Audit: our native doctrine vs Zakaria "Long-Format Native Ad SOP 2.0" — 2026-09-03
Source: https://native-ads.manus.space/ (14 sections, 8,712 words; text saved at knowledge/formats/zakaria-long-format-native-sop-v2-2026-09-03.txt). Our side: native-copy SKILL.md + 16 references, DawnBands brand docs, DR-copywriter formats; line-cited inventory at /tmp/subagent-native-inventory-20260903-1512.md. Fish's read going in: "we don't really have relatability or worldbuilding, introducing our sub-avatar like we did in our Alevia templates."
Verdict in one line
Fish is right, and it is narrower than "we don't have it": the WORLD, the loop stack, dialogue-delivered mechanism and story-disguised objections all exist in our doctrine, but only as Mode B (2,500w+) rules. T2, the only architecture that has ever scaled for us, and every compressed body we shipped this week, carry none of them. That is why the SA2 v1 mechanism was "not stated enough": the T2 budget gives it 100 to 150 words and no dialogue.
Part A — what the SOP has that we are missing or under-specify
| # | SOP element (section) | What we have today (file:line) | Status | What to add, where |
|---|---|---|---|---|
| 1 | THE WORLD (§04 Part 2, lines 4-150, ~20% of the ad): the before, the gradual decline as 3-5 small losses, failed attempts, emotional toll, relationship impact. "Make the reader feel SEEN." | Mode B Beat 2 "Identity contrast" (witness-native.md:190, gated, Mode B only); T1 "human identification 150-300w" (templates.md, untested Jacob delta); T2 = "identification by scene 80-120w" (templates.md §T2) = a scene, not a world; Mode A: nothing. | GAP in T2 / Mode A | T2 gets a WORLD section, 150-250w, three required sub-beats: who she was before the problem (one receipted line), the decline as 3-5 small losses in her words, who notices and what they said. Budget comes out of the failed stack, not the mechanism. |
| 2 | Avatar research protocol (§01): daily experience, internal monologue phrases, failed attempts in order, their language, relationships (who notices, who picks up the slack, who said the thing that stung), the fear they will not say. | Evolve card: desire / experience / emotion / behavior / demographic (research doc §SA2). native-copy "7 avatar detail categories" (SKILL.md) covers failed stack, anticipation, scene after, relationship cost, VOC bank. Missing as card slots: the before, internal-monologue phrases, who-notices, the unspoken fear. | PARTIAL | Add four slots to the sub-avatar card, each VOC-receipted or empty: BEFORE (one quote), MONOLOGUE (3-5 verbatim self-talk lines), WHO NOTICES (one quote), UNSPOKEN FEAR (one quote). These feed the WORLD section; nothing without a quote goes in (9-03 relatability gate). |
| 3 | Relatability as an engine (§00 names it; §11 Mistake 2: "if your details could apply to anyone, they resonate with no one"; §13: "this is MY story"). | Never a named engine. matrix-angle-execution-map.md:19 "relatability" as an ANGLE with a lines-3-8 VOC-specificity test; master-brief.md:128 "Gate 3 recognition". | GAP (no definition) | Define it once in SKILL.md: relatability = two kinds of specificity kept apart. NARRATOR TEXTURE is hers, receipted, as specific as the quote allows (the toothbrush with the paste on it). READER QUALIFIERS are only the card's categories. The SOP's "specific enough to be MY story" and Fish's "no qualifier we can't verify" are the same rule seen from two sides. |
| 4 | Open-loop architecture (§05): stack loops, open the next before closing the current, main loop spans at least half the ad, vary WHAT/WHY/HOW loops, loops at transitions not every line. | Mode B YAML: primary + at least one secondary loop, stop-if-missing (witness-native.md:451-480; mode-b checklist:10). Elsewhere: "cliffhanger" mention (master-brief:137), "open loops at major transitions" (overlays.md:61). No stacking rules, nothing for T2/Mode A. | GAP outside Mode B | Cross-format loop rules in SKILL.md: (a) the hook's loop resolves no earlier than the reveal; (b) one new loop opens inside each section's last two lines; (c) loops open through EVENTS and SPOKEN LINES, never narrator announcements ("what she said next changed everything" is our banned announcer transition; "she wrote one comment under my post" is legal). |
| 5 | Mechanism through dialogue (§06): authority character talks, layered revelation, narrator reactions between layers (3-4 lines of mechanism to 1 line of reaction), narrator asks the reader's questions, "that's why the X did nothing" connect-backs after each layer. §11 Mistake 3: mechanism should be the longest section after the world. | Named stranger speaks the CATEGORY (gated, witness-native.md:276; alevia-live-anatomy). Mechanism itself is exposition everywhere else: T2 budget "100-150w, one delivery style"; 9-03 marksman bodies = 4 lines; SA2 v2 = 12 lines of facts in her voice with no reaction beats. | GAP (the defect Fish called on v1) | T2 mechanism 200-300w, delivered as the named forum mother's comment or DM (forum register, not clinic), in three layers with a reaction line between layers and a connect-back after each ("that is why the calm voice held until nothing moved"). Authority stays inside our gate: forum mother, teacher note, coworker, the adult who was that kid. Never a clinician. |
| 6 | Objection handling disguised as story (§07): narrator asks about each thing she tried, in the order she tried it; each gets acknowledge, gap, one-line summary; reader crosses everything off herself. | Mode B Beat 8 has the pattern (witness-native.md:296-307); overlays.md:51 names the phrase once. Mode A doctrine is the inverse: "objection killer" as a standalone variant that LEADS with skepticism (master-brief:183). Our failed stacks say what happened, never why it only touched part of the mechanism. | PARTIAL, Mode B only | T2 failed stack: each item gets one plain gap line tied to the mechanism ("louder reaches the room, not the filter"; "the calm voice works, and it is still her at the door"). Genericized items, prices stay, no analogy in the summary line. |
| 7 | The close (§04 Part 8): restate the stakes as loss, generational stake, risk reversal, one CTA, final emotional line. | P.S. ladder + two CTAs + guarantee (offer floor); rung 4/5 stakes in the ladder. No loss-aversion restate, no final emotional line as a rule. | PARTIAL | Add two lines to the close block: one loss line (the verified rung-4 stake in the future tense) before CTA #1, and a final emotional line as the last words before the last link. |
| 8 | Inciting incident introduces the authority (§04 Part 3). | Rupture object is gated (Beat 5); the discovery chain follows separately (Beat 6). | PARTIAL | Optional in T2: the rupture night IS the night she reads the thread. We already did this in SA2 v2 by accident; write it as the default order. |
| 9 | The 8 killers (§11) as a pre-publish check. | Six are covered by gates (early reveal, tone switch, overclaiming, awareness mismatch, weak incident, stakes beyond the individual). Not checked: "rushing the mechanism" and "details could apply to anyone" (shallow avatar). | PARTIAL | Add both to the final check: mechanism word share >= world word share × 0.8; every world/rupture line traces to a card quote. |
Part B — where the SOP and our doctrine conflict, and which one wins
| SOP says | We say | Ruling |
|---|---|---|
| Every mechanism needs an analogy (§03, §06); objection summaries are analogies ("closing the blinds in a burning building"). | No similes or analogies anywhere, permanent (Fish 8-25, 8-27 third strike, 9-01 fourth). | OURS. The summary line is a plain fact in her words. |
| Product name at 75-80% (§08). | 55-70% (Alevia median 52%, v10 at 64%, both receipted). | OURS. 70-80% is a legal test band on T1 only. |
| Week-by-week results, weeks 1 to 8, with a relationship-resolution peak (§09). | No outcome timeline without a same-narrator buyer receipt; victory = desired absence (witness-native.md:323-351; master-brief:121). | OURS, with a flag for Fish: fable-v10, the only scaled native, carries a results timeline under authorized dramatization. If a receipted-buyer lane exists (Okendo verbatims), future pacing can reopen there. The SOP's "there you are" relationship peak can be written as a desired absence in the future tense today. |
| The authority is a doctor, specialist, researcher (§04, §08). | No invented clinician who recommends anything (Curvani's fake MD byline got zero distribution; 9-01 validator). | OURS. Authority = named forum mother, teacher note, coworker, the adult who was that kid. |
| Unaware hook: pure story, no hint of a problem (§02). | Unaware lane parked (Fish 8-31); every hook names the reader by symptom in lines 1-2. | OURS. The SOP's problem-aware hook ("the reader self-selects within the first 3 lines") is our rule already. |
| One sentence per line, blank line every 1-3 lines, NO emojis, no caps (§10). | Mode A sentence-per-line; Mode B paragraph prose (Alevia measured, 0/240 one-per-line); compressed bodies close on an emoji checklist. | SPLIT BY FORMAT. Story natives (T1/T2): SOP formatting, no emoji but the 👉 link. Short/static primaries: keep the ✅ checklist (Alevia's longest-running ads are 55-word checklists). The 9-03 marksman bodies crossed the two: story open, checklist close. Stop mixing. |
| Loop-opening phrases: "what she told me next changed everything", "she wasn't finished" (§05). | Banned announcer transitions (said-out-loud test). | OURS. Loops open through events and spoken lines. |
Part C — the patch set (nothing applied yet)
native-structure-templates.mdT2: insert WORLD 150-250w after the hook; mechanism budget 200-300w as dialogue; failed stack gets the gap line; close gets the loss line + final emotional line. Total band moves to 950-1,250w.native-copy/SKILL.md: new section RELATABILITY ENGINE (two specificities) + OPEN-LOOP RULES (cross-format) + MECHANISM-THROUGH-DIALOGUE pacing (3-4:1, connect-backs, authority inside the gate).- Sub-avatar card schema (research doc template + Evolve build): BEFORE / MONOLOGUE / WHO NOTICES / UNSPOKEN FEAR slots, VOC-receipted.
validate_native_copy.py: mechanism word share, reaction-line count inside the mechanism, at least one loop opener in the last two lines of each section (heuristic), emoji ban for T1/T2 bodies.- Final check: the two missing killers.
- SA2 v3 as the pilot: rerun the Fable/Sonnet lanes under the patched T2 and read v2 vs v3 on the same hooks and images.