| Off-LAN (Midhurst `trouble@100.96.20.61`) | `OFFLAN-OK`, `"all_ok": true`, `"http_code": "200"`, `checked_at 2026-10-09T23:43:35Z` → `audit/bilingual-geo-check.json` |
+
+Determinism: `passes3-pass44.txt` (`01075d0e…`, **thirty-one consecutive** passes 14→44); `after-couple-rows.json` + `after44-couple-rows.json` md5 `129e2a45…` == passes 14–43; `after-fire-rows.json` + `after44-fire-rows.json` `4af12e59…` == passes 15–43; `after44-mixed-rows.json` `96665bbf…` == passes 15–43; `after44-mixed-reachable-rows.json` `98c29f24…` == passes 26–43; `after44-mixed5256-rows.json` `53d0dec5…` == passes 33–43. **Frames** (verified by md5 + PIL band diffs; this harness's model declares no image input, so images are not viewed, matching the passes 24–43 policy): canonical couple redraw `after-couple-table.png` == `after44-couple-table.png` **`cbac51ed…` md5-identical pair this pass** (chart-animation-phase variant, == passes 5/8/10–14/21–23/28/29/32–34/36/39/43-canonical; PIL band diff vs the tagged canonical frame `eec99021…` of passes 15–20/24–27/35/37/41–43 confined to **y=234–899** = the G1 animation band; table JSON byte-identical, so the evidence row set is unchanged); `after-fire-table.png` == `after44-fire-table.png` `685fd711…` **byte-identical pair** (canonical frame, PIL bbox `None` — pixel-identical); `after44-mixed-table.png` `222c81a1…` — **byte-identical to `after43-mixed-table.png`** (documented frame of passes 14–26/29–34/36–38/43; PIL bbox `None` — pixel-identical); `after44-mixed5256-table.png` `6d792f12…` — **byte-identical to `after41–42-mixed5256-table.png`** (band-limited variant; PIL bbox `None` vs pass 42, band y=234–899 vs the `36b82193…` frame of pass 43); `after44-mixed-reachable-table.png` `87594d5a…` (documented frame of passes 26–28/31/33/35/37/38/40–42; PIL band diff vs the `2e5aa424…` frame of pass 43 confined to **y=234–544** ⊂ the G1 animation band); `after-fire-step-390.png` **md5-identical pair** with `after44-fire-step-390.png` `761e8055…` 167,045 B 1170×2532@3× == passes 26–43, **pixel-identical** (PIL bbox `None`).
+
+Harness note: as in passes 39–43, the sandboxed shell routes `TMPDIR` to a read-only cache, so all Playwright invocations ran with `TMPDIR=/home/ubuntu/projects/retire-planner/.tmp/pw44` (pure temp dir; no script change). The three mixed probes, `capture_tagged.py`, `legs42.js` and the rowsweep were `cp`-copied byte-identically (md5-proven at destination) into `.tmp/mathfix44/` and run with argv `after44` (probes resolve `OUT` to `audit/math-fix` two levels up — identical results to the in-place runs of passes 37–43).
+
+**No code change required** — engine/renderer/i18n/tests byte-stable vs pass 43, deployed build verified identical 34/34. All plan acceptance criteria (1–6) re-verified against the deployed build. Final: **MATH_FIX_DONE**.
diff --git a/audit/bilingual-geo-check.json b/audit/bilingual-geo-check.json
index 8875795..81698c2 100644
--- a/audit/bilingual-geo-check.json
+++ b/audit/bilingual-geo-check.json
@@ -5,7 +5,7 @@
   "has_i18n_script": true,
   "i18n_js_fr_strings": true,
   "i18n_js_size": 39689,
-  "checked_at": "2026-10-09T23:38:48Z",
+  "checked_at": "2026-10-09T23:43:35Z",
   "vantage": "trouble@100.96.20.61 (Midhurst, off-LAN)",
   "all_ok": true
 }
\ No newline at end of file