Reflection — Saturday, 16 May 2026

1. Contradictions and Corroborations

Fresh corroboration today: iAds –MYR 104K loss is confirmed by two independent pipes — BQ daily revenue trajectory (MYR 22,650 May 12; MYR 32.7k May 11) and today's management accounts close. Neither source was fabricating; they now agree on the same loss figure through different lenses.

Subscriptions' 91-signed/zero-live finding is corroborated by ClickUp (project_blocked signal, 22 days escalating, magnitude 8) and the new Subs Affiliate Programme sheet wired in v1.11.0 — two sources that weren't previously paired landing on the same structural failure.

Contradiction: CMV Video's "complete activity halt from May" (facts.md May 5) conflicts with a deal book that grew from 69 to 77 during the halt before contracting to 76. The halt framing has been restated three days running without resolving what it actually means — Umar is confirmed host of a live shoot today, three deliverables are overdue, and 76 deals remain open. "Halt" is doing no analytical work here.

Hand-waved: The "71% pending inventory drop reads as pipeline contraction" claim (BUP-4, today and yesterday) has been asserted for two days with no BQ approval-inflow data cited. The brief states the interpretation, not the evidence.

---

2. Cross-Source Bridges

Strongest triangulation in the dataset: TikTok Shop as an existential platform dependency is now confirmed by four independent sources — BQ brand mix (iAds 43.82% TikTok contribution, AG Deepleaper USD 30k/day gap), Sheets (1.4M TTS conversions halted), ClickUp (project_blocked magnitude 8, 2 days), and facts.md Watch (9 days running). No single source makes this case; together they are unambiguous. This is promote-ready.

New pairing unlocked by GA4 (v1.8.0 + v1.11.0): GA4 cashiu app analytics (screen-level funnel via ga_aggregated_event) can now be paired with BQ cashiu.conversion_click_performance. The May 13–14 transaction freeze diagnosis — "payment gateway or offer-configuration failure" — is partially testable: if GA4 shows users reaching checkout but BQ records zero conversions, the break is post-checkout (gateway). If GA4 shows checkout abandonment, it's upstream. This question was unanswerable before May 14.

Active disagreement between sources: Today's signal store entry for inactive-publisher dormant recovery reads "MYR 396.9M in expected dormant earnings." Facts.md (May 15) records the same cohort as MYR 396.9k. One is wrong by three orders of magnitude. The May 15 facts.md entry explicitly superseded the prior "MYR 397.5M" error. The signal store's claim text has not been corrected and is re-injecting the wrong magnitude into tomorrow's pipeline.

---

3. New-Source Clarification

v1.11.0 (shipped May 15) wired the Subscriptions Affiliate Programme tracker. This single sheet changed the Subs story from "15 active accounts, low MRR, sales weakness" to "91 signed agreements, zero activations, unnamed integration owner blocking the entire programme." Before the sheet, BQ and ClickUp could see conversion concentration and account-count plateau but couldn't identify the upstream cause. The sheet reveals it: the 80/20 concentration finding Ed made on May 8 is not a publisher-mix problem — it's an activation backlog problem. The network can't convert what it can't activate.

What this grounds in facts.md: The May 9 entry "conversions concentrate in PG-managed publishers" is now better-understood as a symptom, not the root cause.

What this contradicts: Facts.md May 5 treats the 15-account figure as a floor with organic growth potential. The new sheet shows 15 is a ceiling constrained by implementation failure — the 91-backlog won't convert to active accounts without a named integration owner.

New questions now answerable: Age distribution of the 91-backlog (which agreements are oldest), and whether the 58 new signings / 102 rejections rate this week is accelerating or normal — neither was visible before May 15.

---

4. Pattern Emergence

project_blocked accounts for roughly 22 of 35 signals in today's store, spanning every BU. This is not a departmental ops crunch — it is a company-wide structural state. The pattern cuts across Group Services Engineering (CVEs, portal-legacy), CMV Video (AR stack, deliverables, Oppo KLCC), Subscriptions (affiliate handoff), Enterprise (CEP kanban), Corp Dev (iPrice). When every BU reads project_blocked, the signal is about governance, not execution.

The five CVEs are triply-counted: they appear as separate project_blocked signals under group_services/operations, group_services/engineering, and group_services/product. One unassigned issue, three signal entries — inflating apparent severity and diluting signal quality. The data model needs deduplication logic for the same underlying issue showing across boards.

Multi-year blocked items (commission upsize bug: 1,400 days; three Q1 2023 ops issues: 780 days; P&A Dashboard: 144 days; Mahadir's iADS Media Buy: 170 days) are surfacing daily alongside 2–3 day items. This creates false equivalence. They are permanent technical debt, not operational signals. Continuing to surface them alongside today's Oppo KLCC crisis implies they are actionable on the same horizon — they are not.

decision_pending_owner signals in marketing (Travetopia 7 days escalating, Klook Paid Post 9 days overdue) are publisher incentive decisions stalled at approval. They don't break through to Top 6 because the revenue impact is indirect — but delayed publisher incentive approvals erode publisher trust and belong on someone's weekly review list.

---

5. Promote / Demote / Retire

[Promote] TikTok Shop as structural platform dependency: BQ, Sheets, ClickUp, and Watch-list evidence converging for 9 days across four independent sources. Proposed fact: "TikTok Shop is IA's single largest revenue channel (iAds 43.82% brand contribution) and single largest operational dependency — platform validation failures, publisher violations, and conversion duplicates can compound simultaneously, as demonstrated May 7–16 2026." Promote or the Watch item should be retired as deliberately unresolved.

[Promote] Subscriptions affiliate programme as implementation-layer structural failure: "Subscriptions growth is bounded by an unnamed integration function. 91 agreements signed, zero live, no SLA on record as of May 16 2026. Account count ceiling is an implementation bottleneck, not a sales problem." Evidence: May 8 (Ed's 80/20 structural finding), May 16 (91/zero confirmed via new Sheets source).

[Promote] Wing Loon as COO-layer single point of failure: TikTok validation (Wing-owned), five CVEs (Wing's org), Jessie Tong's June compliance stack (no backstop, Wing's org), Vinita's dashboard (Wing's org). Four concurrent critical dependencies with no named deputy across any of them. Evidence: May 12–16 across multiple sources.

[Demote] [2026-05-14] iAds validation collapse: realization rate 87.4% (Jan) → 5.6% (May) — the May 15 CORRECTION explicitly supersedes this diagnosis. The original entry persists alongside its own correction, creating live ambiguity. Flag for human removal of the May 14 entry; the correction is the durable fact.

[Demote] [2026-05-12] Cashiu May 11 install anomaly: 32.7M installs against zero revenue/orders, unreconciled — explained by the Cartesian join bug correction documented May 13. Persisting as "unreconciled" is incorrect; it is reconciled. Flag for human review to mark superseded.

[Retire] [2026-05-01] Apr 30 group revenue anomaly: 48-hour unresolved — confirmed final MYR 494,723 on May 2. No longer load-bearing.

[Retire] [2026-05-03] iAds Apr 30 offer count revised four consecutive times — the settlement-lag pattern this exemplifies is now doctrine (May 4 revenue-freshness entry). This specific instance adds nothing further; retire.

[Retire] [2026-05-13] Cashiu AOV collapsed ~56% on 11 May 2026 — the same-day correction established May 11 AOV (MYR 46) was within the trailing 7d normal range (mean MYR 53, std-dev ~25); May 10 at MYR 106 was the outlier. The "collapse" framing was the distribution-anchoring error the facts.md rule was written to prevent. Retire the original entry; the correction rule stands.

[Flag — data integrity] Signal store today cites "MYR 396.9M" for inactive-publisher dormant earnings. Facts.md (May 15) records this as MYR 396.9k. The May 9 "MYR 397.5 million" error was explicitly corrected but has re-entered via the signal store's claim text. This will contaminate tomorrow's brief if not patched in the signal extraction layer before next run.