docs(linkedin-studio): S31a+b hardening log entries + STATE session-end — multiplier core done, S31c descriptive-% queued

Log entries for the two multiplier-scrub batches (S31a reference-docs 35ed480,
S31b commands/agents c850352) with full TEST/HARDEN/KEPT/VERIFY/SURFACED, and
STATE.md overwritten for session-end.

Captures the new measured-data-class rule (S31b): "Nx"/"N%" that reports or
hypothesizes from the user's OWN measured analytics — or a pure placeholder
[X]x — is not a platform claim and stays (the ab-test:418 "your data shows ...
test whether this holds" pattern); distinguish from generic-claim TARGET. S31c
must apply it to the descriptive-% sub-class (some bands are unsourced platform
norms to scrub, some are sourced benchmarks / user-measured % to keep).

Also records the jerndisiplin reminder that grep output does not count as a
Read for the Edit safety check (three S31b edits were correctly blocked until
re-read with the Read tool).

Push held: origin = open/ (public), STATE.md must not go there; ahead 14 parked.

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

docs/hardening/log.md

@@ -2259,3 +2259,71 @@ like", "quality comment ≈ 2x a like" — SSOT medium) and (b) frequency counts
 engagement rate") + `report:215` "content format multipliers" reference.
 
 ---
+
+### S31a multiplier-scrub (reference-docs) — unsourced "Nx" reach/format multipliers → SSOT
+
+**CLASS: multiplier-scrub / hardening** (surgical SSOT-reconciliation of unsourced "Nx" reach/format/
+growth multipliers asserted as fact; same discipline as S27/S28/S30: read-and-show → grep-confirm →
+re-grep final). Operator: "run everything, multiple sessions"; classification + batching (S31a refs →
+S31b commands/agents → S31c descriptive-%) operator-approved before edit.
+
+**TEST (tool-grounded).** Plugin-wide grep `[0-9]+(\.[0-9]+)?(-[0-9]+(\.[0-9]+)?)?x` (214 raw hits)
+classified into KEEP (SSOT-backed save ≈ 5x / comment ≈ 2x / ≈ 1x), OUT (frequency counts, pixel dims,
+pricing + size-ratio multipliers), TARGET (the Nx reach/format/growth class), and a DESCRIPTIVE-% sub-class
+→ S31c.
+
+**HARDEN (26 edits, 7 reference files).** Vocabulary 1:1 with S30 ("tend to / more reach (multiplier
+unverified)", "no reliable multiplier", format ranking kept + number dropped → SSOT).
+- `collaborations-guide` (10): :3 10x reach, :32 2-3x normal, six per-format "Reach multiplier: Nx" fields
+  (:165/:187/:210/:235/:252/:277), :476 10x growth + the label-only :292 (no number) folded in for
+  consistency → "Reach:".
+- `linkedin-growth-playbook` (10): :57/:132 3x faster growth, :175 1.6x reach, :205 1.4x engagement,
+  :234 1.64x poll, :246-251 format-multiplier list → SSOT ranking (numbers dropped), :386 3.6x respond,
+  :422 3x engagements, :493 5.6x follower growth, :636 40x search.
+- `poll-strategy-guide:7` (1.64x), `articles-strategy-guide:28-29` (2-3x/10x), `video-strategy-guide:308`
+  (2-3x distribution), `ab-testing-framework:49` (1.17x-1.6x), `trajectory:40` (2-3x saves) + `:85` (3x).
+
+**KEPT INTACT.** SSOT-backed engagement-order multipliers. OUT: frequency (5x5x5, Nx/week), pixel dims
+(1080x1350), pricing (2-5x market rate), size-ratio (2-3x follower count).
+
+**SURFACED → S31c (not edited here).** `growth:58` (130% follow), `growth:177/178` (Buffer 178%/247%),
+`trajectory:39` (30-40%). Flagged absolute-figure class outside S31: `growth:421/423`. Sourcing overclaim
+to revisit: `growth:492` "Official LinkedIn data".
+
+**VERIFY.** Re-grep residual targets in 7 files → 0 (the `:292` label caught by re-grep, folded per
+operator nod). Routed siblings + KEEP intact. `git diff --stat` = 7 files, +32/−32. Gate 81/0/0 exit 0,
+counts 29/19/26 + v0.5.0. Commit `02a8ffd` (local; push held — `open/` public).
+
+### S31b multiplier-scrub (commands + agents) — generic "Nx" claims → SSOT
+
+**CLASS: multiplier-scrub / hardening** (S31 batch b). Generic-claim Nx multipliers in commands/agents.
+
+**TEST.** Grep the 6 S31b files for the multiplier class → no new siblings beyond catalog. Close read
+revealed a real distinction (see KEPT): some "Nx" lines report/hypothesize from the USER's own measured
+analytics, not a platform norm.
+
+**HARDEN (5 edits, 5 files).** `report:249` (3.4x reach multiplier) → top organic format; `outreach:652`
+(2-3x reach, ref collab) → above-baseline, aligned w/ S31a; `engagement-coach:120` (3.6x respond) → far
+more likely (figure unverified); `strategy-advisor:241` (3x faster growth) → faster (figure unverified);
+`content-repurposer:591` ("Expected reach multiplier: [2-5x original]") → "Expected reach: above the
+original (no reliable multiplier)".
+
+**KEPT — measured-data class (operator-ruled, NEW RULE for S31c).** `report:303` ("[format] … 2x more
+shares" example), `report:335` ("your one carousel got 3x" rendered example), `ab-test:418` ("Your data
+shows … 2x … test whether this holds"), `outreach:1234` ("Average reach multiplier (collab vs solo):
+[X]x" output field). These report/hypothesize from the user's OWN analytics — the number is an
+illustrative-computed example or a pure placeholder; removing it adds no honesty. Also KEPT:
+`engagement-coach:148` SSOT comment ≈ 2x. OUT: `strategy-advisor:239` 4-5x/week frequency.
+
+**VERIFY.** Re-grep generic-claim targets → 0; measured-data class + KEEP/OUT intact; gate 81/0/0 exit 0,
+counts 29/19/26 + v0.5.0. Commit `e28e357` (local; push held).
+
+**SURFACED → S31c (descriptive-% sub-pass, queued).** Suppression bands (`analyze:230` / `troubleshooting:180`
+"50-70% / 75%+ drop"), engagement-rate benchmark (`growth:198` "11%", Socialinsider-sourced), `report:215`
+"per 2026 content format multipliers" reference, `video-strategy-guide:304` "20-30% more feed real estate",
+`trajectory:39` "30-40% of new follower growth", `glossary:216` "130% follow probability", `growth:58/177/178`
+(routed from S31a), "2.7x pattern interrupts" (`engagement-frameworks:84` / `glossary:153/198`). **S31c MUST
+apply the measured-data-class rule:** distinguish a platform-norm % (scrub) from the user's measured/computed
+% or a sourced benchmark (keep).
+
+---