THE EMAIL OFFER LAYER — FULL FORENSIC RECORD

Verified against, not from

Executed

• Message 11515 fixed (750 TMA | Welcome Sequence After Quiz E6). The trial promise lived in the hidden preheader div and the separate preheader_text field. Replaced with "50% off, 60-day guarantee — and an honest reason to start now", matching the fleet's own 5 Aug substitution. Full backup, whole object echoed back, read back from the API to verify.
• Estate-wide sweep: all 32 automations, all three surfaces, restricted to emails that link the charge-immediately offer — zero trial claims remain.
• email_lint.py now reads the preheader (_with_preheader() folds it in as a hidden div so every existing rule sees it, without rewriting any rule). Proven to block a preheader-only claim and pass a clean one. ac_email_read.py's --grep now searches all three surfaces.

Attempted and cleanly reverted

Wiring campaign 3150 into 758 Nurture Sequence 2. The API creates steps but ignores ordernum — position is set by parent, and every block in that arc is parented to the START block, so the step lands at position 2 or returns HTTP 500. One attempt sat at position 2 for ~60 seconds; it was caught on read-back, deleted, and confirmed sent to nobody on the campaign's own counters (send_amt=0, sdate=None) — not on the delete returning 200. Both 758 Nurture Sequence 2 and the probe automation read back as exact matches to their backups.

Structural findings

• 🟢 Every welcome and nurture arc already ends in an "enter another automation" step, all pointing at 751 Nurture Sequence 1. Aga's ruling is therefore a repoint, not a build.
• 759 Nurture Sequence 3's zero entrants is not a severed chain — 758 Nurture Sequence 2's last step has 0 sends, so nobody has reached the hand-off yet.
• send_collision_check.py with the offer declared returns PASS — an evergreen 50%-off drip and a 50%-off campaign are "the SAME offer family — one conversation, not a collision".

Left open

1. One builder click — campaign 3150 into 758 Nurture Sequence 2 after email 7 (NIC-87).
2. 756 Part 1 - Abandoned Cart Reminder must not be unpaused first — its missing unsubscribe link is still live and the compliant rebuild is parked, not applied (AGA-115).
3. The published copy of this document is still the 5 Aug version — it needs estate_sync.py to catch up.

Claimed, 5 Aug

Stop trial copy on charge-immediately links · 975 Onboarding | After The Purchase's leaked scaffolding · 756 Part 1 - Abandoned Cart Reminder's missing unsubscribe · 987 LIFETIEM SALE | EXITPOPUP- Tag Added:lifetime_checkout_exit_new's $297→$497 mismatch. Effort: Low.

Verdict

Executed on the evening of 5 Aug — 79 verified API message writes — and most of it holds. Two defects survived, and one of them is on the single highest-volume offer email we own.

What I opened

What it returned — defect by defect

🔴 The defect that survived, and why

Message 11515 — 750 TMA | Welcome Sequence After Quiz email 6, 15,495 sends, the biggest offer email in the estate — still reads "50% off, nothing charged till day 8". Its only offer link is themovementathlete.com/50promo?utm_campaign=750-tma-welcome-sequence-….

I fetched all three checkouts that page sells. Every one returns HTTP 200, title "Register and pay", a live pk_live_… key, and zero occurrences of "trial", "free trial", "nothing charged" or "day 8".

Why it survived: the 5 Aug sweep verified charged until day 8 = 0 and was satisfied. The copy says till, not until. The message's mdate is 2025-10-26 — it was never rewritten. A verification grep that does not match the copy reads exactly like a clean pass.

Do this now

One single-slot copy fix through /email-fleet on message 11515 — either drop the day-8 promise, or point the email at a checkout that genuinely carries a trial. Then re-grep on the word actually used, not the word expected.

Claimed, 5 Aug

Biggest single cash lever and the clean falsifier — read the $78.50 count for 72h and causation is settled either way. Effort: Low.

Verdict

Premise confirmed and sharpened. Aga ruled on 6 Aug to build it for her approval; it is filed as AGA-111. One date in the original is wrong, and there is a caveat on "biggest lever" that she should see before she presses send.

What I opened

What it returned

• Last 50%-off broadcast: 31 May 2026 — TMA Memorial Sale 2026 - E10 - Sunday - LAst EMAil, 16,593 recipients. ✅ Exactly as stated.
• 🔴 Last broadcast of ANY kind: 12 July 2026, not 20 July. The 20 Jul send was a drip inside 973 Nurture Sequence 1 NEW COPY FROM CLAUDE, per this document's own §3.8 correction. That is 25 days of complete silence as of today.
• The audience is there. The 2 Jul launch reached 44,747; the engaged-leads tier held 12,023–13,460 right through 12 Jul. Core (opened/clicked ≤90d) ≈ 5,090.
• send_collision_check.py was run 6 Aug: within caps for a Core broadcast.

🔴 The caveat on "biggest single cash lever"

Our most recent launch converted badly. Across 2–12 Jul, the lead-side sends total 274,788 for 594 clicks — a 0.216% CTR — and 1,195 unsubscribes (0.43%). The Memorial arc's closing days ran near 0.9% unsub.

That does not kill the action — it prices it. This is still the cheapest falsifier we have, and 25 days of silence is its own risk. But expect a modest click volume, and read the $78.50 count rather than the open rate.

Do this now

Aga's approval on AGA-111 is the only blocker. Send to Core (~5,090), not the full 44,747 — the collision check already cleared that shape. Then read $78.50 for 72 hours.

Claimed, 5 Aug

The complete evergreen 50%-off launch arc already exists and has 29 lifetime entrants. We do not need to build an evergreen launch — we need to point people at the one we own. Cheapest cash in this document. Effort: Low.

Verdict

Fully confirmed, every element. Aga ruled on 6 Aug: route 760 School of Strength (NEW) as the bridge AND commission a new Jeff-Walker-PLF arc alongside it. Filed as AGA-113.

What I opened

What it returned

🔴 One correction to Part 6, in this action's favour

Part 6 lists "entry triggers cannot be read; builder view only" as [UNVERIFIED]. That is half wrong: the automations list endpoint returns a meta.starts array giving the trigger TYPE per series. 760 School of Strength (NEW) shows submit + subscribe.

Reading is possible. Writing still is not — so repointing the trigger remains Aga's click in the builder. That is the whole of the work.

Do this now

Aga repoints one entry trigger. Before it goes live, fix the Marcus persona collision found 6 Aug — the same member story opens 750 TMA | Welcome Sequence After Quiz E6 and 760 School of Strength (NEW) E1.

Claimed, 5 Aug

Put a real HTML offer into 758 Nurture Sequence 2 BEFORE migrating anyone else into it — as built, the migration moves 14,619 people from a weak offer to none. Effort: Low.

Verdict

🔴 The window this action asked for has closed. The migration ran to completion while the recommendation sat unactioned — so this is no longer a precaution, it is the largest live exposure on the board.

What I opened

What it returned

• The migration is done. added_to_758 = 10,252 · removed_from_973 = 11,014 · first row 29 Jul, last row 5 Aug 15:40. It was Aga-approved on 29 Jul as a nurture consolidation.
• 758 Nurture Sequence 2 now holds 14,620 records, of which 14,609 are ACTIVE. It is where essentially the whole nurture population now lives.
• It still has zero offer. 11 emails, 0 offer hrefs, 18 /start-training/ hrefs — exactly as reported. The fleet's own instrument grades it TRIAL-DOOR ONLY.

🟢 And it is cheaper to fix than this document assumed

The offer email already exists. Campaign 3150, "The training built for you — now half price", is finished HTML carrying 2 real /50promo/ hrefs already tagged utm_campaign=758-nurture-seq, and its automation field still reads 758.

It was written for this sequence and simply is not wired as a send step. §3.11 filed it as an orphan; live, it is an orphan of this exact automation.

So this is a wiring job, not a writing job — one step added to a live arc, which is single-slot and EVOLUTION-LAW compliant.

🔧 ATTEMPTED IN CODE, 6 Aug 2026 — and it needs one builder click

Aga authorised execution. I tried the API properly and it cannot place the step. Nothing was sent and nothing was left changed — 758 Nurture Sequence 2 reads back as an exact match to its pre-attempt backup (24 blocks), and campaign 3150 still shows send_amt = 0, sdate = None.

What was established, in order:

• Block creation IS supported — POST /api/3/automationBlocks works; proved on 757 Part 2 - Abandoned Cart Reminder (paused, 0 contacts) and cleaned up to an exact restore.
• 🔴 ordernum in the payload is ignored. Position is set by parent — the new block lands immediately after whichever block is named. Proved by passing a deliberately bogus ordernum=99 and watching it land in the parent's slot.
• 🔴 Every block in 758 Nurture Sequence 2 is parented to the START block — a flat structure. So naming a parent cannot express "after email 7": the API drops the step at position 2, and naming the email-7 block instead returns HTTP 500.
• The first attempt did land a step at position 2 for about a minute. It was caught by the read-back, deleted, and sent to nobody — verified on the campaign's own counters, not on the delete succeeding.

✅ The collision question is settled, though

send_collision_check.py was run against Aga's promo week with the offer declared: VERDICT PASS. An evergreen 50%-off drip inside this sequence and a 50%-off campaign are "the SAME offer family — one conversation, not a collision". So wiring this does not damage the promo, and the two do not need to be sequenced apart.

Do this now — one click, in the builder

In 758 Nurture Sequence 2, drag a "Send an email" step in immediately after email 7 ("Ross owns a gym. His training was killing him."), preceded by a 5-day wait to match the arc's rhythm, and point it at the existing campaign 3150 — "The training built for you — now half price". Do not write anything: the email is finished and already tagged utm_campaign=758-nurture-seq.

Why email 7 and not the end: the last step has 0 sends — nobody has reached the end of this arc yet. At email 7, roughly 11,700 of the 14,609 have not passed that point, so they collect the ask as they arrive. Tracked as NIC-87; AGA-112 is already ratified, so nothing gates it.

Claimed, 5 Aug

Add HTML offer hrefs to 988 TMA | Welcome Sequence AFTER APPSIGNUP and 992 TMA | Welcome Sequence AFTER APPSIGNUP AB TEST Aug4th — these replaced the app-download-then-sell funnel and carry no clickable offer. Effort: Low.

Verdict

The measurements are exact. But the action's scope is too narrow, and it prescribes the opposite fix to the queue row now standing against it. Aga has to pick one.

What I opened

What it returned

The href counts reproduce exactly — 42 and 45. Both arcs carry the 50%-off wording in plaintext only (19 and 15 messages respectively), which almost nobody renders.

🔴 The action misses the biggest instance. 766 Facebook Lead Form | LM: Building Muscle After 40 | has the identical defect with 913 entrants — more than 988 and 992 combined.

🔴 The conflict Aga must resolve

This action says add an offer to the HTML. Queue row NIC-88 says the opposite: "The HTML is already correct: this is plaintext regeneration, NOT a copy rewrite" — i.e. delete the ghost from the plaintext so both surfaces agree that there is no offer.

Both cannot be right. The 5 Aug pass found this pattern inverted on 967 TMA | Welcome Sequence After LEAD MAGNET: the HTML was the new, deliberately offer-free email and the plaintext was a stale leftover. If that holds here, these arcs were intended to be offer-free — and §3.5's reading, that they replaced the app-download-then-sell funnel and should sell, is the strategic case for the other direction.

✅ AGA RULED, 6 Aug 2026 — and it dissolves the conflict

"Yes — all welcome sequences should end in a launch sequence."

Neither side of the conflict was the right answer, and her ruling supersedes both.

• The welcome arcs stay educational. We do not bolt an offer into the middle of them — so this action, as written, is withdrawn.
• NIC-88 is correct as hygiene and proceeds: regenerate the plaintext so both surfaces agree there is no mid-arc offer. That kills the ghost rather than feeding it.
• The sell moves to the end. Every welcome arc hands off into a launch sequence, which is where the priced offer, the real deadline and the close live.

This applies to all five welcome-shaped arcs, not the two named here — 750 TMA | Welcome Sequence After Quiz, 967 TMA | Welcome Sequence After LEAD MAGNET, 988 TMA | Welcome Sequence AFTER APPSIGNUP, 992 TMA | Welcome Sequence AFTER APPSIGNUP AB TEST Aug4th and 766 Facebook Lead Form | LM: Building Muscle After 40 |.

Do this now

Run NIC-88's plaintext regeneration across all three ghost arcs including 766 — never a mix, which is what produced the ghost in the first place. Then build the hand-off: each welcome arc's last step exits into 760 School of Strength (NEW) or the new PLF arc, per action 2. That is now a structural job, not a copy job.

Claimed, 5 Aug

Move the ask from email 18 of 23 to ~email 7, where 751 Nurture Sequence 1 proved it works. Restores an offer to ~11,000 people getting none. Single-slot, EVOLUTION LAW compliant. Effort: Low.

Verdict

Every diagnostic claim is confirmed. The cash claim is not — those ~11,000 people are no longer in this sequence. Fixing it today would restore an offer to seven people.

What I opened

What it returned — the diagnosis holds exactly

• The ask is at email 18 of 23 (step 35 of 45) ✅ · 0 sends ✅ · reach 0.0% ✅
• Send-decay reproduces to the number: 10,262 → 6,748 → 3,657 → 1,688 → 39 → 21 → 0. The last email that ever sent is number 15. Emails 16–23 have never delivered once.
• 751 Nurture Sequence 1's first ask is at email 7 of 12 ✅, and 1.25% is the best CTR of any offer email in the estate — I ranked all 21 that have ever sent, and it tops them ✅

🔴 Why the cash claim no longer holds

• 973 Nurture Sequence 1 NEW COPY FROM CLAUDE now holds 42 records — 7 active. The migration in action 3 removed 11,014.
• Its entry trigger type reads none. It is taking no new entrants; entered has not moved since 5 Aug.

Two numbers in §3.6 that need restating

"Reached the offer — 14,033, 67% of entrants" mixes denominators. 14,033 ÷ 24,838 entrants = 56.5%; ÷ 21,223 first-email sends = 66.1%. The instrument's reach for 751's first ask is 74.3%. The 67% figure is against sends, not entrants.

"1.25% — best in the estate" is true of offer emails only. 751 Nurture Sequence 1 email 2 clicks at 1.83% carrying no offer.

Do this now

Demote NIC-86 and fold its intent into action 3. The lesson — put the ask at email 7 — is exactly what action 3 should apply to 758 Nurture Sequence 2, where the 14,609 people actually are.

Claimed, 5 Aug

Only two coupon ids exist in 8 years (promo, getstarted50off2018), both undifferentiated. This makes invoice.discount.coupon.id a campaign-level revenue key — cheaper and faster than NIC-77, no dev dependency. Effort: Low.

Verdict

🔴 The headline fact is wrong. There are 1,109 coupons, not two. But the recommendation survives — and it is easier than claimed, because the convention it proposes already exists and is already working.

What I opened

What it returned

• 1,109 coupons exist, oldest 21 Nov 2018, newest today.
• Only 21 have ever been redeemed, across 6,552 redemptions.
• The defensible version of the claim: getstarted50off2018 (3,128) + promo (1,507) = 4,635, or 70.7% of every redemption in 8 years. That is concentration, not scarcity.
• 1,088 coupons have zero redemptions — overwhelmingly auto-minted 30%-off codes with random ids and random 7-character names, one every few days since 2025. That is its own hygiene finding.

🟢 The convention already exists — and converts

Named, purpose-scoped, non-guessable coupon ids are already in production and already attributing. So this action is not inventing a scheme — it is extending a proven one from checkouts to campaigns.

Do this now

Mint tma_bcast_<campaign>_once for the action-1 broadcast and read invoice.discount.coupon.id. It gives that send its own revenue key with no dev dependency — which is the whole point of the original recommendation, intact.

Claimed, 5 Aug

Fix AUTO_PREFIX so the big sequences are visible — otherwise the fleet keeps optimising 2,495-send sequences, blind to 54,826 entrants. Effort: Low.

Verdict

Executed the same day, and more thoroughly than the action asked for.

What I opened

What it returned

• AUTO_PREFIX now sits at line 124, not 107, and is explicitly demoted in the source: "🔴 THIS IS A FALLBACK, NOT THE SOURCE (fixed 5 Aug 2026 — it was the source until today)."
• The real fix is a new instrument: offer_reach_check.py walks the automations' own send blocks, so nothing depends on a campaign name any more.
• It writes campaign_automation_map.json — 231 campaigns mapped on today's run. Previously 231 of 254 campaigns, including 100% of the offer emails, were invisible.
• 973 Nurture Sequence 1 NEW COPY FROM CLAUDE is now visible on the board at 12,141 sends.

Do this now

Nothing. Worth knowing that the EVOLUTION LAW's inputs changed on 5 Aug, so any "worst eligible slot" named before that date was chosen from the blind board.

Claimed, 5 Aug

Never assert an offer without checking the HTML surface. 47 of 236 emails are divergent. This defect class produced a wrong claim in this very document. Effort: Low.

Verdict

The rule is right, is now enforced in code, and I applied it throughout this pass. The specific counts do not reproduce — they are regex- and corpus-dependent, and the corpus has since changed.

What I opened

What it returned

Two honest reasons the numbers move. The corpus shrank when Aga deleted 983 J4 Lifetime· Checkout Abandoner and 987 LIFETIEM SALE | EXITPOPUP- Tag Added:lifetime_checkout_exit_new, and grew as arcs were edited; and "deadline language" has no canonical regex, so the count is a function of the pattern chosen.

🔴 The finding gets worse, not better, when the two classes are separated. The dominant defect is not the missing deadline — it is that 52 emails carry an offer link only in the plaintext, which is the mechanism behind every GHOST OFFER verdict in action 4.

Do this now

Restate AGA-106, which currently asserts the un-reproducible "34". Any count of this class must ship with the regex and the corpus definition beside it, or it will drift again. The rule itself needs no change — it is already a gate in offer_reach_check.py.

Claimed, 5 Aug

Diagnose the −81% broadcast audience collapse — an offer is worthless without an audience. Effort: Medium.

Verdict

🔴 There was no audience collapse. This action was already refuted by the correction appended to §3.8 of this document, and the live send record confirms it independently. It should be struck and replaced.

What I opened

What it returned

• The six-point "−81%" series was six drip sends inside 973 Nurture Sequence 1 NEW COPY FROM CLAUDE, not six broadcasts. It is §3.6's cohort send-decay measured a second time on calendar dates.
• The list did not shrink. The 2 Jul launch reached 44,747. The engaged-leads tier held 12,023–13,460 right through 12 Jul — the last broadcast day.
• The real problem is a famine. Nothing at all has gone out since 12 July 2026 — 25 days.

Do this now

Retire action 9 and read it as action 12. The number is now owned: Core/Mid/Dormant tier sizes belong to the tma-broadcast-cadence-owner seat, weekly. Nothing needs diagnosing — something needs sending.

Claimed, 5 Aug

Wire the orphaned evergreen deadline arc — the launch mechanics are written; they are simply unconnected. Effort: Medium.

Verdict

The three named campaigns are genuine orphans, exactly as described. But the action is now largely redundant, and §3.11's orphan list contains two entries that are not orphans at all.

What I opened

What it returned

§3.11's claim about 3495 holds on the strict surface: its HTML carries "$157" and "$78.50" together with two real /50promo/ hrefs and five deadline phrases. It is the most explicit annual sell we have written, and it has never reached a human. 3499 correctly carries no offer — it is the cart-closed email.

Why it is now redundant

We do not need a second evergreen deadline arc. 760 School of Strength (NEW) is already wired, already carries real deadline mechanics, and already grades OK on reach. Building a second one competes with it for the same traffic.

Do this now

Fold this action into 2 and 3. Use 3150 for action 3 — it is the ready-made 758 Nurture Sequence 2 offer email. Keep 3479/3495/3499 as the copy bank for the new PLF arc Aga commissioned on 6 Aug, rather than wiring them as a rival sequence.

Claimed, 5 Aug

Decide 756 Part 1 - Abandoned Cart Reminder — 0 entrants, finished 50%-off email waiting. Highest-intent audience we have, unserved. Effort: Medium.

Verdict

Confirmed, with one correction and one hard sequencing constraint: turning this on in its current state would ship a compliance breach on the first send.

What I opened

What it returned

• 0 entrants ✅ and the offer is real — 6 /50promo/ hrefs across its two messages ✅
• 🔴 Correction: it is status 2 — PAUSED, not active. Part 4 described 756 and 757 Part 2 - Abandoned Cart Reminder as ACTIVE. They are not, and that is why the missing unsubscribe has not yet cost anything.
• 🔴 The compliance defect is still in the live message. Zero unsubscribe tokens and three raw markdown bolds still rendering — **THE BRUTAL REALITY:**, **HERE'S THE DEAL:**, **THE BOTTOM LINE:**. All three messages carry mdate 2025-10-22: untouched.
• A compliant rebuild exists and passed email_lint.py on its first run — but the ledger records it as "756 rebuild parked". It was never written back to ActiveCampaign.

Do this now

Order matters, and it is not the order the action implies.

1. Apply the parked rebuild to message 11456 first — it is written, linted and waiting.
2. Re-run email_lint.py against the live message, not the file.
3. Only then does Aga's on/off ruling on AGA-115 become safe to act on.

Unpausing before step 1 sends a CAN-SPAM/GDPR breach to the highest-intent audience we have.

Claimed, 5 Aug

Re-establish a standing promo cadence — the one selling motion that demonstrably still works. Effort: Aga's call.

Verdict

Still Aga's call, but it is no longer an open-ended one. The doctrine, the owner and the collision checker were all built on 5–6 Aug, and she has already ruled the pricing posture. What is missing is a calendar.

What I opened

What it returned

• The doctrine exists — how often we broadcast on top of the live automations, to which engagement tier, and how a double-send is prevented. Every number in it was measured, not assumed.
• The owner exists — tma-broadcast-cadence-owner, now a blocking seat before any campaign or promo burst, and the owner of the weekly tier sizes.
• The checker exists and runs — send_collision_check.py was run on 6 Aug against the proposed action-1 broadcast and returned within caps for a Core send.
• Aga already ruled the posture on 5 Aug: new-member window plus calendar sales, with $157 remaining the anchor.
• The evidence for "demonstrably still works" is the promo calendar — but note the honest asymmetry recorded there: the May 50%-off netted ≈+$688, while the 2–12 Jul lifetime at $297 netted ≈+$13,800.

✅ AGA RULED, 6 Aug 2026 — the standing question is answered

"It's 50% off annual, and we will be having evergreen AND calendar events."

So it is not either/or. Both delivery mechanics run, against one price.

• 50% off IS the annual price for a new member — which the data already said: 90.7% of new annual sales landed there and zero ever sold at $157.
• Evergreen — the per-contact launch window every welcome arc now hands off into. This is what 760 School of Strength (NEW) already does, with a real per-contact deadline.
• Calendar — declared, dated promo bursts on top, owned by tma-broadcast-cadence-owner and cleared by the collision checker.
• $157 remains the anchor, per her 5 Aug pricing ruling. The discount is the window, never an ambient price tag.

This is consistent with OFFER_LADDER rung 3, ratified by Aga the same day under AGA-112: "50% off — the new-member window", with exactly two legitimate placements — a real per-contact deadline, or a declared calendar burst. Nothing ambient.

Do this now

The cadence owner turns this into a dated calendar, and the evergreen half is delivered by actions 2, 3 and 4. Nothing is blocked on a decision any more.

Preserved verbatim. The re-verification above does not overwrite the original recommendation — it sits beside it.