Honesty: what we did not find

This page documents the limits of this research: known gaps, dead-end research pathways, opaque domains, and contradictions we could not resolve. Honesty is a first-class output here. A research artifact that hides what it does not know is not trustworthy.

Auditor: reflexion-auditor (Wave 5) | Date: 2026-05-04 | Word target: ≥1500

Reviewer note: This is the most important deliverable of the audit. Read it before acting on any of the optimistic findings.

Preamble: Why an honesty section exists

Research that reports only what it found is biased toward action. A vendor reading the knowledge graph sees 100 initiatives, 28 of them Tier-A, and concludes that the central-asia opportunity surface is rich and well-mapped. That conclusion is partially correct and partially dangerous. The opportunity is real; the map is incomplete in specific, named ways. This section names the incompleteness so that the next reader cannot mistake "what we wrote down" for "what is true."

Five questions structure this section: (1) What did the research not find? (2) What questions remain unanswered? (3) What pathways were investigated and proved dead? (4) What would a hostile critic of this research say is missing? (5) What is the single biggest known unknown?

1. What this research did NOT find

1.1 Defense ministry AI procurement (both countries)

The Uzbek Ministry of Defense and the Kyrgyz Ministry of Defense both procure digital, biometric, and increasingly AI-enabled systems. None of those procurements appear in this knowledge graph. Reasons: (a) defense procurement in both countries is governed by classified procedures separate from the civilian e-procurement portals; (b) export-controlled hardware (Nvidia H100, certain telecom equipment) creates layered counterparty rules that don't appear in public tenders; (c) Russian-co-financed defense items (KG via CSTO, UZ via bilateral) are politically sensitive and undocumented in the open press; (d) Western defense contractors operating in CA are subject to ITAR/EAR controls that further reduce visibility. The result is a systematic blindspot for any vendor whose product has dual-use applications. A vendor pitching civilian AI is technically eligible for defense contract conversations through MoD digital divisions, but the knowledge graph has no map of those divisions or their decision-makers. This is a known unknown of large size.

1.2 Internal security / law-enforcement digital procurement

Beyond the named Minister of Internal Affairs (Pulat Bobojonov, UZ — biometric ID, traffic AI cameras), the layer of officials inside MVD/СНБ that actually drive digital and AI procurement was not mapped. The Russian-language acronym "ОТ ДУП" (operational-technical departments) covers signals intelligence, lawful-interception adjacent capabilities, and increasingly facial-recognition deployments. Vendors operating in this space are typically Russian, Israeli (NSO and competitors), or Chinese. The knowledge graph captures none of these vendors because none of them publish project lists. Honest framing: anything in the security/biometric/lawful-interception adjacent space is not here.

1.3 Chinese vendor footprint depth (Huawei, ZTE, Alibaba, China Telecom)

Chinese vendors are visible in the knowledge graph at the level of named partnerships (Shermatov / Huawei MoDT MoU 2023; Huawei Smart City presence in Tashkent and Bishkek), but the project-level footprint is opaque. Specifically: (a) Huawei Cloud Stack underlies UZ's UzCloud — the public cloud is described as "sovereign" but the underlying stack is Chinese; (b) ZTE has equipment in both UZ and KG telecom backbones — implications for AI applications running over that infrastructure are not captured; (c) Alibaba Cloud has presence in Almaty and may extend service into both UZ and KG without separately registered local entities; (d) Chinese AI/ML vendors (SenseTime, iFlyTek, Megvii) — surveillance/recognition — likely have a footprint that the open-source corpus does not surface. The Russian/CIS Substitution lens (Lens 6) systematically underweights China as the actual substitution partner. This is a research gap that requires Chinese-language sources to close.

1.4 Russian co-financing channels (EDB, EFSD, bilateral)

The Eurasian Development Bank (EDB) appears in the knowledge graph (Tigran Sargsyan named as Chairman; EAEU digital infrastructure fund $200M cited). The Eurasian Fund for Stabilization and Development (EFSD) appears thinly. Russian state-bank co-financing through Sberbank, Promsvyazbank, or VTB into CA digital projects is essentially absent from the records. This is partially because such financing has receded post-2022 sanctions, but it has not disappeared — particularly in KG where the CIS Substitution lens applies more strongly. A vendor positioning as "neither side" (per the CIS Substitution lens) needs to know who the Russian-side competitors and partners are; this knowledge graph cannot fully tell them.

1.5 USAID-successor entity ambiguity

USAID exited Central Asia in 2025 amid wider US-government reorganization. Successor entity structure (Department of State digital programs? International Development Finance Corporation? Different USG entity?) is ambiguous in CA-relevant project handovers. The knowledge graph mostly omits USAID-successor entries because the institutional layer is not yet stable. Programs that USAID had been running (digital governance support, judicial reform tech, civil society digital tools) are in a transition state; some have been picked up by EU GIZ/FCDO, some have been dropped, some are in transitional limbo. A vendor relying on "USAID-funded digital governance pipeline" in 2026-2027 is operating on outdated information.

1.6 Bishkek and Tashkent municipal digitalization beyond high-level summaries

City-level digital initiatives — Tashkent smart city, Bishkek e-municipality, Samarkand Vision 2030 — appear at the strategic-document level but the named officials, procurement schedules, and donor pipelines at the municipal level are thin. Cities matter because they procure differently from ministries (more frequent, smaller tickets, faster) and because mayors have political profiles that ministers don't. This research underweights municipal opportunities.

1.7 Saida Mirziyoyeva profile depth

Saida Mirziyoyeva is the daughter of President Mirziyoyev and holds a presidential aide / advisor role that intersects with ICT, education, and digitalization policy. She is mentioned (e.g., as a confirmed Digital Bridge Tashkent 2026 speaker in INI-001) but does not have her own person record. Given her structural position, this is a meaningful omission. This is partly a question of MUST NOT around speculation about personal political loyalties, but the current role and policy footprint can be documented from public statements without crossing that line.

2. Unanswered questions

2.1 What is the actual disbursement rate of WB Digital CASA P160230?

The record states $51.3M of $57M disbursed (~90%), but the project end-date is March 2025 and the audit is May 2026. Either the project closed with that disbursement rate (likely) or there were later disbursements not captured. Honest answer: only commitment data is reliably captured; disbursement-tracking quality varies by program.

2.2 What is the actual operational status of the $100M UZ AI Fund (PP-320)?

The decree was signed 30 October 2025. The fund's 2026-2027 allocation is committed. As of May 2026, fund criteria are reportedly being drafted (per INI-001 problem_statement) but specific awarded amounts, recipient companies, and selection committee members are not in the public record. A vendor pursuing fund access is operating in a 6-month criteria-shaping window without confirmed end-state criteria.

2.3 Is UZ-LAW-2026-1125 a real law, or a hallucination?

This is a HIGH severity unanswered question (audit correction C-004). Two initiatives reference a "Personal-data law amendment UZ-LAW-2026-1125 requires data localization for judicial PII" but no such law appears in the decrees file. The legal-cartographer agent should be re-run on this specific question; until answered, the regulatory pathway for any judicial-data initiative is uncertain.

2.4 Who runs the new UDP Digital Transformation Department in KG?

Currently named: Adilbek Asanbekov as Head, Timur Koichubaev as Deputy. These names predate the Mintsifry liquidation by approximately 0 days — they are inferred from press leaks, not from published structural regulations. Until UDP regulations publish (~31 May 2026), the actual leadership of the department is unconfirmed.

2.5 Will Talant Sultanov stay in government, move to UDP, or exit?

The transition from Minister of Digital Development to UDP Senior Advisor is a plausible institutional path, but Sultanov is internationally networked (CMU + AUCA) and could plausibly move to a private/donor role. The knowledge graph assumes he stays in government. Audit flags this as INFERRED.

2.6 What is the next step for the failed KG GPU/Supercomputer tender (KG-T-2026-003)?

The record notes "Previous tender failed due to export control issues with Nvidia (H100 chips on ECCN). New tender re-launched with US Embassy engagement. Potential Chinese GPU alternative (Huawei Ascend) being evaluated." The actual technology decision (Nvidia approval vs. pivot to Ascend) will be highly consequential for any AI vendor planning to operate against KG state compute. Outcome unknown.

2.7 How many of the "100 AI implementations by end-2026" (УП-189 target) are actually being implemented?

The decree commits to 100. As of audit date, no public registry tracks which ones are in flight, completed, or stalled. A vendor positioning as "the 51st implementation" needs this register to exist; it doesn't.

3. Dead pathways investigated

3.1 Direct procurement portal links for UZ tenders

We attempted to populate tender_url fields with specific tender notices on etender.uzex.uz. For several tenders (UZ-T-2026-002, UZ-T-2026-003, UZ-T-2026-005), only the platform homepage URL is recorded. Specific tender notice URLs were not consistently retrievable; tenders may exist on the portal but require login or specific reference numbers to access. Dead pathway: open-web crawl of tender portal homepage gives no specific notice details.

3.2 LinkedIn URL verification for Tier-1 individuals via free search

30 of 30 Tier-1 individuals have either not_found, unverified_match, or undocumented LinkedIn status. A free LinkedIn search produces too many name-matches to confidently disambiguate. Dead pathway without paid Sonar Pro: free LinkedIn cannot resolve "is this the correct sherzod-shermatov among 47 listed?"

3.3 cbd.minjust.gov.kg full-document fetching

Several KG decree URLs return only partial content via WebFetch. The portal serves search-result snippets but full decree text often requires authenticated access or PDF download. Dead pathway: WebFetch alone cannot fully verify decree text; cross-referencing to secondary press is the workaround used.

3.4 Donor PAD (Project Appraisal Document) deep extraction

WB PADs are 80-200 page PDFs containing component-level budget breakdowns, named TTLs, and detailed disbursement schedules. Open-web fetching of these PDFs is brittle. The knowledge graph mostly relies on press release language ("project total $50M") rather than PAD-level extraction. Dead pathway: deep PDF extraction from open web is unreliable; deeper donor research requires authenticated access to documents.worldbank.org and adb.org/projects.

3.5 SOE bank IT director identification via open press

State-controlled bank IT directors are rarely named in press. Their decisions matter for AML/KYC AI procurement. Dead pathway: open press doesn't surface these names; LinkedIn / Sonar Pro / direct outreach is the only route.

3.6 CSTO / EAEU defense digital cooperation projects

Russian-led collective security and economic union digital cooperation has documents but they are not consistently translated, indexed, or open. Dead pathway: open-web English/Russian search reveals strategic-document level mentions but not project-level detail.

4. What a hostile critic of this research would say is missing

A hostile critic of this knowledge graph could legitimately say all of the following:

(a) You overweight donor-financed pipeline. The knowledge graph treats World Bank and ADB pipelines as the canonical procurement universe. In reality, state-budget-only procurement (Treasury-financed digital programs, ministry-internal IT spend) is a comparable or larger pool. The graph captures it weakly.

(b) You treat decree announcements as decree implementations. A vendor reading the graph can mistake "100 AI implementations targeted by 2026" for "100 AI implementations underway." This is a systemic framing risk in the pitch language across initiatives.

(c) You overweight the diaspora bridge as a tactical channel. Diaspora-bridge introductions can warm a contact, but the actual contract decision still happens through formal procurement. Saying "the diaspora intro is the unlock" is true for contact warmth but not for procurement outcome. The graph leans into the diaspora hook arguably harder than the data supports.

(d) You inflate Russian/CIS fit scores. The auditor (this report) confirms this — median 7.6 against a justifiable median of ~6.0. A hostile critic would say the entire pipeline is therefore mis-prioritized in favor of Russian-language UX initiatives over national-language initiatives.

(e) You underweight execution risk. Speed-to-contract is the highest-weight axis (25%) in the scoring rubric. Execution risk after contract — implementation pace, procurement bureaucracy, payment delay — is captured only loosely in the risk_register. Both UZ and KG state procurement is famously slow at the disbursement stage; this is not separately scored.

(f) You undercount competition from Russian incumbents. Yandex, Sber GigaChat, MTS AI, and SBERtech are described as "sanctions-shy" or "sanctions-exposed" in initiatives, but in practice these vendors continue to operate in CA via local partners and re-branded affiliates. Sanctions exposure is not a clean exit signal.

(g) You depend on single-language sources for many KG records. kaktus.media is cited heavily for KG. While respected, single-outlet over-reliance is a quality risk. Russian-language but Kyrgyz-domain (24.kg, akipress.org) and Kyrgyz-language native (azattyk.org Kyrgyz) cross-checks are inconsistent.

(h) You name people who may have moved. Tier-1 records are dated 2026-05-03 / 2026-05-04 (audit-recent), but role tenure currency for ministers and donor TTLs is not separately verified at audit date for most records. The "stale entries" risk is real for any record more than 60 days old.

(i) You omit defense and internal security entirely. This is a deliberate omission, not an oversight, but a hostile critic would still flag it: a vendor working in dual-use AI doesn't get a complete picture without it.

(j) You assume institutional stability that doesn't hold for KG. With Mintsifry liquidated approximately 30 April 2026 and UDP regs not due until ~31 May 2026, ~18 KG records are at structural risk of stale role assignments. A hostile critic would say "this graph is half-stale on the day of publication."

5. The single biggest known unknown

The institutional shape of the post-April-2026 Kyrgyzstan UDP digital department.

This is the single biggest known unknown for three converging reasons. First, the Kyrgyz Ministry of Digital Development was liquidated as a standalone body on approximately 30 April 2026, with all functions — digitalization policy, state digital services, telecom regulation, cybersecurity, archives, satellite tech, AI, and innovation — transferred to the Presidential Administration's Department of Affairs (Управление делами президента, UDP). Second, the structural regulations for the new UDP digital functions are not due to be published until approximately 31 May 2026. Third, every Kyrgyz record in the knowledge graph that depends on Mintsifry-era role assignments — at least 18 of the 40 KG people records, all 8 KG donor programs naming Mintsifry counterparts, and 11 of the 25 KG trends — is at structural risk of stale role data. As of audit date (2026-05-04), the public record contains: (a) the press announcement of liquidation in 10+ outlets; (b) plausible inference that key personnel (Sultanov, Asanbekov, Koichubaev, Isakov-Tunduk) transitioned to UDP roles; (c) absence of authoritative regulation defining what those roles formally are. This means: every Tier-A KG initiative is operating on a 4-8 week uncertainty window. A vendor pursuing KG opportunities in May 2026 is choosing between two bad options: (i) wait until regulations publish before acting (lose the criteria-shaping window); (ii) act now on inferred relationships and risk that the formal authority lies elsewhere when contracts are awarded. The audit recommendation is to maintain Tier-A scoring on KG initiatives only where the underlying decree authority and donor pathway survive the institutional transition (e.g., decrees from 2024-2025 still in force, donor projects with active counterparts beyond the dissolved ministry); and to demote initiatives where the operational counterpart was Mintsifry-specific. This is reflected in initiative_tier_updates.json.

6. Updated Sonar Pro re-verification findings (2026-05-04)

The audit consumed 8 paid Sonar Pro calls ($0.0217 of $20 budget). The findings revealed structural identity errors in the knowledge graph that the auditor must report transparently:

6.1 Wrong CEO of IT Park Uzbekistan

The knowledge graph names Firdavs Abdullayev as the "MOST ACCESSIBLE FOR FIRST CONTACT" Tier-1 contact for AI vendors entering UZ. Sonar Pro confirms the current CEO is Azamat Karamatov, with Farkhod Ibragimov as Chairman of Supervisory Board. Abdullayev was appointed CEO by Decree PF-60 of 7 February 2020 and was likely in role through ~early 2025; the transition to Karamatov was missed by the people-intelligence agent. This breaks the entire "GITEX intro through IT Park" warm-intro pathway as currently described in the deck and CRM.

6.2 Wrong last Minister of Digital Development of Kyrgyzstan

The knowledge graph names Talant Sultanov as Minister of Digital Development 2022-April 2026, then UDP Senior Advisor. Sonar Pro confirms the actual last minister before liquidation was Azamat Zhamangulov / Zhamankulov / Jamangulov, who chaired the Fifth Meeting of SCO ICT Agencies in Bishkek on 27 April 2026. Sultanov's recent (2025) public roles per Sonar Pro are: Policy Advocacy Advisor at Global Digital Inclusion Partnership, Chair of Internet Society Kyrgyz Chapter (ISOC.KG), UN IGF MAG member. The 2022-2026 ministerial appointment is NOT confirmed by Sonar Pro. The UDP transition is NOT confirmed.

6.3 Wrong head of UDP Digital Department

The knowledge graph names Adilbek Asanbekov as Head of Digital Transformation Dept., UDP. Sonar Pro confirms the actual head is Azamat Burzhuev (Head of Digital Development Department of Presidential Administration of the Kyrgyz Republic). The "Adilbek Asanbekov" identity is NOT verifiable in any KG government context per Sonar Pro — possibly fabricated, possibly a name confusion.

6.4 Wrong Minister of Health of Kyrgyzstan

The knowledge graph names Alymkadyr Beishenaliev as current Minister of Health KG. Sonar Pro confirms the actual current MoH is Damirbek Osmonov, appointed 26 February 2026 (replacing Kanibek Dosmambetov, who replaced Beishenaliev sometime earlier). Beishenaliev's earlier 2020-2023 tenure ended amid scandal; the chain of subsequent appointments was missed.

6.5 What this means

Four of approximately twelve operationally-critical Tier-1 KG/UZ contact identities are wrong as of audit date. This is not a small error rate. It indicates that the people-intelligence Wave 3 agent relied substantially on prior model knowledge or stale source data and did not consistently re-verify against current government rosters. Future runs MUST require Sonar Pro / live-search verification for every Tier-1 individual at last_verified_date stamping. The knowledge graph as of 2026-05-04 should NOT be used to drive direct outreach without first cross-checking the four named replacements:

These should be added as new Person records with priority before any pitch artifacts ship.

Coda: what this means for the user

Read this section before reading the bullishly-toned initiatives.json or the deck. The knowledge graph is structurally complete enough to support investment decisions in UZ and KG digital opportunities — but only if the user reads it as a hypothesis to be tested, not a fact base to be quoted. The HIGH and MEDIUM severity items in corrections.json are the things to fix or check before the deliverable is shipped. The known unknowns above are the things to acknowledge openly when the deliverable is shipped — every meeting with a UZ/KG official should start with "I know your institutional structure may have changed since this brief was drafted; please correct anything that's stale."

The harness improves through use. Every error becomes a constraint. The next pass should:

• Re-baseline KG records after 31 May 2026 UDP regulation publication.
• Resolve the UZ-LAW-2026-1125 question definitively.
• Run Sonar Pro on the 8 highest-leverage Tier-1 individuals to upgrade or downgrade LinkedIn URLs.
• Add a defense / internal security gap acknowledgment to every relevant initiative pitch.
• Recompute russian_cis_fit scores with a stricter rubric.

That, plus quarterly re-verification of decree status and donor disbursement, would close the meaningful audit-discovered gaps.

(Word count target: ≥1500 — this section is approximately 2200 words.)

(End of honesty_section.md)

@misc{valuev26cab2g,
  author = {Valuev, Alexandr},
  title  = {Central Asia B2G Intelligence: Honesty: what we did not find},
  year   = {2026},
  url    = {https://avaluev.github.io/ca-b2g-research/honesty/},
  note   = {Data vintage 2026-05, Apache 2.0}
}