docs(linkedin-studio): S31c hardening log entry + STATE session-end — S31 complete (a+b+c)

S31c descriptive-% log entry (full A/B/C/straggler breakdown + VERIFY). STATE: the Nx-multiplier
+ descriptive-% platform-norm class is now fully SSOT-reconciled; next phase = post-hardening
expert panel (operator-triggered), then command testing + GUI. Push parked (ahead 17; open/ is
public, STATE must not be published).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016qgzo6rxthw7KuxHjn5vyE

docs/hardening/log.md

@@ -2327,3 +2327,55 @@ apply the measured-data-class rule:** distinguish a platform-norm % (scrub) from
 % or a sourced benchmark (keep).
 
 ---
+
+### S31c multiplier-scrub (descriptive-% sub-pass, final) — platform-norm percentages asserted as fact → SSOT
+
+**CLASS: descriptive-%-scrub / hardening** (S31 batch c, final). Percentages asserting platform /
+algorithm / audience behavior as fact, unsourced — distinct from S30 discrete-% (already swept), from
+content-mix ratios (70/20/10), goal-progress benchmarks (50–80% of goal), and example-hook copy. Same
+discipline as S27/S28/S30/S31a/b: read-and-show (Read tool) → grep-confirm → re-grep final; 8-step gate;
+classification its own gate (NOT pre-approved with S31a/b — operator "Enig" before edit).
+
+**TEST (tool-grounded).** Per-claim exact greps over `commands` / `agents` / `references` / `skills`. The
+catalog (S31a/S31b SURFACED) was the seed; the plugin-wide sweep surfaced a LARGER new sibling set beyond
+it (the 64% / wrong-window-70% / Stage-2 / mobile / retention clusters) — surfaced + operator-approved
+into S31c before edit (scope is its own gate; not silently widened). Measured-data / sourced-benchmark
+rule applied: a platform-norm % is scrubbed; the user's measured/computed % or a cited benchmark stays.
+
+**HARDEN (24 edits, 12 files, +26/−26).** Vocabulary 1:1 with S30/S31a/b ("tends to / figure unverified",
+"proportion unverified", "multiplier unverified", directional kept + number dropped → SSOT pointer).
+- **Catalog (A):** suppression bands `analyze:230-231` (50-70% / 75%+ drop → partial / sharp reach loss),
+  `troubleshooting:180` (50-70% → partial); follow-probability `growth:58` + `glossary:216` (130% →
+  follow-graph signal, figure unverified); `report:215` ("per 2026 content format multipliers" → SSOT
+  pointer); `video-strategy-guide:304` (20-30% feed real estate → proportion unverified); `trajectory:39`
+  (30-40% follower growth → major driver, proportion unverified); "2.7x pattern interrupts" `glossary:153/198`
+  + `engagement-frameworks:84` (→ markedly more, multiplier unverified).
+- **New siblings (B), operator-approved in:** "64% more follow-up" `engagement-coach:52/81/199/313` +
+  `engagement-frameworks:350` (5× → tends to earn more, figure unverified); "70% reach" with the WRONG
+  window (60min/1h) `engagement-coach:75` / `post-feedback-monitor:71` / `glossary:125` / `linkedin-formats:164`
+  → reconciled to the SSOT window (first 15–30 min decide ~70%; the **~70% KEPT** — SSOT-sourced l.81, high);
+  Stage-2 distribution `post-feedback-monitor:72` + `glossary:130` (6-10% of connections → small slice,
+  proportion unverified).
+- **Borderlines (C), operator-ruled:** "70% of retention" (video) `video-strategy-guide:80` +
+  `content-creation/SKILL:235` → disproportionately determines (number dropped); "70% mobile"
+  `linkedin-formats:147` → the majority. **KEPT C1:** "~45% less engagement" (AI comments)
+  `engagement-coach:195/455` + `linkedin-formats:176` — already hedged "(correlational, medium
+  confidence)" + SSOT-tied (l.96); the hedge does the honesty work.
+- **Straggler folded in (surfaced, not silent):** "~3% of posts reach save-worthy status" `glossary:216`
+  (same sentence as the 130% edit) + `growth:59` → "a small fraction". Not in the original map; surfaced +
+  folded with operator veto open.
+
+**KEPT INTACT.** Sourced / computed benchmarks: `growth:177-178` (Buffer 178% / 247%, computed from the cited
+14,001 vs 5,033 / 381 vs 110), `growth:198` (11% engagement rate, Socialinsider 2024). SSOT `~70%` reach
+figure (kept; only the window corrected). `firsthour:4/22` already used the SSOT window → untouched. OUT
+(not the class): content-mix %, goal-progress benchmarks, scoring-rubric percentiles, word-count ratios,
+example-hook copy ("73% / 84% / 87% / 67% / 40% more").
+
+**VERIFY.** Per-claim re-grep of every scrubbed pattern over the FINAL files → 0 residual; SSOT `~70%` remains
+only in the 5 reconciled-window lines + SSOT + firsthour. Gate 81/0/0 exit 0, counts 29/19/26 + v0.5.0.
+`git diff --stat` = 12 files, +26/−26. Commit local (push held — `open/` public).
+
+**S31 COMPLETE (a + b + c).** The "Nx" multiplier + descriptive-% platform-norm class is fully reconciled to
+SSOT. Hardening queue → post-hardening expert panel (`docs/expert-review/plan.md`) · command testing · GUI.
+
+---