00 — Scope, Spine & Benchmark Set

Step: 1 · Persona: Principal Product Strategist · Date: 2026-07-26 Status: ✅ GATE 1 CLEARED — Head of Product, 2026-07-26. Benchmark set, spine amendments, grounding corrections and the Group 5 demotion all approved as recommended. See §8 for the ruling and the constraints it introduces. Evidence base: 5 parallel verification brackets, ~300 Tavily operations. Raw findings in research/raw/{grounding-corrections, group1-verified, group245-verified, group3-verified}.md; source tables in research/raw/sources-*.md.

1. The decision this research must inform

Which AI investments should SatuSatu fund in the next four quarters, in what order, such that each one is traceable to a named money mechanism on SatuSatu's own contribution margin — given a four-direction strategy (curated Pass, horizontal catalog, partner dashboard, reseller API) that is being pursued simultaneously by a company with 253 SKUs, no disclosed funding since November 2022, and no AI currently running in its product.

Every clause in that sentence is load-bearing, and three of them are findings from this step rather than inputs to it.

2. Grounding amendments — the plan's §1 does not survive verification

The plan instructed subagents to challenge the grounding block. It needed challenging. Five premises are wrong, stale, or overstated. Full detail in raw/grounding-corrections.md; the four that change the research follow.

2.1 🔴 SatuSatu is no longer Bali-only — and may already be doing D1

[VERIFIED, Tier A — live site] The footer lists 15 locations, 7 outside Bali (Sleman 7 SKUs, Yogyakarta City 6, North Jakarta 5, Malang 2, Magelang 2, Bantul 2, West Jakarta 1). The catalog also sells Singapore: Universal Studios Singapore, Singapore Oceanarium, River Wonders — plus KidZania Jakarta, Jakarta Aquarium Safari, Saloka Theme Park. Catalog size is a hard 253 SKUs.

[INFERENCE, high confidence] Universal Studios Singapore displays "4.6 (11k+ ratings) · 99k+ sold" — against a plan-stated ceiling of "1k+". Six-figure inherited counts on cross-border branded parks is the signature of ingested wholesaler inventory, not locally-contracted Balinese supply.

If that inference holds, three things follow at once:

• Distribution-in is not a plan, it is a fact. The mapping/dedup/taxonomy/quality-gate cluster can be sized CURRENT, not TARGET — pulling an entire opportunity class out of the plan's own "sized against a hope" trap (risk §1.3(1)).
• Risk §1.3(6), single-destination concentration, is partly obsolete. It must be re-dated, not restated. Any opportunity deferred to Later because "it only pays at multi-destination scale" loses that justification.
• The curation-vs-completeness tension is live today, not a future risk. A brand promising "Experience Indonesia, The Local Way!" and "Discover authentic Bali experiences" is currently selling Universal Studios Singapore.

⚠️ This is the single most important thing for you to confirm or deny before Step 4 runs, because it changes what Step 4b researches, not merely what Step 5 sizes.

2.2 🔴 The +50% margin claim is correlational, and single-source

Every Tier A/B source is sequential, never causal: "following deployment" (TechNode), "after it deployed" (Tech in Asia), "setelah penerapan mesin AI" (VOI). The only causal phrasing in the corpus — "boosting contribution margins by 50%" — is BeamStart, a Tier C aggregator rewriting TechNode. The causal verb is BeamStart's addition. Do not cite it.

Three compounding problems: all five press accounts derive from one TipTip release dated 2026-05-04 (Tech in Asia flags its own sourcing as "🔗 Source: TipTip"); "+50%" is a relative lift off an undisclosed base; and the same release offers two competing explanations that need no AI at all — the GTV-based take-rate pivot replacing tips, and >50% self-onboarded promoter supply removing sales cost.

2.3 🔴 There is no evidence the AI transferred, and the transfer is structurally hard

[VERIFIED, Tier A — live product] SatuSatu exposes zero AI surfaces. Keyword/location filtering only — no semantic search, no itinerary generator, no personalisation. The concierge is explicitly marketed as human: "access to a real Bali local who plans your days."

The four sources linking AI to SatuSatu all say "AI-based ticketing/booking infrastructure" and cite instant confirmation, same-day reservation and skip-the-queue — integration features, not model outputs.

And the structural argument is stronger than the evidentiary one. Concert ticketing is a small number of high-value, date-fixed, single-shot events with a pre-sale demand curve. TAA is 253 evergreen, daily-departure, low-ASP SKUs with no comparable pre-sale signal. It is a different forecasting problem, not a port.

→ Register rule: the demand-prediction engine is an unvalidated, non-transferred asset. It must not be scored as an existing SatuSatu data asset or capability.

2.4 ⚠️ Corporate structure — reword plan §0

[VERIFIED, Tier A] SatuSatu's T&C and Privacy Policy, both dated 2026-01-19: "SatuSatu is a system operated by PT Tiptip Network Indonesia." The App Store developer of "TipTip: Event & Concert Ticket" is the same entity.

This is not parent-owns-subsidiary. One legal entity operates both brands. SatuSatu is a product line, not a company TipTip owns.

Plan §0 requires every P&L to "land on SatuSatu's line, not the group's." That rule survives — but as a management-accounting boundary, not a legal one. There is no separate SatuSatu P&L, cap table, or engineering org to assume. Practically this makes internal reuse cheaper than the plan assumes — if the capability were relevant, which per §2.3 it is not.

⚠️ Jakarta Globe's "SatuSatu, in partnership with TipTip" (2026-05-06, B) is wrong. Do not propagate.

2.5 Smaller corrections that still matter

3. Value-chain spine — validated, with three amendments

The 20-stage spine is sound and I am not proposing a restructure. Three amendments, all evidence-driven:

Amendment A — Stage 3 is retagged D0 D1, not D1

Per §2.1, distribution-in appears to be already live. A stage tagged as future-only will be under-researched by Step 4b and mis-sized as TARGET by Step 5b. Contingent on your confirmation.

Amendment B — Stage 5 is the true cross-cutting prerequisite, and no direction owns it

Stage 5 (inventory & availability — real-time capacity for operators with no booking system) is already tagged D0 D1 D2 D3. That tagging is correct and its implication is under-drawn. Evidence from three independent brackets converges here:

• Arival/Phocuswright (B): the experiences market is "highly fragmented with hundreds of thousands of mostly small, private operators… no definitive registry of businesses exists"; >70% are small or micro; offline direct sales still dominate many local markets.
• Group 3: every connectivity player of consequence — Bókun, FareHarbor, rezio, Ventrata, the Expedition group — exists precisely to solve this, and none of them has a verified Indonesia footprint. GlobalTix is the sole exception.
• Group 1: even Klook, with an NYSE F-1 and ~310,000 offerings, ingests via third-party channel managers rather than solving operator connectivity itself.

→ Stage 5 is simultaneously the hardest problem, the most defensible asset, and the hard blocker for D2 and D3. Step 6 must place it in the prerequisites lane and Step 5 must not let any D3 candidate assume it away. See §6(b).

Amendment C — Stage 6 must carry deferred-liability/float management explicitly

Stage 6 currently reads "Pass bundle composition and breakage economics." Two Group 2 findings make this its own risk surface, not a pricing sub-topic:

• 🔴 Sightseeing Pass suspended operations ~June 2025 and stranded prepaid customers — three independent confirmations, including Go City publishing a customer-poaching landing page within days. ⚠️ Wind-down details unverified; no trade press covered it.
• 🔧 Go City tightened both float levers simultaneously in 2025: Explorer redemption window 60 → 30 days (post-2025-04-01), and pre-activation shelf life 24 → 12 months.

SatuSatu's Pass carries a 90-day activation window — a deferred liability against unredeemed prepaid passes. The category leader moved decisively in the opposite direction, and the category's most recent failure was exactly this risk realised.

All other stages and D0–D3 tags: confirmed as written.

4. Validated benchmark set

Every name in plan §4 was unverified. All are now checked. Verdict: 6 promotions, 4 drops, 5 reclassifications, and 6 additions the plan never named.

Group 1 — Destination-state model [D1 D2 D3] · full teardown

All five confirmed live and retained. But four of the plan's five B2B artefact citations are wrong — fix before Step 3 or it will read the wrong documents:

🔴 PROMOTED INTO GROUP 1 — Klook Pass. [VERIFIED, Tier A live product] Klook Bali Pass exists today: choose 2/3/4/5 activities from 40+ Bali activities, "save up to 45%". Also live in Singapore, Greater Tokyo, Hong Kong, Seoul, and 7 other cities.

This is a direct, same-destination, same-mechanic competitor to SatuSatu's hero SKU — 40+ Bali activities against SatuSatu's "40–50+". And it inverts the plan's central risk. Risk §1.3(2) fears SatuSatu goes horizontal and loses curation. The demonstrated risk is the reverse: a horizontal incumbent bolts a pass on in a week. Klook did it as a merchandising wrapper over an existing catalog. Any D0 defensibility argument that does not address this is incomplete.

Group 2 — Curated / pass / concierge [D0]

The curation question resolves cleanly, and one pattern is survivable:

➕ Concierge/DMC placeholder — filled. There is no Bali-native concierge/DMC technology company worth studying; that slot is SEO order-form sites and USD 5–20k chatbot agencies, all Tier C. Three real entities replace it:

• Elite Havens ⭐ — ~300 luxury villas across Bali/Lombok/Phuket/Maldives, ~70–80k guests/yr, integrated concierge and villa management. Dusit Thani PCL, September 2018 (acquirer's own newsroom, A; ~$15M, B). → The closest real-world analogue to "dedicated human concierge, curated by locals" in Bali, running 28 years, with a resolved ownership outcome. Also a plausible distribution partner — captive high-spend Bali guests needing exactly SatuSatu's inventory.
• Holibob — $12M Series A (July 2022, A). GraphQL API; positions on "collecting, curating, and distributing tours and attractions". The only verified company productising curation-as-a-B2B-service — literally the plan's question. ⚠️ No metric since 2022.
• Canary Technologies — $80M Series D (2025-06-12, A). Hotel AI concierge with activity-recommendation surfaces. → Defines the channel through which a Bali hotel guest is most likely to be sold an activity in 2026, and is the direct answer to "can software replace a human concierge?" Include as adjacency, not head-to-head.

Group 3 — Connectivity & B2B infrastructure [D2 D3] · full teardown

✅ The vertical-integration thesis survives cleanly, Tier A on every leg. Tripadvisor→Bókun (2018-04-20), Booking→FareHarbor (announced 2018-04-19), KKday built rezio. GetYourGuide is the deliberate exception — confirmed by positive proof, not a null result: its own Supply Help Center (updated 2026-02-10, A) names its connectivity partners as Bókun and FareHarbor among others. GYG routes supply through the systems of its two largest competitors. That is a nameable strategic vulnerability and Step 3 should treat it as one.

⚠️ The consolidation map is a 2023–24 snapshot and has missed three events. Refreshed to July 2026:

🔴 Three plan errors to fix: Redeam is not an independent network (acquired a year ago). Palisis is not a standalone (owns TourCMS, and both now sit under Experience Technology Group). TrekkSoft is alive and has a parent — the plan listed it as an independent unknown; both halves are wrong.

⛔ A Tier C trap to exclude explicitly: businessmodelcanvastemplate.com asserts Accel-KKR acquired Rezdy and merged the three. False — Accel-KKR's own portfolio page (A) lists none of them. This is exactly the class of SEO-circulated error that inverted v1.0 of this plan.

⚠️ "Independently-owned" needs a footnote everywhere it appears. The phrase is Rezdy's own marketing (2023-08-08 blog) — Tier A as a quote, Tier C as a fact. In this sector it now means "PE-owned rather than OTA-owned", and Expedition's PE-owned independence is run by the founders of the Booking-owned incumbent. Palisis+Prioticket now claim the same superlative. Repeat neither unqualified.

🌏 GlobalTix — individual attention, and it earns it. ✅ Independent, Singapore, founded 2013. Series B S$6.5M led by Tin Men Capital, 2024-10-14. ⚠️ Double-count trap: a second release (2024-12-19) headlined "$5 Million Series B" is the same round in another currency. Do not sum. 🔥 Indonesia footprint is the strongest in the bracket and materially relevant to D3:

• InJourney Destination Management — an Indonesian state-owned tourism sub-holding — appointed GlobalTix official channel manager for the Borobudur Temple Compound, announced 2025-10-07/08 (GlobalTix release, A).
• 10 Asian offices including Indonesia; a dedicated GM Indonesia (Helmy Pietersz).
• Indonesian inventory already includes Waterbom Bali, GWK Cultural Park, Taman Safari Indonesia.

⚠️ Downgrade the plan's scale figure. The 15,000 partners / 180,000 experiences pair appears only on LinkedIn and a LinkedIn-scraping aggregator (C). GlobalTix's own homepage, fetched today, says 12,000+ agents / 150,000+ experiences / 25M tickets annually. Use the company-controlled numbers, keep the Tier C reliability label. The meaningful signal is 12M → 25M tickets between Oct 2024 and now, not either absolute.

Group 4 — Indonesia & SEA [D0 D1 D2]

🔴 The D2 finding: the partner dashboard enters contested, not empty, space.

[VERIFIED, Tier A live product] Golden Rama already operates two-sided B2B distribution infrastructure. rols.golden-rama.com is a live reservation system carrying both an "Agent Login" and a "Supplier Login". Its consumer site carries a dedicated Attractions category. Offices in Jakarta, Bandung, Surabaya, Denpasar, Makassar.

➕ And an unplanned direct threat to D3: TBO Holidays — a live B2B travel portal advertising 200,000+ global sightseeing products (A). "Who already competes for the agent partner" is answered better by a global B2B wholesaler with a live sightseeing catalogue than by anything currently in the plan.

❌ "WhatsApp-first SEA players" — refuted as a category. Three search framings, zero named consumer-facing companies. What exists is a vendor layer (Trawex, AiSensy, respond.io, Spur — all C) selling WhatsApp engines to travel businesses.

🔴 The competitively significant fact inverts the placeholder's premise: WhatsApp is table stakes for Indonesian incumbents, not a differentiator. Antavaya publishes a named WhatsApp concierge — "AMY", +62817 768 838 — on its homepage; Golden Rama runs a WhatsApp widget; Pigijo lists WhatsApp as primary contact. Any D0 defensibility claim resting on "we serve on WhatsApp" is unsupported.

Bali DMC layer — partially fillable, and the fragmentation is the finding. Only Pacto and Panorama Destination are nameable at scale. Below them: SEO-marketed long tail with no disclosed volume, headcount or funding — not dressed up as a competitive set. [INFERENCE] For D2 this cuts both ways: no dominant incumbent to displace, but no concentrated set of logos to sign — a high-volume, low-ACV motion, with direct consequences for CAC and whether D2 can carry human-assisted sales at all.

Group 5 — Disintermediation watch [threat only]

🎯 Verdict: the discovery layer is being disintermediated now. The booking layer is not.

As of 2026-07-26, the number of tours, activities or attraction tickets any general-purpose AI assistant can book end-to-end on its own rails — without handing off to an OTA checkout — is effectively zero. The most important event of the last twelve months was a retreat.

• 2025-10-06 OpenAI DevDay — Apps SDK with Expedia and Booking.com as first travel partners; MCP + Stripe checkout.
• 2026-02-16 — Instant Checkout ships for US users via ACP.
• 🔴 2026-03 — OpenAI pulled back. Instant Checkout deprecated in favour of merchant-run apps; OpenAI told Skift it is "prioritizing making ChatGPT search and product discovery great, with the Agentic Commerce Protocol serving as the infrastructure" (Skift, PhocusWire, B).
• Klook is genuinely live on ChatGPT — and its own announcement gives the game away: "you'll get a curated list of relevant options from us. And when you're ready, just tap 'View on Klook'."
• GetYourGuide's 2026-04-22 "Spring release" (A) is review summarisation and search on its own funnel — a conversion layer, not an external agent.
• Trip.com TripGenie reports ~60% booking-related interactions (2026-03-16, B) but books only Trip.com inventory.
• Payment rails (ACP, Google AP2, Mastercard Agent Pay, Visa TAP) are real standards with real bank participation, but essentially all substantive sourcing is Tier C vendor content, and there is no Tier A/B evidence of a single activity booked end-to-end over any of them.

→ Sequencing consequence: disintermediation is a distribution and SEO problem on a 12–24 month clock, not an existential booking-layer threat. The nearer-term threats are conventional and already live: Traveloka × Trip.com pooling attractions inventory (June 2026), Airbnb expanding human-supplied Experiences, and Golden Rama already running an agent-and-supplier portal in Jakarta.

5. Explicitly out of scope

1. All tiptip.id ticketing, creator, and sponsorship opportunities. No opportunity will be scoped, sized, or recommended for them. TipTip appears only as an asset/constraint inventory — with the §2.4 correction that it is the same legal entity, so the boundary is management-accounting, not legal.
2. Non-AI product and growth work — pricing strategy, brand, paid-channel mix, supplier commercial terms — except where an AI opportunity depends on one, in which case it is named as a dependency, not analysed.
3. Implementation-level architecture. Step 5a assesses feasibility, data readiness and inference cost. It does not choose models, vendors, or frameworks.
4. Full financial audit of TipTip group. Group figures are context only, and per §2.2 are single-source, company-announced, unaudited.
5. Legal, tax, and regulatory opinion on B2B contracting, cross-border settlement, or Indonesian tourism licensing. Flagged as risks where relevant; not resolved here.
6. Primary research — no customer interviews, supplier surveys, or partner discovery. Every external claim is desk-verified. Where the answer needs primary data, Step 5 names the cheapest experiment that would resolve it (plan §7.13).

6. Coherence check on the four-direction strategy

The plan asks for this before ~12 subagents build on the strategy. It is not a rubber stamp.

(a) Can D0 (curated, concierge, premium) and D1 (horizontal, complete, self-serve) coexist in one brand and one storefront?

Yes — but not with the promise as currently worded, and the tension is already live rather than hypothetical.

The evidence splits three ways and one pattern survives:

• Fever went horizontal by acquisition and the curation brand became a fee complaint.
• Musement was absorbed, and TUI is now re-narrowing toward own-branded curated product on margin grounds.
• Civitatis grew to ~90,000 activities and kept the claim intact — by anchoring it to something verifiable and non-scaling (Spanish-language guiding) rather than to catalog size.

SatuSatu's promise is currently anchored to the catalog — "locally curated experiences", "Experience Indonesia, The Local Way!", "Discover authentic Bali experiences". A catalog-anchored promise breaks mechanically as the catalog grows, and per §2.1 it is already breaking: the storefront sells Universal Studios Singapore under a banner promising authentic Bali.

Recommendation: re-anchor the promise to the concierge and the local operator relationships — both verifiable, both non-scaling, both genuinely SatuSatu's. Then the catalog can go horizontal underneath without falsifying anything. This is a positioning decision, not an AI decision, and it should be made before D1 accelerates rather than after.

Precedent for separation, since the plan asks either way: Go City runs four pass structures under one brand; Klook runs a pass as a merchandising wrapper over a horizontal catalog. Neither needed brand separation. What they needed was a promise that did not depend on the catalog being small.

(b) Does D3 have anything exclusive to sell?

Yes — exactly one thing, and it is also D3's hardest blocker. That coincidence is the strategic crux of this whole research.

Working through what is not exclusive first:

• Branded parks and cross-border inventory (USS, Singapore Oceanarium, Jakarta Aquarium) is feed-sourced and therefore definitionally non-exclusive — any reseller reaches it through the same channels.
• Major Indonesian attractions are increasingly spoken for: GlobalTix holds the Borobudur mandate from an Indonesian SOE and already carries Waterbom Bali, GWK, Taman Safari.
• The API itself is table stakes — Klook, GYG, Viator, KKday and Headout all ship one, and Viator's is an industry default with formal certification.

What remains exclusive is the long-tail Balinese supply — the ATV, jeep, snorkel, spa, transfer, temple/dance and car-charter operators. And the reason it is exclusive is precisely that >70% of operators in this market are small or micro, offline direct sales still dominate, and no connectivity player of consequence has an Indonesia footprint (Arival/Phocuswright B; Group 3 sweep).

So D3's only defensible inventory is the inventory that has no real-time availability — which is the one thing an API cannot sell. Spine stage 5 is simultaneously the moat and the blocker. Step 5c must not let any D3 candidate assume availability away, and Step 6 must sequence stage 5 ahead of the API surface, not alongside it.

(c) What is the stated position on D2/D3 channel conflict with D2C?

Unknown — {{B2B_PRICING}} (Q15), and it is cheap to answer. Without it, Step 5c must assume worst-case cannibalisation and will kill D2/D3 rows that a stated policy would have saved. Answering this one field protects the candidates the plan is most worried about under-generating.

Note the additional pressure from §4: Golden Rama already runs an agent-and-supplier portal, and Traveloka is already buying and reselling third-party inventory. D2/D3 are not entering an empty market, so net-rate discipline will be tested immediately rather than eventually.

(d) Which direction carries the platform investment the other three free-ride on?

Nominally D1. Actually stage 5 — and no direction currently owns it.

• D1 carries the visible platform cost: feed ingestion, product mapping, dedup, taxonomy normalisation, automated quality gating. D2 and D3 free-ride on it entirely — partners cannot be sold an empty or bad catalog.
• But the deeper shared prerequisite is stage 5, real-time availability, and it is not owned by any of the four directions. D0 needs it for instant confirmation (already an advertised USP). D1 needs it for bulk SKUs to be sellable. D2 and D3 are hard-blocked without it — a partner dashboard that cannot quote live availability is a quote form, and an API that cannot confirm is a catalogue feed.

Recommendation for Step 6: the prerequisites lane should be ordered stage 5 → stage 3 quality gate → D2/D3 surfaces, and this chain should be presented as dictating sequence more than RICE does (the plan anticipates exactly this in Step 6).

(e) An unasked question I am raising: is four directions coherent at this size?

The plan asks me to question the strategy, so: 253 SKUs, no disclosed raise since November 2022, a just-achieved EBITDA-positive position, and no AI in the product today — against four simultaneous directions, three of which are pre-launch.

I am not recommending dropping a direction; that is a leadership call and the directions are individually sound. But two things follow that the register must respect:

1. Capital discipline is a hard constraint, not a preference [INFERENCE from §2.5]. Any bet with a long pre-revenue build should be ranked accordingly, and Step 5a should treat "buy/partner" as the default rather than the fallback.
2. {{CAPACITY}} (Q16) is more decisive than the plan treats it. If D2/D3 platform work already consumes the engineering team, then every AI opportunity competes with the strategy rather than supporting it — which changes the question from "which AI bets" to "which AI bets fit in the gaps." That is a different report, and I would rather know now than at Gate 2.

7. Open Questions

Seeded in 99-open-questions.md — 28 entries, each tagged by blast radius and by the recommendation it destabilises. Steps 2–8 append continuously.

Answered by this step: Q21 (ownership map — survives, with corrections), Q22 (TipTip↔︎SatuSatu — same legal entity), Q23 (GlobalTix scale — downgraded to 12k/150k/25M, self-reported), Q26 (dropped SEA candidates — four recovered and promoted), Q25 (agentic booking — effectively zero, and OpenAI retreated).

Newly raised: whether distribution-in is already live (§2.1) — the highest-value single question in this document; whether the Pass float exposure has been modelled against the Sightseeing Pass precedent; and whether the curation promise will be re-anchored before D1 accelerates.

8. ✅ GATE 1 — ruling, 2026-07-26

8.1 Constraints introduced by the Step 0 answers

These bind every subsequent step and must be carried into each subagent brief.

🔴 {{CAPACITY}} = none dedicated. The D2/D3 platform build consumes the team. Effort is the binding constraint, not impact. Consequently:

• Every candidate competes against the company's own stated strategy for the same engineers — Step 5b must state what each row displaces
• Buy/partner is the default; build carries the burden of proof in Step 5a
• "Now (≤90 days)" is restricted to near-zero-engineering interventions — vendor-delivered, configuration-only, or ops-process changes
• Rank on impact per eng-week, not raw RICE, which would surface unstaffable rows
• Register shape (approved): keep plan §5's ≥3 candidates per D1/D2/D3 to guard against D0 gravity, but mark rows that cannot be staffed within two quarters [UNSTAFFABLE — documented for sequencing, not proposed] rather than padding the ranked table

🔴 {{MARGIN}} = thin, on third-party net prices. Sizing and feasibility must split by supply pool. Still outstanding: numeric bands per pool.

🔴 {{B2B_PRICING}} = tier by supply source.

→ D2/D3 is built on Pool B. Pool A is catalog filler for coverage, not a B2B product.

8.2 Three findings that follow, and that Steps 2–5 must carry

1. Stage 5 is the gating dependency for the entire B2B strategy — not one prerequisite among several. D2/D3's commercial rationale rests on Pool B, and Pool B is exactly the inventory whose operators have no booking system. Step 6 sequences it first, ahead of any partner-facing surface.
2. Growing Pool B is the B2B revenue lever — not growing total SKU count. Horizontal expansion via aggregators buys coverage and search-success, but no B2B revenue. The register must stop treating these as one goal.
3. The constraints stack: no capacity + thin margin + no raise since Nov 2022. Together they argue for cheap, bought-not-built interventions against costs that already exist. Mechanisms (d) cost-to-serve and (e) leakage reduction survive this filter best; anything with a long pre-revenue build is disadvantaged on principle, not on preference.
4. ⚠️ Single-feed concentration risk is now live. GlobalTix is simultaneously the incumbent supplier, holder of the Borobudur mandate from InJourney (an Indonesian SOE), and a direct D3 competitor. That is a dependency worth a register row of its own.