What to do, what to click, where to go

RULED BY AGA 4 Aug — she chose Option C: '$97 for your first year, then $157/year from 4 August 2027. Cancel any time.' It fixes the contradiction by naming $97 as a FIRST-YEAR price (not a locked-forever one) and naming the renewal date, which the live line never did. 🔴 ONE IMPLEMENTATION NOTE FOR NIC: the date must be COMPUTED as purchase-date + 12 months, not hardcoded — '4 August 2027' is only correct for someone buying on 4 Aug 2026. Render it from the same clock the charge uses. Copy lives inside the quiz bundle (index-*.js), not a WordPress page, so it cannot be edited from the SEO tooling — needs Nic. Goes to his brief once Aga approves it there.

AGA RULED 13 Aug 2026: **3 MINUTES**. Canon confirmed — it matches her 3 Aug ruling (AGA-75) and GA4's measured 3.2 min average, so the ruled number and the measured number agree. CONSEQUENCES: (1) 3 minutes is the ONLY duration any TMA surface may state for the web quiz. (2) The ~10-minute figure belongs to the IN-APP Layer 2 calibration and must never be attached to the web quiz again — that borrowed-claim swap is what put 'about ten minutes' into the AGA-146 activation email. (3) 🔴 THE LIVE PAGE STILL DISAGREES and Aga cannot fix it herself — quiz.themovementathlete.com 302s to /2026-quiz-checkout-v1/, whose bundle states 'An 8-minute assessment that maps your…' and 'This 8-minute assessment maps your 9 movement…' as the PRODUCT claim, plus 'This free 5-minute assessment' elsewhere; only a narrative line says 3. Filed for Nic as the page fix. UNTIL THE PAGE SAYS 3, new copy states NO duration rather than a number the destination contradicts — an email promising 3 minutes that lands on a page headlining 8 mis-sells twice.

DONE 13 Aug 2026 — but the enable was only half the job. Access WAS switched on (by Aga or a parallel session) and it locked Aga out of her own docs: 'That account does not have access'. VERIFIED LIVE via the Cloudflare API — root cause was that the ONLY identity provider was 'cloudflare' (sign in with a Cloudflare ACCOUNT), so the policy's email agnieszkanazaruk@gmail.com could never be asserted unless her CF account used exactly that address. FIX APPLIED: created the onetimepin identity provider (id 2c7763ce-4396-47e9-a14e-d6221aea5bf9) — the login page now offers Email + a 6-digit code, verified live in the rendered page. SECOND FAULT FOUND AND FIXED: the allow policy was NAMED 'Aga + Nic' and contained ONLY Aga; Nic was never in it. Added nicmrayce@gmail.com (confirmed as Nic from the GTM change log + his GitHub login, NOT from the 21-hit torterotot.nicolas@gmail.com which is a CUSTOMER). App a52306b1 on docs.themovementathlete.com, policy dff48f3f, service-token policy left untouched for the fleet.

Probe row only — created 12 Aug to verify the WIP cap exempts owner_code FLEET. It does (cap refused a human row at 43 open; this one minted). No work implied.

DONE, verified live 12 Aug 2026: AC campaigns 4146-4153 (all 8 slots of 750 TMA | Welcome Sequence After Quiz) confirmed type=split, status=1, three arms each = 24 live variants. Per-arm data IS readable via GET /api/3/campaigns/{id}/campaignMessages — each arm row carries its own send_amt, uniqueopens, uniquelinkclicks and unsubscribes (e.g. c4146 arms 13682/13683/13684 at 21 sends each). The self-learn loop therefore has live, readable experiments, which was this row's whole purpose. NOT closed by this row and carried forward on AGA-141: 991 NEW Re-Engagement (3 send steps, 0 splits) and 973 Nurture Sequence 1 NEW COPY FROM CLAUDE (23 send steps, 0 splits) have no splits — but the 750 subject test named in this row's title is live.

2026-08-12 VERIFIED STALE: 'nothing ever emails them' has been false since ~7 Aug on three counts — Nic's Django T+1h abandon email is LIVE (SendGrid checker: 24 sends 9-12 Aug), the hello@ personalised lane sent 39 on 7 Aug + drafts daily at 06:05, and the 30-min AC bridge stamps persona/plan/tag on every abandoner. The job audit ('all 20 jobs, nothing emails them') predates com.tma.quizabandon (3 Aug) and com.tma.abandondraft (10 Aug). ~14/day superseded by the reconciled denominators (§3a: pool ~315/mo). NO surviving work of its own — folded into FLEET-42 (756/757 personalised rebuild) + E11 (Nic upload/activation) + FLEET-46 move 4 (v3 hourly reconcile — decided/D5-approved, still never executed, the one genuinely undone piece). Historical >7-day pool = AGA-118's send.

CLOSED 12 Aug 2026 — nothing left in it. It was SPLIT on 31 Jul into AGA-34/35/36; all three are now finished (AGA-34 superseded, AGA-35 superseded, AGA-36 done). The parent stayed rendered on her page because the generator deliberately keeps 'split' rows visible with their status — correct as a rule, wrong once every child dies. Verified each child's status live before closing.

DONE 12 Aug 2026 — Aga granted the service account Editor on property 177471782, and the fleet created the dimension itself the same minute. VERIFIED BY READ-BACK, not by claim: POST returned properties/177471782/customDimensions/15427123787 {parameterName: method, displayName: 'Login method', scope: EVENT}, and a LIST call now returns 5 dimensions with 'method' among them (was 4). The Data API also already ACCEPTS customEvent:method — no more 400. 🔴 NOT YET ANSWERED: every row currently returns blank or '(not set)' because the dimension is not retroactive and all events so far predate registration. Read it again on 14 Aug. 7d baseline for that read: login iOS 132 / Android 72 · social_login_attempted iOS 30 / Android 0 · signed_in Android 5 · sign_up web 15. Standing gain: the fleet can now create and manage GA4 config itself — no GA4 config task should ever be filed on a human again.

2026-08-12: Aga confirmed in session she has SENT the follow-up to Sovat (restore into STAGING, never over production). Her action on this row is complete. Standing next step is the fleet's, triggered by his reply: if he produces a staging restore, diff it against production BEFORE anything else — no restore ever touches production (memory: wpengine_staging_envs_are_stale_never_push_2026-08-12).

DUPLICATE of AGA-165, folded into it 12 Aug 2026. A parallel session reached the same conclusion 40 minutes earlier from Aga's forwarded dev correspondence; this row reached it independently by measuring the staging environments directly. The ASK is identical, so this row is cancelled and its evidence merged into AGA-165 and into the draft reply — Aga sees one row, not two. Evidence carried over: (1) the fingerprint proof (CA production and CA staging both return page id 205530 slug 'tma' modified 2020-06-11T14:22:36 — identical, which proves the staging-push cause from the artefact rather than inferring it from correspondence); (2) BWTA staging measured directly — newest content 30 Jul, and across 14 recently-edited pages production carries 31 quiz-root links against staging's ZERO, 14 of 14 pages at nil; (3) TMA staging newest content 2 Aug, also behind; (4) CA's active theme is STILL Divi 4.4.8 despite the dev reporting the theme updated; (5) deactivating Page Builder Everywhere removed the site's copyright line from every page (Divi default footer now renders instead); (6) WP File Manager 8.0.4, WP phpMyAdmin 5.2.2.01 and ClickFunnels 3.1.1 are all still ACTIVE on CA; (7) live administrator lists pulled for CA and BWA (4 each), which answers the dev's open question directly instead of asking him again.

✅ SENT by Aga 12 Aug 2026. The message went with the measured evidence in it: the page-205530/2020-06-11 fingerprint proving the staging→production DB copy, the 14-page BWTA prod-vs-staging diff (31 quiz-root links vs ZERO), the four hard asks (no staging DB push · tell me first if a DB change is needed · full backup + restore refs before touching BWTA/TMA · confirm what was pushed to CA and when), the three CA corrections (Divi still 4.4.8 · PBE deactivation removed the copyright line · WP File Manager + phpMyAdmin + ClickFunnels still active), the live admin lists answering his 30,659-users non-answer, and the mysql.sql SFTP confirmation Rocket.net's file-type limitation makes hard-blocking. AWAITING his written confirmation on point 1. 🔒 The wait is now GUARDED, not just remembered — see FLEET-67: a daily checker compares each production site against a stored baseline and alarms if production content goes BACKWARDS, which is the signature of a staging push. Sent copy: marketing/seo/DEV_REPLY_STAGING_PUSH_2026-08-12.md

Withdrawn on Aga's correction and re-verified against AC: 751 = 12 slots, 206,401 sends, last send 2026-08-12, healthy on every slot. 973 = retired per Aga. The finding transfers to 751, which shares 973's copy: 9 of 12 subjects fail Gate 2 (8 over the 50-char cap, 1 age reference) plus the same misattributed members. Work moved onto 751.

DONE 12 Aug 2026 — Aga ruled OPTION A ('yes follow the recommendation'): repair the door rather than preserve the comparison cohort. All 21 live messages of 988 TMA | Welcome Sequence AFTER APPSIGNUP repaired via tools/email-sequences/fix_988_door.py --apply, backups in marketing/email-sequences/_AC_988_DOOR_BACKUPS_2026-08-12/, read-back OK on every message, converges (21 already clean on re-run). VERIFIED INDEPENDENTLY after the write: start-training 0 (was 2 per message, 42 total) · quiz-root links 42 (was 0) · plaintext drift 0 · grey canvas 0 · bare %FIRSTNAME% 0. This arc uses a DIFFERENT template from 992 — no tma-cta anchor; every target was re-derived from its own live bodies (p.cta-head above a.cta-btn, plus a second checkout link in the P.S. with per-slot anchor text). Beyond the links: 8 P.S./body sentences that described CARD mechanics were rewritten (they were honest under a trial button and false under a free-quiz button — e.g. 'The card goes in to start the week'), four caption phrasings were swapped (the fourth, E10's 'Charged on day 8 only', matched none of the first three — the exact way B2 slipped through on 992), and the arc's own pre-existing hygiene defects were fixed in the same pass: grey canvas on every message and bare %FIRSTNAME%, plus two over-length paragraphs in E4's arms. Lint ALL PASS on all 21 candidates before writing.

RULED 12 Aug 2026 — Aga: 'leave this alone.' NO REFUNDS. The 15 double-paid members ($1,662.86 across overlapping periods since 2024) are left as they are; the row is closed as a decision made, not as a problem fixed. Her standing rule stands and is reinforced: the fleet surfaces a money finding and STOPS — no refund, no credit, no member contact, ever, without her word. 🔴 IF A MEMBER RAISES IT: that is a fresh decision for Aga, not a precedent set by this closure — do not cite this row as 'we decided not to refund' to anyone outside. WHAT CONTINUES WITHOUT HER (unchanged by this ruling): diagnosing from Stripe HOW the overlap happens — whether one specific checkout path leaves the prior subscription running — so the next person is not double-billed. Prevention was never part of the refund question.

2026-08-12 VERIFIED LIVE (all 12 wired send-step campaigns of 751 Nurture Sequence 1, HTML AND plaintext read separately): Aga was right — the fix is in. Every email the automation actually sends now reads 'you signed up at The Movement Athlete'; zero occurrences of 'took the assessment' in any wired message, either surface (the known 751 plaintext-lag failure mode is absent). Residue, not member-facing: 10 ORPHANED unwired campaigns under seriesid 751 still carry the old line (plus one with HTML/plaintext divergence) — unreachable by contacts (no send block references them); cosmetic cleanup only, folded into routine estate hygiene.

✅ ANSWERED 12 Aug 2026 — no WP Engine support ticket needed. Aga forwarded her dev correspondence and it explains it: Sovat built staging environments (calisthenicstg / movementathstg / themovemen3stg), did the plugin+theme+PHP 8.2 work there, and on 10 Aug 08:17 wrote 'now that you've confirmed everything is good, I'll move forward with the production updates.' The CA staging clone was OLDER than production, so pushing it over calisthenicsacademy.co reverted that site's database to its 2020 state — newest surviving edit site-wide was 2020-06-11, uploads intact (database-only), and the application password went with it because an app password is a DB row. That matches every symptom exactly and needs no restore-point theory. 🔴 SUPERSEDED BY AGA-165, which is the ACTIONABLE half: BWTA and TMA have NOT been pushed yet and are verified intact (BWTA banner 416/459, re-checked page by page against the previous day's baseline, zero losses; TMA's 11 Aug edits present), and Aga's own question to him — 'when you push them to production... how can we avoid this?', asked 11 Aug 10:54 — was NOT answered in his 10:36 PM report. A BWTA push on the CA pattern would silently destroy 416 banner pages, the 4.8 sweep, 686 paragraph splits, 18 301s, the FAQ schema fix and the title change under measurement. Draft reply written for her: marketing/seo/DEV_REPLY_STAGING_PUSH_2026-08-12.md

DONE 12 Aug 2026 — Aga exported the CSV and it is verified as the right file. LANDED: marketing/campaigns/paid-ads/fb-lead-exports/2026-08-12_build-muscle-with-40plus.csv (263 rows). PROVENANCE CONFIRMED, not assumed: every row carries form_id 886259157389251, form_name 'Build Muscle With 40+-higher intent' and campaign_name 'MOF - Lead Form'; is_organic=false on all 263; date range 2026-05-26 to 2026-08-11, which is exactly the inside of Meta's rolling 90-day window; 263 usable unique emails, ZERO blank or malformed; first_name present on all 263; platform fb 212 / ig 50. (The zip's second file lead.csv was a one-row CRM sample with no email — kept as 2026-08-12_lead-crm-sample.csv, no value.) 🔴 MY EARLIER CLAIM WAS PARTIALLY WRONG AND IS CORRECTED HERE: I said these leads were 'definitively NOT in AC'. Now that the emails exist I checked all 263 against live ActiveCampaign one by one: 237 (90.1%) are genuinely absent, but 26 (9.9%) ARE present. The inference was right in direction and overstated in degree. HOW THE 26 GOT IN — they were never delivered by the dead integration; they found their own way: 20 of 26 carry QuizUserCreated/QuizPasswordProvided (they did the quiz independently), 4 carry free_signed_up_mobile_&_web, and 3 carry the facebook-lead-ads tags but with contact dates of 18/25 Nov 2025 — repeat form-fillers whose July 2026 submission did NOT re-tag them, which is independent confirmation the sync is dead. 🔑 THE FINDING THAT MATTERS MOST: 6 of the 26 carry stripe-integration and 4 carry P1-ANNUAL-JUL2026 — people who filled this lead form BECAME PAYING CUSTOMERS after reaching us by another route. This audience converts, so the 237 who never received a single email are a genuinely warm list, not junk. SEGMENT-INTEGRITY PRE-CHECK PASSED: all 237 checked against ever_paid_emails.json (16,080 emails, rebuilt 12 Aug 02:10) — ZERO payers among them, so the mailable candidate count is 237. NEXT, AND IT IS FLEET WORK NOT AGA'S: the import, tagging and first-email design move to FLEET-70. The reconnect half of this row stays retired — the campaign is paused, so there is nothing to catch.

BOTH CALLS EXECUTED ON AGA'S GO, 12 Aug 2026, in one widget save. CALL 1 — text-69 (the 2016 'INSANE SKILLS?' advert, an image link to the bare TMA homepage with NO UTM) REPLACED with the assessment card; widget title cleared because the card carries its own heading. CALL 2 — text-47 (the ebook) KEPT, per the recommendation, with its link made countable. 🔴 THE FIX WAS NOT THE ONE PLANNED, AND THE PLANNED ONE WOULD HAVE BEEN A SILENT NO-OP: adding UTMs to /calisthenicebook does nothing, because that 301 STRIPS THE QUERY STRING — measured, ?utm_source=bwta... came back with a bare redirect_url. So the link now points DIRECT at https://calisthenicsacademy.co/calisthenics-fundamentals-book/?utm_source=bwta&utm_medium=sidebar&utm_campaign=sitewide&utm_content=sidebar-ebook, verified to preserve the query string end to end. LIVE-VERIFIED in a real browser on /push-up-progressions/ at BOTH 1440px and 390px: card renders (208x581 desktop, 390x374 mobile), CTA -> quiz ROOT with UTMs, the 2016 image is gone estate-wide, ebook href carries its UTMs, no horizontal overflow on either. Tap target was 43px on mobile — under Google's band — so the card's padding went 13px -> 15px and it now measures 47px. Backup: tools/seo/apply_backups/bwa-widgets-sidebar1-20260812T074524-PRE-AGA149.json (full raw, every pow_* key preserved). ALSO RETRACTED IN THE CARD'S OWN SOURCE COMMENT: the claim that ClickFunnels' 'unpause your account' advert renders on the order page — measured false, zero occurrences. 🔴 WHAT THIS DOES NOT YET DO: the ebook CLICK is now tagged, but calisthenicsacademy.co carries GTM-TKGDZR9 and a dead UA-70796991-1 with no GA4 measurement ID found, and the ThriveCart cart carries fbq but not G-ZBQN3P8WKW — so the SALE is still invisible. That half is AGA-143 and is not closed by this row.

MEASURED 12 Aug 2026 — the premise is refuted on all three counts, and the fix it proposed is unnecessary. (1) GA4: a WIRE TEST (headless Chromium reading /g/collect, not a source grep) shows BOTH tma.thrivecart.com carts AND the CA ebook page load GTM-TKGDZR9 and fire GA4 on our own G-ZBQN3P8WKW. (2) The CA site is in the property with 239 sessions/28d carrying page_view, scroll and click. (3) MONEY: ThriveCart revenue lands in Stripe account 2 (STRIPE_API_KEY_2, 'Thrivecart lifetime-sales recovery account'), which TEN existing tools already read — stripe_mrr_pull.py, rebuild_daily_revenue.py, refresh_cockpit_daily.py, ever_paid_build.py, segment_guard.py, funnel_stage_export.py, customer360.py, flow_metrics_pull.py, reply_fulfil.py. Account 2 holds ,374.58 across 79 charges lifetime, ,208.94 in the last 90d. 🔴 HOW I GOT IT WRONG: I grepped page source for a GTM-INJECTED tag, which CLAUDE.md explicitly forbids ('NEVER infer what is tracked from page source'). The tag is invisible to a grep and obvious on the wire. TWO REAL THINGS SURVIVE, both small and neither needing an Aga decision: (a) tma.thrivecart.com shows 0 sessions in GA4 over 28d DESPITE the tag firing — a hostname/data-stream config question, filed as FLEET work not an Aga call; (b) the ebook barely sells — 7 charges <=$35 across account 2's entire 79-charge history and ZERO in the last 90 days, which is a real number against the 'keep it, it is live revenue' case and should inform the 4-week review of the sidebar widget (AGA-149 call 2, now live with UTMs). CONSEQUENCE FOR THE ~25 AUG CAPTURE READ: I told Aga it would 'undercount while looking clean'. WITHDRAWN — the rails are counted; the read stands as designed.

2026-08-12: Verified LIVE against https://tma-customer-journey-2026.netlify.app (Quiz journey tab) on Aga's ask ('I believe this has been done'): all six decisions are ruled and documented — J1 SIMPLIFIED BY AGA 7 Aug eve (shape decided; one input parked as a Nic question) · J2 yes with the nightly-refresh caveat · J3 evergreen enters at journey exit, AGA-113 repoint NOT executed (reconfirmed by Aga 12 Aug: 760 School of Strength (NEW) is SoS-lead-signups ONLY) · J4 $97 in-window / $78.50 evergreen · J5 surgical per Evolution Law · J6 3-email activation micro-arc. Outstanding items are NOT this row's — they live on AGA-134 (welcome-rebuild GO, still awaiting Aga), AGA-135, AGA-161.

2026-08-12 EXECUTED BY FLEET on Aga's GO ('yes fix this'): the 7 in-flight contactAutomation records (5 unique contacts: 354249 x2, 136502 x2, 371561, 372074, 372045) were removed from 973 Nurture Sequence 1 NEW COPY FROM CLAUDE via API — all 7 DELETEs returned HTTP 200, readback confirms 0 in-flight (35 completed records remain). Nobody can now receive the three false-deadline emails (steps 18/19/22). Entry stream was already dead since 3 Aug (AGA-98 repoint). No builder visit needed; the automation itself left ACTIVE-but-empty per Aga's 'deprecated, will drain' framing.

2026-08-12: Aga asked what was already decided on abandonment; folded per the decided record (7-10 Aug: journey-page Abandonment tab + her four-email ruling). Do NOT arm the pre-personalisation email: the decided path is FLEET-42 (rebuild 756/757 E1-E4 on quiz-signal personalisation) -> E11 (Nic uploads into fresh shells) -> ONE arming decision with rendered emails in front of Aga. Interim coverage already exists (Nic's ~1h login email + the Tier-A personalised-notes lane), so the audience is not unserved while the rebuild lands. 757 stays off either way.

2026-08-12: Aga RULED option (c) — the two annual prices are DELIBERATE and documented: $97 is the QUIZ-WINDOW price (paywall + the re-pitch while their quiz window lives); $78.50 is the EMAIL second-chance price, reached only through the 50%-off sequence (993 SALES SEQUENCE / the evergreen event). Her words: 'the quiz has 97 its correct, then we pitch it to them again, and then through email they enter the 50% off sequence and then we pitch 78, that's it.' Consistent with J4 on the journey page. Ruling appended to OFFER_LADDER.md; estate propagation sweep minted as its own FLEET row.

2026-08-12: Aga ruled in session — '760 School of Strength (NEW)' is ONLY for School of Strength lead signups; the repoint premise is NOT correct. TMA already has its own sales sequence implemented and active in AC. Do not route 750/751/758 traffic into 760; it keeps its own SoS-signup audience. This is the ruling the 7 Aug hold note was waiting for (J3's 760 sub-question): 760 is NOT the evergreen sales entry.

CLOSED 12 Aug 2026 — re-measured live, then emptied of Aga-work. Four of the five checks were answered by measurement with no ad-UI visit (see the row's amended title and note for each: the Audience Network lock would have destroyed a live Advantage+ A/B arm and AN is 0.07%% of Meta spend; the +43%% budget step never reached spend, £34.84/day Jul vs £34.98/day Aug; no VIDEO campaign exists and T14's 69,076 'advertising' views traced to the now-paused 4th-of-July Demand Gen campaign plus 10 paused UAC campaigns; and the ASA premise was false — our own brand is the only thing Apple Search Ads runs, already settled by DEC-2026-08-11-keep-the-3-branded-apple). The fifth — item (a) — was real but the row mis-stated the ruling it cited: google-category-search-only is a CHANNEL rule, so 2 of its 3 named campaigns do not breach it, and PMax is the only live breach. Its two live consequences were split out and moved off Aga: NIC-108 (PMax placements — £60.85/30d, 280 clicks, zero conversions) and NIC-110 (Competitors Branded budget). 🔴 AGA REASSIGNED THE REMAINING DECISION ON 12 AUG: 'this needs to go on the to-do list for Nick' — so the keep-or-kill call on Quiz Funnel | Competitors Branded | US is now NIC-110, approved, with a directive recommendation (cut £10→£5/day after NIC-108, hold until after AGA-157's 13 Aug read, kill condition pre-registered for the 10 Sep gate). Nothing on this row remains for Aga. Full working: quiz-funnel-versions §PA.11.

SHIPPED + LIVE-VERIFIED 11 Aug 2026. Redirection rule created (/30days/ -> https://themovementathlete.com/start-training/, 301, enabled) AND the source page unpublished (BWA page 210113 -> draft, backed up to tools/seo/apply_backups/). Verified on the LIVE url, not from the POST: 301 -> https://themovementathlete.com/start-training/. Front-page guard confirmed 210113 is not BWA's front page (206880 is). The NIC-98 warning on the target was re-measured live first: the DEAD $297 is GONE (zero occurrences of '297' on the page) so the blocker that deferred this row no longer exists; the retired 4.9 rating survives but is ONE string, not the 12 the row claimed (that count matched inside $24.97 and SVG path data) and was already exposed to all of that page's existing traffic. 🔴 TRAP FOUND AND FIXED IN CODE: cache-busting a redirect with a query string DEFEATS the Redirection match — /30days/?cb=99 returns 404 while /30days/ returns 301. retire_dead_pages.py now busts cache with Cache-Control/Pragma headers and never a query string.

RULED BY AGA 4 Aug after seeing the live comparison (artifact 4ecb2164): 'of course use the built, gate-passed one.' = REPLACE. The new end-banner block REPLACES the existing end-of-article CTA on BWA's 162 pages and the 8 TMA pre-footer Spectra instances — it does not stack. Evidence that made it obvious: a live 390px capture of bodyweighttrainingarena.com/push-up-progressions/ shows the article currently ending in THREE stacked CTAs, one of them a Calisthenics Academy image and one an ebook advert. 🔴 Before applying: the banner renders 4.7 stars and must be updated to 4.8 per her same-day ruling, and the CTA button to gold #FFB000 per AGA-91. ||| 🔴 CORRECTION APPENDED 11 Aug 2026 — AGA'S RULING STANDS UNCHANGED; TWO FACTS IN MY ORIGINAL EVIDENCE LINE DID NOT. This resolution originally read '(the brand retired in 2018)' and 'an ebook we no longer sell'. BOTH ARE FALSE and are struck. Verified live 11 Aug: calisthenicsacademy.co is up and serving, its own nav feeds into TMA ('TAKE ASSESSMENT!' → themovementathlete.com/welcome/, 'Login' → app.themovementathlete.com/register), and the ebook 'The Ultimate Guide to Getting Started with Calisthenics' really sells at $27 through a working ClickFunnels order form (posts to /order, Stripe + PayPal referenced). Aga caught this herself. WHY IT MATTERED: this line was the ORIGIN of the error and it propagated into FLEET-15's title, the SEO weekly report and the revenue-plan page, each time telling a reader that a live revenue path was dead — which is how it would eventually have been deleted as cleanup. The 4 Aug REPLACE ruling concerned the END-OF-ARTICLE CTA and is untouched; the ebook advert is a SIDEBAR widget (text-47) and its fate is a separate open call, AGA-149.

ANSWERED by Aga the same day she was asked, 11 Aug 2026: 'yes nic needs to know about this — one Meta campaign is still sending people to the old quiz page instead of the new one that sells the $97.' RULING = option (1): the 10 Aug RE-AIM instruction stands and supersedes the 7 Aug 'deliberate control arm' note. Consequences executed: NIC-106 unblocked and approved for Nic; the 7 Aug keep-the-control-arm position must now be banner-stamped as overtaken wherever it appears (cockpit quiz-funnel-versions sections 3 and 6, and the Nic attribution brief's 7 Aug Variant-2 card) so it is not raised a third time.

CANCELLED by Aga's ruling, 11 Aug 2026: 'do not do it yourself' / 'this is Nic's job'. This row had moved the MC Control repoint onto Aga's side with the FLEET executing by API, on the grounds that the Meta token carries ads_management. Aga overruled that: capability is not authority, and changes inside the Meta ad account are Nic's job regardless of whether the fleet could make the API call. The work returns to NIC-106, which is restored to open with the corrected recommendation (REPOINT, not pause) and the click-by-click detail. No API write was made — nothing was changed in the ad account by the fleet.

ALREADY DONE — Aga migrated them. Verified live 11 Aug 2026 by fetching both pages and enumerating every app.themovementathlete.com anchor: /start-training/ = trial/1month + trial/3month + trial/12month + 50off/lifetime497; /50promo/ = 50off/3month + 50off/12month + 50off/lifetime497. ZERO /register/ links remain on either page. My row was written from an audit taken earlier the same day, before her change; I reported it as current without re-checking the live page. Also confirmed en route that the v3 URLs DO preselect their plan (browser: /trial/1month/ renders '1-Month plan · $24.97/mo', /trial/12month/ renders '12-Month plan · $157/yr'), so the 'it might become a choose-your-plan page' concern in the row was unfounded.

DONE 11 Aug 2026, verified end-to-end against Apple's own API. Aga uploaded the public key at Account Settings -> API -> Client Credentials -> Generate API Client, signed in as the API user (rayariley87@gmail.com / 'Raya R.'); Apple issued clientId+teamId+keyId, now in .env.local as APPLE_ADS_CLIENT_ID / _TEAM_ID / _KEY_ID. PROVEN LIVE, not claimed: ES256 client-secret JWT -> appleid.apple.com/auth/oauth2/token (200) -> /api/v5/acls returns orgId 3090590 'The Movement Athlete LTD', role API Campaign Manager -> /api/v5/reports/campaigns returns real rows. ASA cost is no longer NOT KNOWABLE: 90d spend $407.32, 163 tap-installs, $2.50/tap-install, 8 of 10 campaigns spent, only 3 RUNNING (all Branded, $1.11-1.27/install). New tool tools/finance/asa_pull.py (--days/--write/--selfcheck) writes data_asa_spend.json, registered in estate_selfcheck as 'asa-spend' (network tier, PASSING). NOTE the row's 5-minute estimate was wrong: the real flow is invite an API-role user -> accept from that Apple ID's PRIMARY email -> then the key box appears.

✅ DONE 11 Aug 2026 — Aga created the 'Claude-access' application password and it authenticated as administrator (id=1) on the first try. Ran tools/seo/ca_repoint_ebook_checkout.py --apply: all 44 dead-rail links across 5 pages repointed to https://tma.thrivecart.com/calisthenics-ultimate-guide/ — /clickbank/ 13, 688 8, 680 8, 636 7, 331 8. Each page backed up first (tools/seo/apply_backups/ca-pages-*-PRE-TCREPOINT.html), URL-only replace with an anchor-count assertion, read back, WP Engine cache cleared, then re-verified on the RENDERED pages: 44 dead-rail links -> 0, 0 ThriveCart -> 44, 5 of 5 pages clean. ThriveCart cart itself confirmed loading (HTTP 200). 🔴 THE CAUSE IS NOW ESTABLISHED AND IT IS BIGGER THAN THIS ROW — see AGA-152: the site was restored from a DATABASE-ONLY backup, and it did not just lose 10 Aug. The newest surviving edit on the entire site is 2020-06-11. Uploads/media are intact (all images 200), so files were untouched — database only. Note the 3 non-link changes from 10 Aug are still reverted and are NOT covered by this row: the ClickFunnels plugin is ACTIVE again, the false 'ClickBank is the retailer' disclaimer is back on /clickbank/, and 3 instances of the dead hello@calisthenicsacademy.co address are back.

REFUTED BY LIVE WRITE, 11 Aug 13:04 MDT. TMA_WP_APP_PW as stored in .env.local (mtime 12:35 today — updated AFTER this row was minted at 08:16) authenticates fully. Proved end-to-end, not by capability label: created draft page 256251, updated its title, force-deleted it — all succeeded. users/me?context=edit returns edit_pages=True publish_pages=True manage_options=True. The three lanes this row named as blocked all respond authed: wp/v2/settings OK, wp/v2/plugins OK (45 plugins), code-snippets/v1/snippets OK (11 snippets). No action needed from Aga.

OVERTAKEN BY EVENTS, verified live 10 Aug PM: 993 SALES SEQUENCE is ACTIVE (API automations/993 status=1, 121 entered, 115 in flight) — Aga un-paused it herself after this row was minted, releasing the held cohort as the first cohort. EVERGREEN SALES 01 · PL1 send_amt=0 at time of check, so the newest upload (16/16 read-back, repetition fixes included) is what everyone will receive. The optional W7 ruling (welcome arc soft-trial vs activation CTA) remains open at Aga's discretion — recorded in REVIEW.html Updates panel and the funnel-cro verdict; no action needed unless she says the word.

DONE — Aga ruled 12 Aug: 'leave them'. The 24 split arms stay as they are. Both of the row's original blockers were already refuted (per-arm stats ARE readable via campaigns/<id>/campaignMessages; the slot-6 offer collision was removed by the 10 Aug no-offer work and a sweep of all 24 arms' rendered prose found ZERO offer references). The remaining concern was power, not correctness: at 2-21 sends per arm no test can reach significance, so the splits are decorative but harmless, and per-arm data now accumulates against the day volume arrives. No builder time spent.

DONE 10 Aug 2026 — Aga ruled: there IS a deadline; do not say it comes back; do not say 'I'm not going to invent a deadline'. Both denials removed and VERIFIED LIVE. (1) 750 W7 — written directly into ActiveCampaign across all three subject variants of the wired campaign 4147 (msgs 13685/13686/13687), read-back OK on each: 'I'm not going to invent a deadline. What I'll say is this: this is the new-member window.' -> 'This is your new-member window, and it closes.' The retired predecessor campaign 3103 was deliberately left alone (unwired since the 10 Aug rebuild; 15,546 historical sends). Tool: tools/email-sequences/fix_750_w7_deadline.py (idempotent, backups in marketing/email-sequences/_AC_750_W7_DEADLINE_BACKUPS_2026-08-10/). (2) 993 D13 — fixed at the builder, rebuilt, 19/19 lint PASS, re-uploaded 16/16 read-back: 'The offer comes back around.' -> 'That price is gone - it closed with your window.' Live grep for 'comes back around' on 993 now returns 0. The 60-day guarantee's 'every cent comes back' is a different promise and was left untouched in all four emails carrying it. D1's 'ten days / we don't do this often' and D11's 'tonight at midnight' were already consistent and are unchanged. REVIEW.html now carries the ruling and the before/after. 🔴 CARRY-OVER: the copy now asserts a deadline that /50promo/ does not enforce - the page never expires. That is NIC-95, now load-bearing for this ruling rather than cosmetic.

CLOSED on artifact evidence by the 10 Aug selflearn run: campaign 3715 is NO LONGER in 750 TMA | Welcome Sequence After Quiz. Aga rebuilt the whole arc in the AC builder on 10 Aug ~08:07-08:18 CDT (A/B/C splits on every slot, one email deleted; backups in marketing/email-sequences/_AC_750_SPLIT_BACKUPS_2026-08-10/). Live flow verified via ac_email_read.py 750: campaigns 4146-4153 only - 3715 is gone, so the pull-3715 premise no longer exists. Verified against the live automation, not a doc.

SENT on Aga's 'send them', 8 Aug 2026 — 39/39 sent, 0 failed, 0 skipped, with a final live buyer-check per person at the moment of send. Message ids: tools/retention/quiz_abandon_sent_log.csv (verified against the Gmail API responses). Copy = Aga-iterated v8: personalised goal line, member proof with real demographics, per-person verified persona checkout, 48h window, humanized. Replies are watched by check_replies.py from tomorrow 05:35 (composed Jesse drafts; STOP auto-suppresses). Recovery = paid-in-Stripe within 96h; fleet reads it from 11 Aug.

CLOSED on artefact evidence, not on the status flag. 750 TMA | Welcome Sequence After Quiz was paused 16:41 and un-paused 18:03 (automations/750 → status 1, mdate 18:03:43). The closing proof is an ARRIVAL, not the flag: automations/750/contactAutomations shows a new entry at 18:22:50, 19 min after the un-pause, where the previous newest was 16:15:10 — so the trigger was untouched and the arc genuinely receives again. Outage ~82 minutes; at the measured ~21-25 arrivals/day that cost roughly 1-2 quiz leads their welcome. No action was ever needed from Aga — whoever paused it reversed it before she saw the row.

APPROVED BY AGA 7 Aug PM — holdout ARMED at 1-in-5. Executed the same session: HOLDOUT_EVERY=5 in tools/retention/quiz_abandon_recovery_read.py, and wired into the LANE's skip chain (quiz_abandon_manual_notes.py) so a held-back address is never drafted — reason logged as 'RECOVERY HOLDOUT (control arm, 1-in-5)' in the manifest. Bucketing is a stable SHA of the address, so the same person is always in the same arm with no state file to drift. 🔑 EXPECT holdout=0 FOR THE FIRST FEW DAYS — the 39 already emailed stay in the touched arm forever (you cannot un-email someone), so the control arm fills only from NEW abandoners at ~2/day; ~20 per arm is the threshold for an honest read, so ~6-8 weeks. The tool now prints 'HOLDOUT TOO SMALL' until then rather than implying a lift exists.

MY OWN FILING ERROR, corrected in the same session. Every other FLEET-nn row lives on the NIC side of the queue — build_todo_page.py filters non-Nic owners out of Nic's brief, so that is where an agent-owned row belongs. Putting this one on the aga side pushed a task Aga does not do onto AGA_TODO_NOW.html as item 24, with no walkthrough. Re-minted verbatim as FLEET-38 on the nic side; no work is lost.

DONE 6 Aug 2026 by the fleet, verified on the live public pages. Scanned all 410 published TMA WP items: the defect was 18 internal->internal utm_source=tma links across 6 items (page 252941 get-started-with-calisthenics x13, posts 214521/216170/214501/214677/251165 x1 each) — these restart the GA4 session and relabel an externally-sourced visitor as internal. All 18 stripped via the WP REST API, raw content backed up BEFORE/AFTER to tools/analytics/wp-backups/aga73-2026-08-06/. Also fixed the mislabel this row names: the sole quiz CTA on get-started-with-calisthenics sits at 3% through the page (the hero) but was tagged utm_content=prefooter-banner — now reads hero (2 occurrences). CRITICAL CORRECTION to this row's framing: utm_source=tma is NOT a blanket bug — it is the deliberate T1.1 measurement spec that tools/seo/tma_conversion_meter.py filters on, and all 542 cross-domain quiz.* CTA links were deliberately PRESERVED. Verified live post-fix: 0 internal-utm links remain on the fixed pages, quiz CTAs intact. NOTE for the strategy session: the meter's own header states utm_source=tma reads 0 by construction because the quiz-root 301 strips UTM params (the one-line $is_args$args fix never shipped) — so this measurement is not yet working regardless.

CLOSED 6 Aug 2026 — proven end-to-end, nothing outstanding. (A) Aga pointed all 10 Leadpages pop-ups at https://themovementathlete.com/resource-thank-you/ — THE thank-you page for ALL lead magnets, replacing the retired /magnet-thank-you/. (B) Fleet built + published the GTM tracking: variable 'URL - magnet (query)', trigger 'TMA - Lead Magnet Thank-You Pageview' on /resource-thank-you/, tag 'GA4 - generate_lead (magnet)' to G-ZBQN3P8WKW with lead_type=magnet + magnet_slug — container v116. (C) FIXED IN PASSING: the page is served by WP code snippet 11 which bypasses the theme header, so the GTM container was not loading there AT ALL (0 refs vs 3 on a normal page) — container injected into the snippet, verified live. (D) PROVEN END-TO-END in a real headless Chromium against the live URL, reading actual network requests: gtm.js loads then GA4 receives en=generate_lead tid=G-ZBQN3P8WKW ep.lead_type=magnet ep.magnet_slug=e2e-fleet-test. Observed on the wire, not inferred from config. (E) Star rating corrected 4.7 to 4.8 (visible number, aria-label, fraction-fill 70pct to 80pct, 3 comments) — 6 of the 11 '4.7' strings were SVG path coordinates and were deliberately NOT touched; asserted the 5 star paths byte-identical after the edit. Verified live. (F) Ruling propagated estate-wide via estate-sync: canonical_strategy fact + enforcement ban, Nic's brief (14 stale refs repointed, Step B rewritten as already-done with a do-not-rebuild warning, wrong measurement id G-20210510 removed), cockpit, SEO master md+html, task page. NOTED not changed: legacy tag 'GA4 - Lead' also fires here via trigger 186 (URL contains thank-you) — different event name so no double-count, pre-existing, and narrowing that trigger is Aga's call since it could affect other thank-you pages.

APPROVED BY AGA 6 Aug — both rows approved and their nic_approved flags set to yes, so they can now reach Nic's brief. NIC-12 is largely already executed by the fleet under AGA-123 (18 x 301 + 17 pages drafted, verified). NIC-07 approved but explicitly RANKED LAST on its own evidence: +13 to +32 clicks/28d, about half a lead a month, and 20 minutes of work; the note on the row tells Nic not to let it displace anything carrying a cash number.

GO GIVEN BY AGA 6 Aug — EXECUTED AND VERIFIED THE SAME SESSION. 18 x 301 created via the Redirection plugin REST API (13 TMA + 5 BWA) and 17 source pages drafted, all backed up to tools/seo/apply_backups/*-PRE-RETIRE.html. VERIFIED LIVE: all 18 return 301 in ONE hop to a 200, and every target page (/blog/, /start-training/, /30-off/, /resource-thank-you/, /thank-you-50/, /you-are-out/, /why-tma-page/, /quiz/, /calisthenics-blog/, /calisthenic-fundamentals-book/) confirmed 200. Nothing deleted anywhere — draft + 301 throughout, so unverifiable external backlinks keep working. The live uncontrolled 70%-off page is closed: /70-off-new-aug24/ now 301s to /start-training/. It had THREE WORKING legacy checkouts on it (/register/lifetime997/black-friday/, /register/quarterly2/, /register/yearly3/ — all 200) plus a lifetime997/special link that 302s to trial/lifetime497, and hardcoded _gl= tokens; anyone who found that page could buy through checkouts that fire GTM-PPGJXHH, a container not in our account. Also fixed: /why-tma-old/ was serving HTTP 500 to ~13 sessions/28d and now redirects cleanly.

DONE BY AGA 6 Aug 2026, same session it was raised — both grants landed and the live read is WORKING. (1) Gmail Postmaster Tools API enabled on project 927298485621: verified by me, the call went from 403 SERVICE_DISABLED to 200 OK. (2) Domain-wide delegation authorised in Workspace admin for client_id 113899261199989199070 with scope postmaster.readonly. Two snags worth recording because they will recur: Google rejected the first paste of the client id as 'Invalid client ID' because it was ONE DIGIT SHORT (20 of 21 characters — a truncated paste, not a typo she could see), and the first API call after authorising returned unauthorized_client because delegation had not propagated yet — it succeeded on the retry ~30 seconds later. VERIFIED LIVE: domains visible = ['domains/themovementathlete.com'], 74 days of trafficStats pulled, written to data/email-analytics/postmaster/SPAM_RATE_SERIES.json with provenance flipped from TRANSCRIBED to API. Hardened while closing: postmaster_read.py now names all four causes of unauthorized_client instead of Google's opaque message, and --save merges hand-written per-day notes forward (the first save silently deleted them; fixed and re-verified — 16 notes survive a genuine API save).

DONE 6 Aug 2026 — Aga supplied vendor number 87822406, written to .env.local as ASC_VENDOR_NUMBER (gitignored; the loader does not strip inline comments, so it must sit on its own bare line). The App Store Connect financeReports endpoint works with the EXISTING Admin key — no Finance-role key was needed. ANSWER TO THE GROSS/NET QUESTION: 'Total Estimated Proceeds' is NET, and so is RevenueCat's App Store revenue — so the 1.02x ratio was two net figures agreeing, NOT a defect. Evidence: Apple's fiscal-May report totals GBP 2,678.85 net vs GBP 3,448.95 gross, and the figure already stored for May was GBP 2,684.01 (0.19% from net, 28% from gross); RevenueCat App Store revenue 3,844 May / 3,085 Jun / 3,089 Jul USD sits within 2-7% of Apple net and 17-23% BELOW Apple gross, three months running. CONSEQUENCE: the '~15% overstated / ~525/mo phantom app cash' claim is REFUTED — no correction to the headline was needed. Two real fixes did land: July iOS was null and is now filled from Apple at GBP 2,365.68, and the kpi_15k_sync guard that fired on ratio>0.95 was REVERSED (it now warns on divergence outside 0.90-1.10) so the false 'app cash may be ~15% high' warning no longer reaches the cockpit. TWO TRAPS FOUND AND FIXED IN THE PULLER: (1) filter[reportDate] is Apple's FISCAL period, not the calendar month — 2026-06 returns 1-28 Mar; fiscal index = calendar month + 3. (2) Refunds carry a negative quantity AND an already-negative customer price, so multiplying them flipped refunds positive and produced a bogus EUR 0.29 ratio. Remaining: Google Play is still manual (no API access) — that half stays on AGA-105.

RATIFIED BY AGA 6 Aug 2026 verbatim ('yes this is correct'). APPLIED IN FULL, same session. (1) OFFER_LADDER.md Rung 3 rewritten to '50% off — the new-member window' with the two legitimate placements (real per-contact deadline / declared calendar burst) and the ban on ambient discounting, plus a note carrying the evidence (488/538 = 90.7% of new annual at exactly 50% off; zero new annual ever at $157). (2) NEW §1 hard rule added: '50off/ CHARGES IMMEDIATELY — never write free-trial copy against it', listing the banned phrases and the instruction to read the href the CTA points at, not the offer in the subject. (3) Append-only change-log entry written. (4) Propagated the same session to .claude/commands/email-fleet.md (KB table row), .claude/agents/email-qa.md (new blocking check 4c) and .claude/agents/tma-email-offer-strategist.md (refusal #4 restated as ratified law). (5) Source doc EMAIL_REVENUE_STRATEGY_2026-08-05.md §5 contradiction ① stamped RESOLVED with the old text kept for the record. (6) ENFORCED IN CODE — tools/retention/email_lint.py gained the permanent check 'no trial copy against a charge-immediately link' (helper _text() strips tags so an emphasis-split sentence still matches; the regex deliberately catches BOTH 'charged until day 8' and 'charged TILL day 8', the spelling that survived the 5 Aug sweep). VERIFIED: 8/8 synthetic cases correct, then run live against 170 messages across 750/751/758/759/973/764/760 — ZERO failures and ZERO false positives, confirming both that the 6 Aug fix held and that the rule can no longer silently return. Re-run: python3 -c "import sys;sys.path.insert(0,'tools/retention');import email_lint" or python3 tools/retention/email_lint.py --ac 750

APPROVED BY AGA 6 Aug 2026, WITH ONE AMENDMENT: 'yes to this but I am going to create and prove the emails.' So the copy is HERS, not the fleet's — /email-fleet does NOT write this send. Everything else stands: Core tier only (~5,090 opened-or-clicked in 90d), never the dormant tier, unique coupon code for attribution, collision check before send, and the $78.50 charge count read at 72h as the falsifier. Deliverability headroom CONFIRMED the same day, not assumed: Postmaster now readable (AGA-107) and the estate has run 0.0-0.3% since 1 Jul with 15 of 21 data days at 0.0%; the comparable Memorial Sale broadcast at 39-45k sends/day peaked at 0.2%, and this send is ~8x smaller. Follow-on work I own is tracked as FLEET-nn (audience build, coupon mint, collision check, 72h read) — this row was the approval, and the approval is given.

DONE BY AGA 6 Aug 2026, AND NOW READ STRAIGHT FROM THE API — not a screenshot. themovementathlete.com is verified in Postmaster and tools/analytics/postmaster_read.py pulls the series live via gmailpostmastertools.googleapis.com (service account impersonating hello@themovementathlete.com). 🔴 THE API CORRECTED MY OWN TRANSCRIPTION: reading her dashboard capture I recorded 26 Jun as 0.0%; it is 0.9%. Corrected totals over 74 data days 9 Apr - 3 Aug: 53 at 0.0% (not 54), unweighted mean 0.22%, peak 3.3% on 28 Jun, TWELVE days at/above Google's 0.30% never-exceed line (not 11), 6 days above 1.0%. THE FINDING IS UNCHANGED AND STRENGTHENED: all six days above 1.0% and nine of the twelve red-line days fall in JUNE, the one month with ZERO broadcasts (verified against AC: no status-5 campaign between 31 May and 11 Jul). The big-volume promo windows were CLEAN — Memorial Sale 22-31 May at 39-45k sends/day ran 0.0-0.2%; the 2-12 Jul lifetime window at ~12k/send ran 0.0-0.3%. Since 1 Jul: max 0.3%, 15 of 21 days at 0.0%. Promo VOLUME is not the complaint driver; the dirty band coincides with the dormant/sunset re-engagement lanes on a small denominator. CURRENT STATE HEALTHY. Re-run any time: python3 tools/analytics/postmaster_read.py live --impersonate hello@themovementathlete.com --save (or 'stored' for the cached series). Caveat stamped in the file: Postmaster's denominator is Gmail-inbox-delivered volume that day and AC cannot supply per-day Gmail volume, so no single day is a verdict.

DONE 6 Aug 2026 — Aga approved with 'fix all'; executed and verified. SCOPE WAS BIGGER THAN THIS ROW SAID: the row described 34 emails carrying a phantom deadline. Live scan of all 419 messages in the 32 automations found 71 divergent — 20 phantom deadlines AND 66 phantom offers (the row counted only the deadline class). Two spot-checks before writing confirmed they were real, not detector noise: 988 TMA | Welcome Sequence AFTER APPSIGNUP msg 13562's plaintext says 'Claim Your 50% Off - This Week Only' with a live /50promo/ link while its HTML makes NO offer at all; 750 TMA | Welcome Sequence After Quiz msg 11446's plaintext says 'for the next few days only, save 50% on any plan' with the same link. Both on EVERGREEN automations, so that clock has never been true for anyone. APPLIED to all 71 via the new tool tools/retention/ac_plaintext_regen.py: 71 clean, 0 needing a look. VERIFIED, not asserted: full message JSON backed up BEFORE and AFTER every write (71 pairs in marketing/email-sequences/_AC_PLAINTEXT_REGEN_BACKUPS/), and diffing those pairs proves the HTML is BYTE-IDENTICAL on 71/71 — only the plaintext changed, so no copy was written or removed and EVOLUTION_LAW was not engaged. 🔴 WORTH KNOWING: 42 of the 71 are 988 and 992 — the same pair the forensic named as having ZERO HTML offer hrefs. Regenerating their plaintext does not add an offer; it stops the plaintext pretending one exists, which makes that offer gap fully visible instead of half-hidden in a surface almost nobody renders. Closing the gap properly (a real HTML offer in those arcs) remains PART 5 work.

DONE 6 Aug 2026 — BOTH halves closed the same session, and NEITHER needed Aga to read a dashboard. Apple: she supplied vendor number 87822406, and tools/finance/asc_finance_reports.py filled July iOS at GBP 2,365.68 from the App Store Connect API. Google Play: she granted the bulk-reports bucket read, so the 30 Jul '403' note is now stale — tools/finance/play_earnings_reports.py pulls Google's official earnings report directly. THE PLAY DIAGNOSIS WAS CORRECT AND IS NOW MEASURED: the file held play['2026-07'] = GBP 318.24, the true net payout is GBP 730.65 — out by GBP 412.41, exactly the half-month artefact predicted (typed 16 Jul, mid-month). May (115.78) and June (378.37) matched Google's report TO THE PENNY, which confirms the file's basis was always right and only July was stale. The Play ratio moves 0.3184 to 0.7311, and Google's fee is exactly 15.0% of charge in all three months — confirming Aga's 'Google takes 15%'. Scoreboard re-run: pace 7,822 vs 15,000, gap 7,178, app_warning None. 🔴 COVERAGE LIMIT: the bucket holds NO Play earnings before 2026-05, so the 'store exports back to 2024' ambition (legacy task 18) cannot be met from store data at all — that needs a different source, and the file's pending list now says so instead of implying someone forgot.

DONE 6 Aug 2026, verified live. Aga enabled the Tag Manager API and added the service account — confirmed by listing all 3 workspaces via the edit-scope API. The container permission had been set to Read rather than Publish; the fleet completed that grant itself using the account-level admin Aga had granted (fulfilling this row's own 'with Publish rights' instruction). Then BUILT AND PUBLISHED the tag: variable 'URL - magnet (query)' id 226, trigger 'TMA - Magnet Thank-You Pageview' id 227 (pageview on /magnet-thank-you/, confirmed 200 live), tag 'GA4 - generate_lead (magnet)' id 228 -> G-ZBQN3P8WKW with lead_type=magnet + magnet_slug. Published as container version 115; verified by reading the LIVE version back and confirming the tag is present. Workspace checked for other pending changes before publishing (only these 3). Every future GTM change is now the fleet's job, not Aga's.

✅ RULED YES by Aga, 5 Aug 2026: "yes of course" — TMA honours the 60-day promise on the two discounted in-app pages. If a buyer emails inside 60 days we make good by whatever route works: an Apple refund request via reportaproblem.apple.com, or compensation on our side (free months / refund on another rail) when Apple declines. The seal copy therefore SHIPS AS WRITTEN — 'Feel stronger in 60 days — or your money back. Train for two months. If you don't feel the difference, email us and we'll make it right.' No re-cut needed. The trial page was never in question (it leans on the platform promise, 'you won't be charged for 7 days', which Apple enforces). ⚠️ The operational half now belongs to support: hello@ must recognise a 60-day refund request from a STORE buyer as honour-it, not forward-to-Apple-and-stop.

RULED BY AGA 7 Aug PM: THE 48-HOUR WINDOW IS BINDING — but the engineering half is NOT urgent. Her words: 'Binding: this is a task for Nick, but it's not urgent. He needs to focus on their things. Put it on Aga's list to then approve it for Nick, but for now we just say it is binding.' TWO CONSEQUENCES, both live from now: (1) DOCTRINE — the window is binding, so the estate stops treating it as an open question; (2) 🔴 THE COPY CONSTRAINT SURVIVES UNTIL ENFORCEMENT SHIPS — the server still never returns expired (AGA-68's probe: TMA_QC_OFFER_EXPIRED never true, lapsed entries re-mint a fresh 48h), so NO email may claim the deadline yet. That is why the 1:1 note's three 48h claims were replaced (D6) and why the AC arc's E3 cannot ship its 'closes tonight' subject. Successor row created for her approval of the Nic task when his queue clears.

ANSWERED BY AGA 4 Aug, both halves. (1) PROVENANCE CONFIRMED: 'yes I did' — Aga assembled marketing/LIFETIME-MEMBER-TESTIMONIALS/LIFETIME TESTIMONILS WALL .png herself from real screenshots. It is therefore a GENUINE capture montage and is the first-choice proof image estate-wide, safe in any context including ad review and press. Consistent with the pixels: real Gmail dark-mode chrome, real Facebook recommendation cards with real profile photos, real full names (Raya Riley, Tanya Richards, Tim Cook, Mpho Moralo, Quentin Vaucelle-Auzel, Ann DesJarlais, Michael Gtz), uneven crops. (2) THE RENDER SET IS ALSO APPROVED: shown the evidence that LT-01..LT-20 were built 1 Jul with faces reused from the member face pool, that 02_proof-walls is composited from those 20, and that LIFETIME_PROOF_BANNER is the same set, Aga ruled 'yes these are fine'. The whole lifetime library is usable. TWO LIMITS RECORDED IN LAW, the entire remaining exposure: never caption a composited one as a real/unedited screenshot, and never submit one where a platform asks for authentic customer evidence (ad review, app store, press, regulator) — for those, use Aga's real wall. STILL BANNED: generating NEW fabricated testimonials; this permits the existing library, it does not reopen the generator.

DONE BY CLAUDE 4 Aug, CORRECTED after Aga caught a real error. All 7 enter-blocks feeding 973 Nurture Sequence 1 NEW COPY FROM CLAUDE now point at 751 Nurture Sequence 1 — NOT 758 Nurture Sequence 2 as I first set them. MY ERROR: 751 and 973 are two versions of the SAME rung (both hand off to 758), so pointing the welcome arcs at 758 made every new lead SKIP nurture 1 entirely — 12 emails of proven content (20,942 sends, 22% opens, closes with 'The honest case for joining now'). I had wrongly generalised Aga's 29 Jul ruling, which was about where the STUCK COHORT already past nurture-1 content should go, into where NEW leads should ENTER. Aga confirmed 751 is the nurture 1 we keep. LADDER NOW VERIFIED END-TO-END: welcome arcs (750 TMA | Welcome Sequence After Quiz, 967 TMA | Welcome Sequence After LEAD MAGNET, 760 School of Strength (NEW), 766 Facebook Lead Form LM Building Muscle After 40, 976 Cancelled Win-Back, 988 TMA | Welcome Sequence AFTER APPSIGNUP, 992 TMA | Welcome Sequence AFTER APPSIGNUP AB TEST Aug4th) -> 751 -> 758 -> 759 Nurture Sequence 3 -> 764 Nurture Sequence 4 -> END. 973 is now fed by NOBODY. All 7 still status=1, block counts unchanged, automation mdates unmoved. 973 holds 8 stragglers; tomorrow's 09:40 run sweeps them to 758 (correct — they already had nurture-1 content), then com.tma.n973to758 can be unloaded and the register row deleted.

RULED BY AGA 4 Aug: HOLD — 'do not message partners.' Nothing sends. The 3 gate-passed drafts (Rick Donohoe, Jari Dohmen, Rob Turp) stay unsent in the outreach log; do not re-raise partner outreach without her opening the subject.

ALREADY DONE — no action needed. Verified live 4 Aug 2026 with header-only curl (no -L): /test-2/ -> HTTP/2 301 location https://themovementathlete.com/start-training/ and /blackfriday24/ -> HTTP/2 301 same target; /start-training/ 200. Both 301s were applied in the 29 Jul promo-retirement batch (PROMO_RETIREMENT_PACK_2026-07-29.md lines 111/128/131). The 3 Aug P11 audit that raised this row hashed the three URLs with curl -L, which FOLLOWED the existing 301s and returned three identical hashes of /start-training/ — the 'byte-identical duplicate money pages' finding was an artefact of the fetch method, not a real duplication. Method fix: check status headers before hashing. Residual (not Aga's): the site-wide Simple Custom CSS and JS snippet gated on window.location.pathname === '/test-2/' (rewires .pboxblack1 to tma.thrivecart.com/speciallifetime297/) is now permanently dead code and can be removed from Code Snippets.

RULED BY AGA 4 Aug: GOLD #FFB000. Applies to CTA buttons on WordPress pages on both sites and to every future block — the magnet thank-you page (already gold) stays as built; the new blog inline + end-banner blocks change from the 3 Aug orange #F09819 to gold. One-token change per file.

RULED BY AGA 4 Aug: DROP. Remove every fabricated/AI-generated testimonial image; keep the verbatim quotes as callout boxes. Replacements come ONLY from the approved real sources she named: the COMPILATION (marketing/LIFETIME-MEMBER-TESTIMONIALS/LIFETIME TESTIMONILS WALL .png, marketing/assets/LIFETIME_PROOF_BANNER COMPILATION.jpg, social-proof-screenshots/lifetime/compilation/LIFETIME_PROOF_BANNER.png|jpg), the PROOF WALLS (02_proof-walls/, 4 files H+V), the 41 individual real captures in 01_individual-testimonials/ (+22 _AIRDROP-JPEG, +9 _montage-extracted), the tagged TESTIMONIALS-DATABASE, and anything already published on our own pages. Her words: 'you can easily use ANY testimonial in our database and on our pages' and 'I have asked you at least 5 times about this'. ESCALATED TO A PROJECT-WIDE LAW binding EVERY fleet (growth, email, social, YouTube, ads, SEO/pages, checkout, quiz) — written into CLAUDE.md as the TESTIMONIAL RULE with the source table, because the 27 Jul ban prohibited fabrication without ever telling agents where the real ones live. Memory: feedback_testimonials_real_only_approved_sources.

ALREADY DONE — no action needed. VERIFIED LIVE 4 Aug by fetching https://themovementathlete.com/calisthenics-assessment/ twice, once clean and once with ?utm_source=youtube&utm_medium=organic&utm_campaign=probe (the exact condition that used to trigger the shim). BOTH responses: tmaUtmFwd 0 occurrences, '/2026-v2-2' 0, no '/2026-' path anywhere, and all 4 quiz links point at the ROOT quiz.themovementathlete.com. Page 214348's stored content also contains 0 occurrences (context=edit read). The shim was removed on/around 3 Aug and this row went stale. 🔴 MY ERROR: I carried the stale row into Aga's runbook and asked her to re-rule something already settled by her standing quiz-root law — a CHECK-BEFORE-YOU-ASK breach. Aga's response: 'why are you asking me again? we already ruled this, ALL redirects need to go to root quiz, that is FOREVER a rule.' NOTE the utm_source=tma overwrite on the nav/footer links is still live and is a DIFFERENT row (AGA-73), still open.

Self-test row for the CHECK-BEFORE-YOU-ASK code gate — confirmed --checked is required on the aga side and that its text persists to the row as checked_not_already_done. No real work; safe to ignore.

RESOLVED by Aga 3 Aug — it was not an error in the doc, it was a MISSING FUNNEL DISTINCTION. Both flows are real: FUNNEL 1 (web/quiz lead) = assessment is FREE, then paywall, then 3 free workouts unlock — the doc was correct for this and its 4 approved framings hold. FUNNEL 2 (app install via Apple/Google ads, never touches the web quiz) = small in-app onboarding, 3 free workouts from the GENERIC library plus an example plan shown, paywall immediately, and the ASSESSMENT REQUIRES THE UPGRADE. Applied: PRODUCT_TRUTH.md now opens the assessment section with a FUNNEL SPLIT warning naming Funnel 1 as its scope, the approved-framings list is labelled Funnel-1-only, and a new locked 'Funnel 2 — APP SIGNUP' section documents the real app flow plus its copy rules (no free-assessment ask, one CTA door to /start-training/, 15-minute floor, buttons say trial things). Committed to memory as reference_two_funnels_assessment_and_paywall and indexed in MEMORY.md. Verify: grep 'FUNNEL SPLIT' on PRODUCT_TRUTH.md. Matches the cockpit's existing Funnel 1 Web / Funnel 2 App tabs.

RULED BY AGA 4 Aug: WIRE IT — 'I am trying to wire it.' Pinterest stays in the team's remit and gets measured. Fleet builds the Pinterest developer app + OAuth (~1h) so the board scores OUTBOUND CLICKS — Pinterest's real currency, judged on 90-day clicks-out, never weekly impressions (a ~3.7-month content half-life makes weekly reads meaningless). Acceptance: platform_registry.pinterest.lane flips DARK to AUTO in tools/social-pipeline/social_status.json and clicks render on the board. Aga may need to approve the developer app in her Pinterest account — fleet will ask for exactly that one step and nothing more.

RULED BY AGA 4 Aug: KEEP 4.8. 'We stick with 4.8 — there is also Facebook and our own internal things.' 4.8 is a BLENDED cross-platform figure (App Store + Play + Facebook recommendations + internal member feedback), not a storefront reading, and it is the number we publish. Store-only figures (4.53 iOS / 4.7 Play) stay inside ASO analysis. Never attribute 4.8 to one storefront; 4.9 remains retired. 🔴 CONSEQUENCE: the new blog end-banner renders 4.7 and must be corrected to 4.8 before it ships to 162 BWA pages; sweep anything else built 30 Jul-4 Aug that adopted 4.7. Written into CLAUDE.md so it is never re-asked. Memory: reference_published_rating_is_4point8_blended.

ANSWERED by Aga 3 Aug. (1) SESSION FLOOR = 15 MINUTES, confirmed ('yes our workouts start from 15 min') — so the arc's 'ten minutes' promise (14 occurrences across 5 of 6 emails, including E1's subject line) is FACTUALLY WRONG and must be corrected to 15. PRODUCT_FEATURES.md line 58 was right. (2) FLOW CONFIRMED: install -> survey -> assessment -> 3 free workouts unlock immediately (matches protected-core PRODUCT_TRUTH lines 20-22) -> those free workouts are NOT personalised (they are the on-demand library) -> to truly unlock features they must UPGRADE and take the ADVANCED ASSESSMENT that positions them across all the skills on the skill tree -> the calibrated progressive engine is what the paywall unlocks. (3) STRATEGIC RE-BRIEF, bigger than the two fixes: Aga ruled the arc's job is NOT just activation-to-first-workout, it is EDUCATE-TO-UPGRADE via direct-response copy, teaching the system, autoregulation, microprogressions, real member results, social proof, the backing and how the system was created, how personalisation actually works, and gamification/mastery/unlocking as the motivation key. The current 6 emails cover personalisation (E2) and microprogressions (E3) and thin proof, but are MISSING: the credibility/origin story, gamification-mastery-unlocking, and critically the upgrade wedge itself (the arc never acknowledges the reader's current workouts are generic). Consequence: this is a re-brief through /email-fleet, not a patch. 968 has never deployed, so the EVOLUTION LAW control lock does not bind (it protects LIVE automations). Approved framing to use, verbatim from PRODUCT_TRUTH line 30: 'The on-demand library gives you real workouts from day one. The progressive system - the part that calibrates to you - is what the trial unlocks.'

REVERSED BY AGA 4 Aug — do NOT shrink the token. Her instruction: 'I actually would like us to use other things like respond to comments and messages etc.' The wide scope set (instagram_manage_comments, instagram_manage_messages, pages_manage_engagement, pages_manage_posts) is now WANTED, not surplus — she intends the fleet to reply to comments and DMs. Keep the current token. HER QUESTION, answered on the fleet side: the security exposure is real but modest and it is about BLAST RADIUS, not breach — the token never expires, sits in .env.local, and was pasted into a chat transcript, so anyone holding that string could post as the brand and message followers. Since the scopes are now deliberate, the mitigations change from 'reduce scopes' to: (1) ROTATE once, because the current string has been in a transcript — same permissions, new secret; (2) keep it out of chat from now on; (3) set a calendar rotation. Superseded by a FLEET row covering the rotation and the comment/DM reply capability she actually asked for.

RULED AND APPLIED 3 Aug 2026 (Aga delegated the call to the CMO seat); leads-tracking addendum applied 4 Aug on Aga's instruction 'reset it but let's still keep tracking the leads'. P4-youtube moved REVENUE -> CAPABILITY tier, graded on VIEWS not leads, exception-only reporting; the withdrawn $200-350/mo makes the $15K gap larger and honest. Tripwire kept: views/28d >= 45,000 (~3x today) re-opens the charter and the lane re-enters revenue. ADDENDUM 4 Aug: added a 'watched' block to the capability entry (4 metrics + their json_paths + the 3 surfaces) and taught aga_console_sync to resolve those values LIVE, so leads are measured daily without being targeted. RE-RUN TO VERIFY: tools/finance/venv/bin/python tools/ops/aga_console_sync.py then grep 'watched, not graded' in docs/company/ANALYTICS/GROWTH-COCKPIT/index.html -> '206 real sessions/28d -> 3 leads · 1.5%'. Confirmed live on tma-growth-cockpit-2026.netlify.app after estate_sync. Same numbers also render on tma-social-scoreboard.netlify.app YouTube tab and in the 06:45 Slack digest.

SOLVED IN CODE, no manual trimming needed — 3 Aug. The 3 videos were over YouTube's 5,000-char limit by only 69/144/263 chars. Root cause found: each carried the same assessment CTA twice, and once rewritten each copy became a long UTM'd quiz-root URL. Fixed automatically by (a) deduping exact repeat lines, (b) capping hashtags at 15 — which YouTube requires anyway, since a video with >15 has ALL its hashtags ignored. All 3 now fit (max description across the channel is 4,878 chars) and are back in the automated pass. Aga does nothing.

SUPERSEDED BY AGA'S OWN RULING 4 Aug. She published the redesign, judged it (not full width, colours off, needs mobile optimisation), then scrapped that line entirely: 'the page is already done here https://themovementathlete.com/resource-thank-you/ — I will remove the other page, THIS is the approved page.' 🔴 CANONICAL LEAD-MAGNET THANK-YOU PAGE = https://themovementathlete.com/resource-thank-you/ (verified live 4 Aug: HTTP 200, title 'You are in — check your inbox', 4 quiz-root links, noindex present). Draft 255927 and the published /claude-draft-youre-in-check-your-inbox/ are BOTH retired — Aga removes the latter. 🔴 One defect on the approved page: it renders 4.7 stars 11x and must go to 4.8 per her same-day rating ruling.

VERIFIED — criteria amended AND the 2 stuck entries resolved, but NOT as approved-on-the-numbers. New bar in CONTROL_EMAIL_LIBRARY.md: a performance_proven control needs LIVE PAIRED metrics (clicks ahead of a NAMED concurrent comparator AND unsub within benchmark, <=0.30% welcome-slot / <=0.50% hard-stop), never a 180/200 self-score. I then tested the 2 pending entries against that bar and they FAILED it: queried the AC API live this session (/api/3/campaigns/3057 and /3065) and BOTH return 0 for sends, opens, clicks AND unsubscribes -- the stored 25.9%/1.3% and 21.3%/2.1% figures no longer reproduce and no unsub figure exists, so requirement 2 is unmeetable. Ruling applied: keep the craft, drop the claim -- both flipped PENDING_APPROVAL -> ACTIVE as control_type craft_exemplar (their techniques[] remain useful to Agent 07), rates tagged [UNVERIFIED] and explicitly barred from being quoted, cited in a brief, or used as a comparator. Consequence to know: the estate has ZERO performance_proven controls today; the first will come from the per-slot scoreboard, which records clicks and unsub together from daily snapshots and will still answer the question in six months.

VERIFIED — one scheme ruled: BOTH forms valid, reality wins. Correction to my own brief: I claimed the validator rejected our live links; I tested it and it PASSES. The real divergence was that utm_builder.py build CRASHED on a descriptive slug while validate accepted it, so the fleet would have minted a second incompatible scheme going forward. Fixed by widening SLOT_RE in tools/email-analytics/utm_builder.py to accept the short slot form (e1/o3/d5) AND the live descriptive slug that ac_utm_tag.py actually stamped on all 1,077 links (100% tagged, per docs/company/ANALYTICS/GROWTH-SCOREBOARD/data_email_utm_audit.json pulled 3 Aug 11:21). No re-tagging of live links needed; history stays joinable across the 29 Aug revenue gate. Re-run to confirm: utm_builder.py build --content r1-you-have-ignored-my-last-19 now succeeds, --content e6 still succeeds, --content 'E1 The Method!' still raises ValueError, and the 968 arc still passes the UTM gate 6/6. Standard updated with the ruling.

VERIFIED — DIRECTIONAL tier RATIFIED and live. Aga approved 3 Aug ('follow whats recommended'). Written into EXPERIMENTATION_DECISION_RULES.md as sA-bis (volume tiers A/B/C + the 7 bounds: pre-registration, >=200 sends/side, big-swings only, winner's-curse x0.5, guardrail auto-revert, may promote or revert but may NEVER mint a CALL/learned rule/playbook entry, confound invalidation) and sA-ter (rollback rule ported from the deprecated EXPERIMENT-REGISTRY.md so deprecating it does not delete the estate's only rollback clause). Re-run to confirm: grep -n 'A-bis' on the file. EVOLUTION_LAW.md Appendix A flipped from DRAFT to RATIFIED. Consequence: EXPERIMENT_LEDGER.json record 2026-W32-968-APPSIGNUP-REMEDIATION is now unblocked on the instrument and waits only on AGA-60 cutover approval.

Aga ruled in-session 3 Aug 2026: the assessment duration figure is 3 MINUTES ('3 min quiz'). All surfaces converge on 3 minutes; magnet-thank-you page updated same session.

WITHDRAWN — the premise was FALSE and I verified it against the code instead of re-asserting it. ac_utm_tag.py matches eligibility with 'any(tracked_host in url)', a SUBSTRING test — and 'app.themovementathlete.com' CONTAINS 'themovementathlete.com', which is already in TRACKED_HOSTS. So checkout/app links were never excluded. My earlier check tested the reverse direction (is the full host a substring of a tracked host) and returned a misleading False. Full dry run across the LIVE estate on 3 Aug returns 8 untagged links in 1 automation, all of them 990's brand/success-stories links (990 had been skipped by AGA-55 only because it is INACTIVE, and the tool walks live automations) — the rest of the estate is 0 dark, consistent with AGA-55's 1,101/1,101. Those 8 are now tagged via ac_utm_tag.py --apply --yes --only 990 after probing both endpoints preserve the query string (Rule 2). No estate-wide change was needed and none was made; TRACKED_HOSTS is unmodified. Also: this class of work should never have been queued for Aga at all — her 3 Aug rule feedback_instrumentation_is_not_a_navigation_change says UTM tagging on live AC emails auto-applies and never queues.

ANSWERED BY AGA 4 Aug — the missing URL is now known and the scope is bigger. CANONICAL DESTINATION = https://themovementathlete.com/resource-thank-you/ (live-verified 200). Her instruction: 'ALL lead magnets should be updated with this thank-you page now' — so this is no longer one leadbox, it is all 10. Folded into AGA-58, which is the same Leadpages screen and now carries the confirmed URL. Note the redirect must be published, not saved — Leadpages 'saved' is not 'published', the trap that cost ring-dips six days.

VERIFIED LIVE 3 Aug 2026. Root cause was the APP's use-case config, not the token: Aga added 'Manage messaging & content on Instagram' + 'Manage everything on your Page' to app TMA Analytics (826836293077394), an app admin approved them, and she regenerated the Analytics Reader system-user token (SYSTEM_USER, expires_at 0) now carrying instagram_basic + instagram_manage_insights + pages_read_user_content + read_insights with all 8 prior scopes intact. Swapped into .env.local via the guarded swapper (backup outside repo, mode 0600). Also pinned IG_USER_ID=17841403844381324 because auto-resolve silently picked @forgemovementathlete (2 followers) over @themovementathlete (43,452). ig_insights.py extended to pull EVERY linked IG account into per-account files (Aga: 'they are both TMA'). RE-RUN: tools/finance/venv/bin/python tools/social-pipeline/scripts/ig_insights.py --check -> two green lines (@themovementathlete primary 43452 followers, @forgemovementathlete); --limit 60 -> 60 + 3 reels with reach/saves/shares. FOLLOW-UP OPEN: the new token carries publish/messaging/WhatsApp write scopes far beyond read-only, and the OLD token is still valid and now carries them too.

DONE 3 Aug 2026 — Aga ruled the quiz root applies across ALL automations, not just 991. Surveyed the whole live estate: 50 quiz.themovementathlete.com links, of which 42 were already at the root and 8 were on the variant path /2026-v2-3/ (967 TMA | Welcome Sequence After LEAD MAGNET x4, 991 NEW Re-Engagement - For people who stopped opening Before Sunset x4). All 8 collapsed to the root with query strings and UTMs carried across intact. VERIFIED: probe shows the root preserves the query (200, 1 hop -> /2026-quiz-checkout-v1/?utm_source=... intact); raw API read of msgs 12296/13623 shows 0 variant-path hits and root+UTM links present; drift check now reports LINK RULE HOLDS. NOT cosmetic - the root serves /2026-quiz-checkout-v1/ while /2026-v2-3/ is a different, older quiz, so those readers were being sent to a stale variant. HARDENED: new tool tools/retention/ac_quiz_root_rule.py (dry-run default, backs up every body, idempotent, and REFUSES to apply unless it re-probes the root and sees UTMs survive the hop) wired into refresh_cockpit_daily.py --check so drift is caught daily.

TAKEN BACK BY THE FLEET 4 Aug — this was never an Aga task; its own text said 'NEXT STEP IS MINE, NOT YOURS' while it sat on her list. Exactly the failure the new DEFAULT-EXECUTOR RULE exists to stop. Investigation DONE, and it found something bigger than the missing UTMs. Live-read of page 252941 via the WP REST API (context=edit): the uagb/forms block carries 36 attributes and EVERY ONE is styling (colours, borders, padding, submitButtonText, variationSelected). There is NO afterSubmitAction, NO confirmationType, NO webhook, NO email target — no integration configured on the block at all. Corroborating: (a) no Spectra form-entry post type exists in /wp-json/wp/v2/types (only spectra-popup, 0 items); (b) no ActiveCampaign list or tag matches this magnet (searched all 100 lists + 200 tags for start/calisthen/getting/beginner — nothing); (c) no Code Snippet hooks the form (11 snippets read, the only match is a CSS button rule). STRONG EVIDENCE THE FORM IS DARK — leads may reach nothing, which would make 'no UTM capture' the lesser half of the problem. NOT PROVEN: Spectra's GLOBAL settings page and any server-side plugin (e.g. WP Fusion) are wp-admin UI, not REST-readable from here. Continues as the FLEET row minted alongside this closure.

RULED BY AGA 4 Aug: BINDING. The 48-hour $97 window is real. Consequences: (1) E3 and E4 may state the window closes, because it does; (2) the engineering half re-opens for Nic — once an email has had a window, never mint a second; enforce at CHARGE time not just the UI; acceptance = an expired address sees $157 and Stripe charges $157; (3) today's probe found the opposite is live — TMA_QC_OFFER_EXPIRED returned FALSE on every address tested including a 90-hour-old abandoner, and lapsed cache entries re-mint a fresh 48h. So the copy must NOT claim a binding deadline until Nic's enforcement ships, or we assert something the code does not do.

TAKEN BACK BY THE FLEET 4 Aug under the DEFAULT-EXECUTOR RULE — this is code/script work, not an Aga decision, and it should never have been on her list. Re-minted as a FLEET row so it stays tracked. (AGA-49 says so in its own text: 'Fleet work, ~30 min, no decision needed'. AGA-65 is a Python change to tools/email-analytics/collect_metrics.py. AGA-63 is a scripted hygiene pass over the 991 backlog before enter_991_reengagement.py resumes.)

TAKEN BACK BY THE FLEET 4 Aug under the DEFAULT-EXECUTOR RULE — this is code/script work, not an Aga decision, and it should never have been on her list. Re-minted as a FLEET row so it stays tracked. (AGA-49 says so in its own text: 'Fleet work, ~30 min, no decision needed'. AGA-65 is a Python change to tools/email-analytics/collect_metrics.py. AGA-63 is a scripted hygiene pass over the 991 backlog before enter_991_reengagement.py resumes.)

DONE 4 Aug — F1 is now structurally closed, and less was outstanding than Monday's board said. (3) the single-writer lock + in-lock id minting already existed in tools/ops/queue_edit.py (built 3 Aug). (2) the missing piece — a TITLE-HASH GUARD — is now in tools/seo/queue_drift_check.py: it records id→title in .queue_title_hashes.json and FAILS when a row's title drops below 0.5 similarity, i.e. when an id is repurposed. Regression-tested by replaying the actual 31 Jul NIC-22 swap (caught, similarity 0.20) and a legitimate prefix edit (correctly ignored). Wired into BOTH output modes, so the 06:40 daily rail runs it. (1) rows repaired: AGA-40's real WP-Engine-bill outcome restored; AGA-41 REOPENED because its true outcome is unrecorded and a fake closure on a security task is worse than an open one; the two orphaned tasks recovered as AGA-94 (967 relabel) and AGA-95 (typography). 🔴 Still needs Aga: one line on AGA-41.

DONE 4 Aug — the new 8-email activation arc is fully implemented and verified in 992 TMA | Welcome Sequence AFTER APPSIGNUP AB TEST Aug4th (Aga's clean copy, 0 entered). All 8 builder steps (campaigns 4115-4122) verified live: every message carries the new copy, every CTA resolves to /start-training/ with UTMs, and zero App Store or Play Store links remain. Each step runs a real subject A/B — variant 1 CONTROL, variant 2 CHALLENGER, identical HTML body per the Agent 16 law that a challenger tests the subject only. 17 messages total. CORRECTION TO MY OWN EARLIER WARNING: I had told Aga steps 9 and 10 needed deleting. Wrong — I had read the campaign list (10 campaigns carry seriesid 992) instead of the automation's actual send blocks. The builder has exactly 8 send steps; campaigns 4123/4124 exist as orphans outside the flow. Aga caught it. Nothing needed deleting. SEPARATELY, on Aga's instruction, 988 TMA | Welcome Sequence AFTER APPSIGNUP had a CTA-ONLY edit applied to all 23 messages: dual store buttons replaced with a single gold #FFB000 'START A FREE TRIAL' block button to /start-training/, CTA headline and trust line matched to it, in-body store links repointed, and five anchor texts that would have lied on a trial page relabelled. Subjects, body copy and imagery untouched. Both automations backed up before any write and restorable by PUT. REMAINING (Aga's call, not blocking): arm 992 by moving the app-signup entry trigger onto it, and leave 988 paused rather than deleted so its per-slot history stays as the comparison baseline.

CLOSED 3 Aug 2026 — done in code, no wp-admin needed. My original instruction ('restore the trashed Divi Library item') was WRONG and cost Aga a look: nothing was trashed, her library shows 206 items all Published with no Trash link. The real cause, read off the GLOBAL column in her own screenshot: the three 'Ready to level up' layouts carry NO globe icon, so none is a GLOBAL item and nothing answers to id 207925. Fix was to detach the dead reference so Divi renders the inline copy it already stores. Shipped with the CTA rebuild in one pass: 162 pages live-verified, 0 failures, 60/60 sampled confirm painted + tagged + legacy gone. 18 DRAFT-class pages (61 clicks/28d) left bare by the guard — see AGA-61.

Aga reviewed the ★ paywall-doorways card on the Nic brief (opened live 6 Aug, all 10 elements verified against the deployed page) and SENT THE TASK TO NIC the same day — her decision-and-dispatch half is done. The implementation now lives on tma-nic-attribution-brief.netlify.app/#paywall-doorways with its own acceptance test (GA4 variant landings →~0, or the variant bundles carry the $97 card, and the two ACTIVE Meta campaigns' ad destination URLs read the new target). Verification of the fix happens against that card when Nic ships — this closure is of Aga's action, not the repoint itself.

DONE 3 Aug 2026, Aga said GO. Ran tools/retention/ac_utm_tag.py --apply on 758 Nurture Sequence 2 (181 links / 34 messages) and on the 12 stragglers in 991 NEW Re-Engagement - For people who stopped opening Before Sunset. VERIFIED against live AC, not the script's own claim: (a) raw GET /api/3/messages/{11555,11565} shows 4-6 html + 4-5 text UTM'd links each; (b) re-run of tools/finance/email_utm_audit.py at 10:48 reports 1101/1101 links tagged (100.0%), 0/21 automations carrying zero UTMs. Estate is fully tagged. CORRECTION for the record: 758's live send surface is 11 wired emails (messages 11555-11565, campaigns 3152-3172), not 34 - campaigns 3129-3151 have 0 sends and are wired to no step. The audit's 34/181 counted 23 never-sent orphans; they were tagged anyway (harmless, idempotent, pre-instrumented if ever wired in).

VERIFIED 3 Aug 2026: Aga deleted the Redirection rule (/calisthenics-assessment/ -> quiz root); after WPE cache expiry a utm-only request now reaches the page (200) and receives the tag-preserving JS redirect (Snippet 9, with noscript fallback). Guard re-run: 'bash tools/analytics/check_link_integrity.sh' -> 19 preserved · 0 dropping, exit 0.

🔴 CORRECTION FIRST: I told Aga the five 25-Jul capsule drafts had been PURGED. THAT WAS WRONG — my probe hit HTTP 403 (missing User-Agent, which WP Engine rejects) and I read it as not-found. All five drafts EXIST and all five carry their capsule. What was true, and independently verified with curl: none of it ever reached a live page. So it was eight days of finished work sitting idle, not lost work. | RULED 2 Aug — Aga chose a SPLIT: auto-apply blog posts, workout pages and progression/exercise guides; homepage and pricing pages become a DRAFT she approves. Encoded as tools/seo/page_class.py (money patterns beat content patterns; unknown defaults to DRAFT; self-test 19/19). Auto-applied pages are still backed up, live-verified, and reported with the exact URL. 🔴 Also ruled: DO NOT TOUCH THE HOMEPAGE in this pass.

DONE 31 Jul 2026 (Aga: 'can you fix it?' — took option (a)).

DONE 31 Jul 2026 — verified at the time: all 8 send-step labels in 967 re-read from the AC API after the write, 8/8 matched their campaigns. Reusable tool: tools/retention/ac_relabel_automation_steps.py (dry-run by default).

DONE — live-verified 4 Aug 08:0x MDT via AC API: 990 Monthly→Annual Upgrade Engine (Jul 2026) status=1 ACTIVE (armed 3 Aug 13:05 CDT), 18 contacts entered 4 Aug (all status 1), E1 campaign 4073 '990 · Monthly→Annual Upgrade Engine — E1 (recognition)' sent 15 / 2 opens / 0 clicks / 0 unsub. E2-E4 (4074-4076) staged, 0 sends, awaiting their wait steps. The stranding trap NIC-32 warned of was cleared: all 22 'upgrade-eligible' tags carry a FRESH 4 Aug 06:37-06:38 tag-add date, and 18 of the 22 fired into the automation. OPEN THREAD (not blocking): 4 of the 22 (203551, 260256, 368290, 369288 — all active subscribers, none bounced) were tagged in the same 06:37-06:38 batch but never entered; reason unverified.

ANSWERED 4 Aug WITHOUT running the panel — our own GA4 held the answer. 1,211 AI-engine sessions over 180d produced ZERO attributed leads, against ~12 expected at the site rate (p≈1/180,000, so not noise). Cause is measurement, not behaviour: (1) every internal quiz CTA carries utm_source=tma, and GA4 restarts a session on campaign change, so an AI visitor's origin is erased the moment they click — cross-domain IS configured (1 self-referral in 90d/108k sessions), so those UTMs buy nothing and only destroy attribution; (2) (direct) holds 446 leads/90d and AI clients strip referrers. AI visitors engage WELL (homepage 169s, /progressing-with-bodyweight/ 541s, and they reach /pricing/). VERDICT: do not fund more listicle pitching on the AI-citation promise — not because it is disproven, but because we could not tell if it worked. Fix the internal UTMs, re-read after 28 clean days, then decide. Full write-up: OUTREACH_MASTER §7a.

TAKEN BACK BY THE FLEET 4 Aug under the DEFAULT-EXECUTOR RULE — this is code/script work, not an Aga decision, and it should never have been on her list. Re-minted as a FLEET row so it stays tracked. (AGA-49 says so in its own text: 'Fleet work, ~30 min, no decision needed'. AGA-65 is a Python change to tools/email-analytics/collect_metrics.py. AGA-63 is a scripted hygiene pass over the 991 backlog before enter_991_reengagement.py resumes.)

DONE 31 Jul — Aga activated Smash Balloon; I verified v6.11.4 ACTIVE via /wp-json/wp/v2/plugins (not assumed), then restored the feed in a proper wp:SHORTCODE block with the button kept underneath. LIVE-VERIFIED: Smash Balloon markup renders, zero raw shortcode text, button intact. The original bug was the shortcode sitting in a Custom HTML block, which emits its contents verbatim instead of executing them.

DONE 31 Jul — Aga: 'this has already been done.' The dunning/autosend double-send risk (job audit finding O1) is handled. Off the risk register.

🔴 CLOSED 31 Jul — AGA RULED: 'this is not retired, leave it alone.' The checkout on checkout.themovementathlete.com is LIVE AND INTENTIONAL. The finding was raised by a parallel session on the assumption that a 4th-of-July build still serving in late July must be a leftover — it isn't. **DO NOT RE-PROPOSE** taking it down, redirecting it, or 'cleaning up' that subdomain. Same discipline as the P4 off-brand prune: the argument was heard and answered. 🔴 Anyone auditing checkout URLs, dead promos or DNS: this one is EXCLUDED BY DECISION, not oversight.

DONE 31 Jul — Aga revoked the old token at myaccount.google.com/permissions. Full sequence completed in the safe order: (1) re-authorised, new token at ~/.config/tma/gmail_hello_token.json chmod 600, outside the repo and not in git; (2) I verified read-only that it authenticates as hello@ and that all four lanes — R1 replies, payment recovery, cockpit email, reply-fulfilment — resolve through the safe path; (3) she revoked. The committed copy is now a dead credential. 🔴 STILL TO DO (Claude, low): delete the dead tools/Gmail Connection/token.json from the working tree. Git history is deliberately NOT rewritten — it was never pushed to GitHub and the token is revoked, so history rewriting would be risk without benefit.

DONE 31 Jul — Aga ran the OAuth flow. Verified independently: authenticates as hi@themovementathlete.com, scopes gmail.send + gmail.readonly only (cannot modify or delete), token at ~/.config/tma/gmail_outreach_token.json chmod 600 in a 700 dir, outside the repo. No password was ever shared. listicle_outreach.py now resolves its sender.

DONE 31 Jul — Aga sent the deletion job to Nic. Remaining follow-up rides with the migration: check Rocket.net's answers to the two questions when they reply.

DONE — Aga attested 4 Aug: 'he is updating the PHP and has access to WP Engine.' The GRANT half of this task is complete and the dev is actively working. The REVOKE half is deliberately NOT closed here — split to its own row so it survives this one closing, because a revoke that lives inside a completed task is a revoke nobody performs.

DONE 31 Jul — DECISION MADE: migrate to Rocket.net Business billed annually (~$830/yr + ~$130 overage), saving ~£2,500/yr ex VAT vs WP Engine's £3,856/yr. Migration ticket SENT 31 Jul; execution tracked on AGA-42. 🔴 RESOLUTION RESTORED 4 Aug — this row previously carried the resolution of a DIFFERENT task (renaming 8 send-step labels in 967 TMA | Welcome Sequence After LEAD MAGNET), an id-collision artefact found by TRUST_AUDIT_2026-07-31 F1b. The 967 rename now has its own closed row.

DONE 31 Jul — Aga raised the Jira ticket with the RN team. Awaiting one figure back: what share of installs ever reach workout 3 (the gate the existing review prompt sits behind). Ticket source: marketing/seo/tickets/TICKET_rn_in_app_review_prompt.md

DONE 31 Jul — Aga: 'I already did this.' The WordPress ticket is with the external dev.

DECIDED + MY OBJECTION WITHDRAWN 31 Jul. Aga ruled: wire it — the form enters list 212 'TMA [LEAD] - from LEAD MAGNETS' and triggers 967 TMA | Welcome Sequence After LEAD MAGNET. Verified live: 212 genuinely IS one of the 11 triggers, so it works exactly as she said. I then raised three objections about the sequence's content and ALL THREE WERE WRONG — I read the automation's STALE STEP LABELS instead of the live campaigns. The real emails are '967 LEADS WELCOME — E1..E8', a clean generic lead-welcome arc: E1 'It was never you that failed' never mentions the assessment, and there is NO 50%-OFF or 48h deadline anywhere in the live copy (grep with context returned zero hits). Aga caught it by simply reading the email. No email-fleet work needed; build the wiring as instructed.

APPLIED LIVE 31 Jul on Aga's approval. /best-calisthenics-workout/ (2337). LIVE-VERIFIED after apply: trophy emoji 11 -> 0 · fabricated testimonial screenshots 2 -> 0 · proof-wall compilation present · 11 tinted benefit+pull-quote cards rendering · gap above the title 214px -> 51px · body text 20px = exactly 75 characters per line (inside the 50-75 standard) · zero stale draft ids in the shipped CSS. Backup: apply_backups/bwa-pages-2337-20260731T114210.html (one-command revert). Draft trashed after apply.

SUPERSEDED 31 Jul — VERIFIED LIVE, all five already shipped. The live page carries: 1) 'On this page' TOC 2) the byline 'Written by the TMA coaching team - Reviewed by Jesse, Head Coach, The Movement Athlete - Updated July 2026' — the E-E-A-T gap is CLOSED and Jesse is correctly Head Coach 3) '22-min read' 4) a KEY TAKEAWAYS box above the fold 5) the one table already has an overflow-x wrapper. Same error as AGA-34 — queued from a stale plan without a live check.

SUPERSEDED 31 Jul — VERIFIED LIVE, no decision needed. The geometry fix is ALREADY DONE. Captured /best-calisthenics-workout/ at 1440px and measured every content block: 153 of ~170 sit on the SAME 340px left edge, i.e. one centred column. The plan's '5 competing axes / paragraphs at 127px / card at 24px / 180 vs 246' describes the PRE-v3 page — v3 shipped 22 Jul, the same day the plan was written, and delivered it. My error: I queued this off a 22 Jul doc without re-checking the live page first.

✅ LIVE — verified 31 Jul on the published page (cache-busted): the raw `[instagram-feed …]` shortcode no longer prints (0 occurrences), H1 count is 1 (was 12), and 17 UTM-tagged magnet links are live. Aga chose the on-brand Instagram BUTTON over restoring the photo feed. Follow-up queued separately: check whether Smash Balloon is ACTIVE if the photo grid is ever wanted back.

VERIFIED FIXED 3 Aug. Aga deleted/re-added the stuck field; the served leadbox config now reads data-url="utm_content" with the comma gone (direct cache-busted fetch of startup2013.lpages.co/serve-leadbox/Zs7oB9HvcwisqTMHWHzCyH/, comma count 0), and the full checker run returns 9/10 with ring-dips capturing all three tags (utm_source=youtube, utm_campaign=probe_campaign, utm_content=probe_slug). The 10th, get-started-with-calisthenics, is NOT a Leadpages page and was split out to AGA-69 — it will always read as 'no leadbox trigger' by design, since that page routes via its hero CTA to the quiz root instead. All 9 Leadpages magnets are wired. Six-day drift closed.

FOLLOWS AGA-97's ruling 4 Aug: 'do not message partners.' The 2 remaining listicle web forms (artofhealthyliving.com, calisthenics.com) are outbound partner/publisher contact and are therefore held too. Do not re-raise without Aga opening the subject.

FIXED BY AGA'S ADMIN GRANT — verified live 4 Aug, minutes after she granted it. She ticked Admin (all permissions) at ACCOUNT level on tma-aso-reader@pc-api-8587882451037229353-904.iam.gserviceaccount.com and asked me to look it up. Re-tested with the devstorage.read_only scope: gs://pubsite_prod_rev_07845367963248148244 now returns HTTP 200 on both the bucket metadata AND an object listing (stats/installs/installs_com.themovementathlete.app_*.csv present from 2022 onward). The 403 was a genuine permission gap, not propagation delay — the earlier read-only bulk-reports tick had not taken effect at account level. ANDROID INSTALL + ACQUISITION DATA IS NO LONGER DARK. Fleet now pulls it; no further Aga step. 🔴 SECURITY FOLLOW-UP (fleet, not Aga): the grant is ADMIN — full control including releases, financial data, store presence and user management — where read-only bulk reports would do. Once the first successful pull is banked, propose stepping it back down to 'View app information and download bulk reports (read-only)' + 'View app quality information (read-only)'.

DONE 30 Jul — Aga created the Admin key 8H2C28GLVY. Verified: analyticsReportRequests is now ALLOWED (was 403 on the old key). ⚠️ salesReports is still refused — that needs the Finance/Sales role, and we don't need it for install-conversion, so not chasing it. 🔴 SECURITY: the .p8 downloaded into data/ INSIDE the repo (the AGA-23 trap). Checked git first — NOT tracked, gitignored via 'AuthKey_*', never in history, so no leak. Moved to ~/.config/tma/AuthKey_8H2C28GLVY.p8, chmod 600. .env.local now points at the new key; the old 7364PZ9TSW file is kept, not deleted. THEN: created both analytics report requests (ONE_TIME_SNAPSHOT f85445c0 for history + ONGOING 46367fb6 for daily) — there were ZERO before, so Apple was never going to generate anything. Puller built: tools/aso/asc_analytics_pull.py.

DONE 31 Jul — Aga filled the 1.70.72 listing fields. 🔴 They do NOT take effect until the next app version is submitted, so the DEAD SUPPORT URL is still live on the store right now. That is the item to watch: a 404 support URL is a known Apple rejection reason, so it must ship with the next build rather than wait for one after it.

VERIFIED FIXED 29 Jul after Aga's purge. Bare paths, x-cache: MISS (fresh from origin): /calisthenics-the-lost-art-of-building-strength/ = 200, 0 hops, canonical self-referencing, 257KB rendered. /...-strenght/ = 301 -> the correct slug, 1 hop, x-redirect-by: redirection (rule 20). The 51-hop loop is dead and the 3,821 hits now land on a real page with a correctly-spelled canonical URL. NOTE: the page renders 0 H1s -- that is the known sitewide theme behaviour (entry-header hidden), pre-existing, not caused by this fix; queued as its own estate-wide audit.

DONE 31 Jul — Aga confirmed with a screenshot: Brave → Settings → Downloads → Location = /Users/foxyloxy/Downloads. 'Ask where to save each file' is also ON, which is a stronger guard than the default. The repo-as-download-folder trap is closed.

RULED BY AGA 29 Jul: '159 is on iOS in the App Store because they don't accept prices like 157, then 157 is on our website.' => the pricing law is WEB-SCOPED. Web/Stripe $157/yr (~$3.02/wk); app stores $159.99/yr (~$3.08/wk) because Apple/Google sell on fixed price points. Both correct on their own surface. Never quote $157 on a store listing, never $159.99 in web content, never present the two as a discrepancy to a customer. Recorded in master §13.2 + §13.4 A10.

Folded into AGA-25 — version 1.70.72 is editable, so keywords + promo text + description + URLs are one sitting, not separate tasks.

Folded into AGA-25 — version 1.70.72 is editable, so keywords + promo text + description + URLs are one sitting, not separate tasks.

FIXED BY CLAUDE 29 Jul — and 🔴 MY EARLIER DIAGNOSIS WAS WRONG. Aga purged repeatedly; a re-probe returned x-cache: MISS WITH a 301, which is a fresh origin response, proving the cache was never the problem and her purges had worked. The loop was live in the rules. Resolution, all via REST: (1) Redirection rule 19 (correct-spelling -> misspelled) DISABLED — note the plain POST to /redirect/19 silently did nothing; the working call is POST /redirection/v1/bulk/redirect/disable with {'items':'19'}; (2) page 1731's slug renamed to the CORRECT spelling calisthenics-the-lost-art-of-building-strength (content byte-length asserted unchanged, 32,493 chars; full pre-change record backed up); (3) NEW Redirection rule 20 created, misspelled -> correct, 301, flag_query=pass, so the 3,821 hits keep landing on a real page independently of SEO Redirection, which I cannot read. Origin verified: correct spelling = 200, and the old URL now has a rule I control. This is the durable structural fix the master doc recommended (rename the slug, let the old URL 301 to it) rather than another purge. ONE cache purge is still needed for the edge to stop serving the old 301s — this time it genuinely is cache, because the origin changed a minute ago.

ALREADY DONE — verified live 5 Aug. bodyweighttrainingarena.com homepage now serves 24 scripts of type 'rocketlazyloadscript', i.e. Delay JavaScript execution is ON (the row's premise was that it had gone 25 -> 0, meaning the whole feature was off). WP Rocket confirmed present. No Aga step remains. Not re-verified: whether GTM/Google Analytics are specifically in the one-click exclusions — but the measurement risk that mattered was the feature being off entirely, and it is not.

DONE 30 Jul — Aga unstarred `Lead`; screenshot-verified: `Lead` gone from the Key-events tab (that tab lists ONLY key events, so disappearing = success), `generate_lead` still starred on the Quiz stream. Nothing further.

DONE 29 Jul. ROOT CAUSE: WP Rocket was delaying GTM on BWA via type='rocketlazyloadscript', so any visitor who did not interact never fired the tag ⇒ BWA UNDER-REPORTED in GA4 (1.79M impr/28d). Aga excluded GTM+GA via One-click exclusions → Analytics & Trackers. Live-verified x3: GTM loads as a normal script. 🔴 CONSEQUENCE: every historic BWA-vs-TMA traffic comparison is suspect, and BWA's '~0 tracked leads' reading is partly a measurement artefact.

DONE 29 Jul — Redirection rule 63 repointed /quiz/ to the quiz ROOT.

Scope re-ruled by Aga 27 Jul: do NOT touch pricing/offer pages (incl. /start-training/, /pricing/, /30trial/) or dead promos. Of the 30 FAQ-carrying pages, 17 dead promo + 8 pricing/offer + 1 gear were excluded; the 4 free-resource pages were done. /start-training/ therefore stays without schema by decision, not oversight.

DONE 29 Jul. BWA: Aga trashed ids 6+8. TMA: Claude trashed ids 6,7,8 via REST DELETE — NOTE the endpoint returns HTTP 500 but the trash operation SUCCEEDS; do not read that 500 as failure. Yoast REST bridge (id 5 both sites) verified still intact after.

CLOSED 31 Jul. 30 Jul: 22 promo pages drafted, 16 301'd one-hop on Aga's live rulings (10 -> /start-training/ incl. all black-friday*; 4 -> /50promo/; /spring-sale/ + /smpromo50/ kept pre-existing high-hit rules). 31 Jul: Aga confirmed /70off/ + /summer70/ dead -> already 301 -> /start-training/ (rules 68/69, re-verified); approved the final 6 -> Redirection rules 82-87 created (/pay-pal-alternative-promo/, /ebook-sales/, /test-2/, /start-training-new/ -> /start-training/; /blog-new/ -> /blog/; /freemonth-new/ -> /freemonth/), all origin-verified one-hop 301 (plain paths serve cached 404 until edge TTL expires - AGA-15 pattern). End-state: 24 pages -> 2 untouched (KEEP rulings), 22 drafted + redirected. Audit in tools/seo/wp_changelog.jsonl.

RULED BY AGA 4 Aug: YES — ship the two invisible ones (BWA push-up-progressions + calisthenics-workouts, structured data only). Her follow-up question is logged as a NEW fleet row: she asks whether the FAQ is 'hidden and not showing on the actual page' — i.e. whether invisible-to-the-eye FAQ schema is the right call at all, or whether these should render visibly for readers as well as for Google.

Replaced by AGA-11, which states the two groups and the exact action for each.

CLOSED 28 Jul — already correct, no action was needed. Live check of Admin → Data display → Events → Key events (15 events): `session_start` is NOT a key event and is not even in the list, so there was nothing to unmark; `generate_lead` is ALREADY starred (stream: Quiz). `landed_on_pricing_page` does not exist as an event — it has never fired, so it cannot be marked. 🔴 The '8,918 session_start conversions' figure in master §2 (20 Jul) does NOT describe this property today — it was either fixed since or wrong then, and it was handed to Aga as a live task without re-verification. Treat §2's GA4 key-event claim as STALE.

DONE 29 Jul — IndexNow v1.0.4 ACTIVE on BOTH sites, 'Automate URL submission: Enabled'. 0 submissions in first 48h is correct (only fires on content change).

DONE 29 Jul by Aga, live-verified via rendered Organization schema on both homepages: name 'The Movement Athlete' both sites (BWA was 'Calisthenics Academy'), TMA logo both, sameAs 5 on TMA / all 7 on BWA.

DONE 29 Jul — BOTH credentials landed and verified live. Apple: Issuer ID 69a6de96-… written to .env.local; ASC API authenticates and returned the app + every subscription price. Play: read-only service account at ~/.config/tma/play-aso-reader.json, Reporting API 200. Immediately settled two things public data could not: the iOS keyword field (100/100 chars of aesthetics/diet terms — see master §13) and the real store price ($159.99/yr, 15% Apple commission). Tooling: tools/aso/asc_client.py + asc_listing_pull.py. STILL PENDING (no action from Aga): install-conversion needs Apple's async reports + Play's bucket ACL propagation (≤24h).

DONE 29 Jul — Aga pasted the 3 CSS edits; live-verified all three render. 🔴 REAL BUG FOUND + FIXED: the nav CTA rules were trapped inside @media (min-width:1000px), so the primary CTA was completely unstyled on every phone. Now top-level = desktop AND mobile. Also killed off-brand #00b6d4 (0 remaining). Claude parsed the CSS twice pre-publish and caught 2 separate brace errors.

Affiliate lane excluded from the 29 Jul run by Aga; and AGA-01 closed as KEEP, so the Amazon check no longer gates anything.

APPLIED LIVE 29 Jul — Aga ruled STRIP. 20 count-asserted replacements across warm-up (216170, 12) + planche-prog (214521, 8). Live-verified: 0 fabricated strings remain, 17 replacements rendering. Backups tools/seo/apply_backups/tma-posts-{216170,214521}-20260729T095503-PRE-HONESTY.html. RESIDUAL flagged to Aga, not yet actioned: warm-up still says 'After analyzing thousands of workout sessions' / 'Based on thousands of training sessions' (same fabricated-methodology family, outside the approved list).

CLOSED 29 Jul — Aga ruled KEEP (no prune). Affiliate lane excluded from this run; both the SEO and affiliate analyses had independently reached the same reject list, but Aga's call stands: the 8 off-brand equipment posts stay indexed.