Reflection — Friday, 15 May 2026

---

1. Contradictions and corroborations

Fresh corroborations. The iAds zero-conversion diagnosis is corroborated by BQ: 88 Raja-managed non-iAds publishers logged 821k conversions on 87M clicks in the same 14-day window, confirming validation infrastructure is healthy and the iAds-specific zero is a real publisher-side outcome. CMV Affiliate gross-vs-net reconciliation closes: 14 May brief flagged the BQ gross ~MYR 39k vs P&L MYR 5,481 gap as unexplained; today's staff-cost decomposition (MYR ~30k) resolves it. Enterprise PG pace reverses the 14 May pessimism: yesterday required "~MYR 99k/day with no contributor at that rate"; today shows ~MYR 140k/day actual, making May closure structurally intact.

Contradiction. The 14 May BUP-1 claimed a validation-pipeline collapse (realization rate 5.6%, 167k USD gap). Today explicitly supersedes it: "Yesterday's validation-pipeline framing is superseded." This is a hard root-cause reversal between consecutive briefs. The facts.md entry from 14 May ("92.7% backlog deep-critical, 53% overdue past alert threshold") now requires correction — these metrics derive from Andrew Tay's internal iAds sheet and don't map to BQ aggregate validation latency as the 14 May brief inferred.

Magnitude error uncaught. BUP-6 today states Group Services May 12 revenue "revised upward to MYR 564M (from yesterday's 477M read)." GS daily revenue has run MYR 532k–600k across all preceding cycle days (10 May MYR 584k, 11 May MYR 576k, 9 May MYR 532k). MYR 564M is ~1,000× the plausible range. The facts.md magnitude-sanity rule applies; the actual revision was MYR 477k → MYR 564k. Pipeline rendered the unit suffix incorrectly.

Hand-waved. TikTok cross-BU SPOF (8 days): the 43.82% iAds brand-contribution figure and USD 30k/day AG gap are unchanged from their origin dates — neither has been refreshed by a new BQ pull nor has an owner been named. The Watch item is extending daily without re-examination. Claude Team Plan (18 days): "still unapproved and uncosted" — no new evidence of decision, slip, or escalation.

---

2. Cross-source bridges

Convergence — Trip.com violation source shift: ClickUp (iAds Stars halt documented, Huynh ICT surfaced as new primary), Google Sheets (iAds Reporting showing violation counts and source attribution), BQ (Trip.com offer revenue concentration in iAds publishers) all triangulate the same finding: stopping iAds Stars didn't reduce violation volume because the source had already shifted to Huynh ICT. No single source caught this; the Sheets + BQ pairing made it visible. Remedy-side: BQ still cannot pair offer × publisher on violations without a new metric join (the bigquery.md gap is documented).

Disagreement — iAds Sheets metric vs BQ publisher data: Andrew Tay's internal sheet shows "92.7% deep-critical backlog, 53% overdue past alert threshold" (cited 14 May). BQ publisher-level data shows validation is processing normally for the other 88 publishers. These two sources were implicitly conflated in yesterday's brief. The Sheets metric likely tracks iAds-internal offer-health or postback-latency fields unrelated to BQ conversion-validation pipeline; its meaning has not been confirmed in writing anywhere.

Disagreement — Subscriptions account-status: Hermes contacts.md records OPPO ID (Dwy) as "Accounts Going Live/Lived This May — Live." The 13 May brief showed OPPO ID regressed to "awaiting paperwork." Fifteen days later, contacts.md has not been updated. Cross-source inconsistency; canonical account state is ambiguous.

New pairing unlocked — GA4 + BQ cashiu.* : GA4 (v1.8.0, 14 May) + BQ cashiu.ga_aggregated_event (1.07M rows) enables an install→session→purchase funnel diagnosis that BQ alone couldn't support. Cashiu's DAU stall (~175 despite fresh installs) could be localised to a specific screen drop-off using GA4 event sequences joined to BQ conversion events. No brief has attempted this join yet.

---

3. New-source clarification

v1.11.0 (today) wired 11 new sheets across ops/escalations, iAds daily P&L, CMV target-vs-actual, Subs Affiliate Programme, and Cashiu daily depth.

Claims now better-grounded. The iAds zero-conversion diagnosis required publisher-level P&L visibility; the iAds daily P&L sheet (newly wired) is what enabled the aff_id 1008463 / 1065674 identification. Previously the brief relied on BQ aggregates and Andrew Tay's Thursday Basecamp posts — publisher-level daily tracking wasn't available. Enterprise PG May pace ("~MYR 140k/day actual vs MYR 105k/day required") is now grounded in the CMV target-vs-actual sheet; the 14 May estimate of "~MYR 99k/day required" was computed without the sheet's denominator. CMV Affiliate gross-vs-net reconciliation used the staff-cost decomposition now visible in the same sheet.

Claim needing revision from new sources. The Subs Affiliate Programme sheet reveals zero integration completions YTD against 91 signed agreements, 135 days running. The brief has been framing Subscriptions around weekly revenue collapse and account-count drift; this source says the structural issue predates both — 91 agreements sitting in limbo for ~135 days is the upstream cause of the conversion concentration problem Ed Guno identified on 8 May.

Questions the brief can now answer. Which specific iAds publishers are zero-converting (identified: aff_ids 1008463, 1065674)? What is the Subs Affiliate Programme integration completion rate (zero, now tracked)? Is Enterprise PG on pace for May (yes — the answer changed today because the CMV target-vs-actual sheet is now wired and the denominator is visible).

---

4. Pattern emergence

velocity_drift cluster — May as a company-wide break. Four velocity_drift signals are simultaneously active: AG team (2 days, -97% MoM initiatives), PG team (15 days, ~5 completions vs April's 65), Subscriptions Weekly Revenue (7 days, -85% WoW, cause unidentified), CMV Affiliate (7 days, near-zero May revenue). All four share April as a high-water mark and May as the drop. This is not BU-specific variance — it is a synchronized velocity pullback across every revenue-adjacent unit simultaneously. No brief has framed it as a company-wide pattern; it has been treated as four separate BU issues.

Trip.com as highest cross-BU signal-frequency entity. Trip.com appears in five distinct signal contexts in the current window: AG decision_pending_owner (click-limit violations, 45 days), iAds decision_pending_owner (iAds Social Project running Trip clicks at zero conversions, 14 days), Enterprise/iAds Stars halt (8 days), GS leadership escalations to Rizky/Barbia unconfirmed (45 days), and facts.md JP adult-content incident (04 May). No brief has assembled this five-source view. The cross-BU story is: Trip.com is simultaneously IA's largest advertiser, its primary compliance flashpoint, and the unresolved commercial relationship at most risk.

decision_pending_owner signals trending into de-facto-accepted state. CMV Video invoices (85 days), Subscriptions Sales Pipeline (60 days), Subscriptions Affiliate Programme (135 days project_blocked), Trip click-limit violations (45 days) — signals exceeding 30 days in decision_pending_owner type are structurally no longer pending decisions. The brief treats them as ongoing urgency; the pattern suggests they have been implicitly accepted as operating conditions without being named as such.

Unbriefd outlier — Group Services Operations MYR 2.61M approval queue (361 days). The signal "MYR 2.61 million in ready-for-recognition conversions blocked in approval queue; oldest cohort 361–726 days" has been in the signal store at days_running: 361. It has never appeared in Top 6. Its magnitude exceeds the Trip Stars RM 95k halt by 27×. If this is a revenue-recognition timing gap (not a cash loss), the brief should state so explicitly and retire the signal; if it represents real uncollected revenue, it should have been in Top 6 twelve months ago.

---

5. Promote / demote / retire

• [PROMOTE] TikTok as cross-BU structural risk: 8 consecutive days on Watch, today's brief flagged "promote next cycle if exposure persists" — exposure persists, no new owner named. Evidence dates: iAds 43.82% brand contribution (first cited ~7 May); AG USD 30k/day Deepleaper gap (8 May onward); GS TikTok validation offline (12 May onward). Promote as: "TikTok concentration is a cross-BU structural exposure with no cross-BU coordination owner."

• [PROMOTE] PG initiative delivery structural reset: 15 days of velocity_drift signal; facts.md 13 May already confirmed "structural collapse, not weekly variance." Evidence: April 65 initiatives → May ~5 through day 15 across all four tracked contributors. Promote as durable structural fact anchored on April-as-peak.

• [PROMOTE] Subscriptions 80/20 network economics: Ed Guno's 8 May finding corroborated across seven cycles without contradiction. Promote as: "Subscriptions revenue concentrates in PG-managed ~20% of publishers; scaling the broader network adds clicks, not conversions — structural, not correctable through publisher volume."

• [PROMOTE] May 2026 as company-wide velocity break: Four simultaneous velocity_drift signals with April as shared high-water mark. Promote as briefing-level structural context, not individual BU flags.

• [DEMOTE] iAds validation-pipeline framing (facts.md 14 May — "realization rate 5.6%, 92.7% deep-critical backlog"): Today's brief explicitly supersedes this. The facts.md entry inferred a validation-pipeline collapse from an iAds-internal Sheets metric whose actual meaning hasn't been confirmed. Flag for human review before correcting: what does "92.7% deep-critical backlog" actually measure in Andrew Tay's sheet?

• [DEMOTE] Subscriptions Changan/CelcomDigi/Comviva "Live" (facts.md 06 May): 09 May brief showed Changan as "Pending Integration" in the tracker; never resolved. Fifteen days later, none appear as active accounts in the brief. Flag for human review: are these currently Live or not?

• [RETIRE] MYR 397.5M dormant publisher balance (facts.md 09 May): Superseded by three-cohort breakdown in facts.md 15 May (Freeze MYR 1.306M, Inactive MYR 396.9k, Freeze→Active MYR 22,920). The 09 May entry is both the wrong number and the wrong frame.

• [RETIRE] Apr 30 revenue anomaly entries (facts.md 01 May, 03 May, 04 May): Fully explained by the T+1 lag rule (facts.md 04 May). No longer load-bearing as individual facts; the T+1 rule entry is the durable anchor.