Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects

NCPAGkilatis: A Policy Note series from the UP NCPAG community POLICY NOTE NO. 003-102725 TRACING CORRUPTION BREADCRUMBS: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects After expressing his deep concern over questionable flood control projects during his 4th State of the Nation Address (SONA) on July 28, 2025, President Ferdinand Marcos Jr. “Bongbong” Marcos, Jr. (PBBM) has directed the Department of Public Works and Highways (DPWH) to submit a comprehensive inventory of all flood control projects initiated or completed from 2022 to the present. He has also created via Executive Order No. 94 an Independent Commission for Infrastructure (ICI) with the main task of investigating “anomalies, irregularities, and misuse of funds” relating to flood control projects in the government. By October 9, 2025, DPWH Sec. Vince Dizon reported having discovered 421 ghost projects out of more than validated 8,000 infrastructure projects. The parallel investigations being undertaken by the DPWH and ICI, alongside the earlier in-aid-oflegislation hearings conducted by the Senate and House of Representatives, have shown how the sophisticated schemes of deception and collusion among private contractors and government officials have made detecting anomalies difficult and exacting accountability even more arduous. In large-scale corruption scandals, timeliness is imperative. The sheer depth and breadth of the problem could overwhelm investigative capacity, undermine momentum, obscure evidence, blunt reform, and ultimately diminish the prospect for justice and accountability. Such is the challenge before us now. This 3rd installment of NCPAGkilatis, a Policy Note series facilitated by the UP NCPAG’s GRIT Labs, presents the results of an AI-enabled approach to detecting potentially anomalous flood control projects. Using both supervised and unsupervised machine learning techniques, combined with geospatial analysis, we were able to identify not only ghost projects, but also other permutations and trends that are not easily detectable and discernible. We hope that this can help the government expedite its investigation in a manner that is data-driven, rules-based, and free of partisanship, which have been our Achilles heel in past corruption and plunder cases. This issue of NCPAGKilatis (Policy Note No. 003-102825) is co-authored by Kristoffer Berse, Kim Robert De Leon, Lloyd Marcelo, July Marañon, Ramon Christopher Caballero, Francis Miguel Garcia, Noriel Christopher Tiglao, Olivia Cassandra Bondad, Mape Estellena, Micah Paula Milante, and John Coby Cabuhat. The views, thoughts, and informed opinions expressed in NCPAGkilatis are solely that of the authors and do not represent the official position of the University of the Philippines National College of Public Administration and Governance. 1 First, what we did Using publicly accessible data from DPWH, Sumbong sa Pangulo, Project NOAH, and PCO websites, three approaches were deployed in the course of our analysis. First, we developed an AI model called Flood-Control AI Labelled Checker (FAIL Checker), which transforms established principles under our procurement and audit laws and regulations (R.A. 9184/R.A. 12009, PD 1445, COA MC 2012-003, etc.) into quantifiable indicators that allow the detection of red flags through data analytics. The FAIL Checker applies rule-based supervised artificial intelligence logic that systematically evaluates every project in the flood control dataset using measurable, legally grounded indicators. Based on the above, we have identified six (6) data-driven categories that mirror common audit findings, as follows: 1. Green Flag Projects - These are projects that show no apparent anomalies based on all measurable indicators. However, it represents only a preliminary presumption of regularity pending validation of an appropriate authority. 2. Chop-Chop Projects - A project is classified as a Chop-chop project when it appears to have been fragmented or duplicated within the same geographic and fiscal boundaries signifying possible artificial segmentation. The FAIL Checker model evaluates proximity, phase references, and textual similarity to detect possible artificial segmentation of a single development site into multiple contract packages. Within the same location and Fiscal Year, any pair of projects are flagged if their coordinates are identical, if the distance between them is less than or equal to 500 meters, or if they are within one kilometer and contain phase-related terms or highly similar project descriptions. Additionally, projects within 500 meters of each other are further grouped into clusters for a deep dive analysis. 3. Doppelganger Projects - A project is classified as a Doppelganger project when duplication across contracts is suspected. This may occur under any of the following situations: Near-identical cost and project description within the same area and fiscal year, where the contract costs differ by no more than two percent (2%); OR Identical contract cost, location and rounded coordinates implemented within one infrastructure year; OR Same contractor and same District Engineering Office with identical start and completion dates. 4. Potentially Ghost Projects - A project is classified as Potentially Ghost when the recorded implementation period is implausibly short or chronologically inconsistent. This may occur under any of the following situations: The completion date precedes the start date; OR The actual duration is less than thirty days (30 days) and the contractor manages more than five active projects within the same fiscal year; OR The actual duration is less than or equal thirty days (30 days); OR The actual duration is less than seventy-five percent (75%) of the contract duration. 5. Siyam-siyam Projects - A project is classified as Siyam-siyam when it demonstrates prolonged or rolling implementation. These are projects that persist beyond reasonable timeframes without closure. 6. Potentially Kamote Projects - Flood-control projects are considered potentially kamote if they are located in areas with zero to low flood risks. These are detected by a spatio-temporal layering of DPWH and NOAH data. 2 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Second, we took a deep dive into potentially repetitive projects over the same area and period by employing spatial clustering (DBSCAN) with the Haversine formula for distance computation to identify simultaneous and geographically contiguous projects. The analysis identified clusters of at least two flood control projects that are within a 500-meter radius of a cluster’s center. Lastly, to determine whether or not the flood control projects correspond to areas where infrastructure interventions are needed most, we overlaid project and contractor data with the flood hazard data from Project NOAH. This geospatial layering, tested under different machine learning models, allows us to quickly and reliably scan and visualize the extent of how a given solution (i.e., flood control project) addresses a particular need (i.e., flood hazard). A mismatch in this regard is tantamount to wasting public funds, even in the absence of corruption. District Engineering Green Flag Rate Total Projects Green Flag Projects Total Contract Cost Green Flag Contract Cost Iligan City District Engineering Office 66.67% 6 4 ₱346,149,111.20 ₱215,185,300.90 Palawan 1st District Engineering Office 50% 18 9 ₱564,325,076.50 ₱158,158,139.10 Cebu 2nd District Engineering Office 50% 12 6 ₱388,756,245.10 ₱186,011,220.70 3 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Out of 198, 30 DEOs were found to have no green flags at all. Collectively, they are responsible for 1,007 projects from 2018 to 2025, amounting to Php 56,041,244,751. Among them, the DPWH North Manila District Engineering Office has the highest number of flood control projects at 185 projects. Other glaringly high-volume DEOS with no Green Flag projects are Metro Manila 3rd District Engineering Office, with 141 projects and Quezon 2nd District Engineering Office with 59 projects. Table. 1. DEOs with more than half of projects tagged as green flags by the FAIL Checker model Green Flag Projects Our FAIL Checker analysis revealed that only 923 out of 9,826 flood control projects (9.4%) do not have potential irregularities (i.e. Green Flag Projects). Conversely, this means that 90.6% of the projects demonstrated deviations from standard procedures laid out in Republic Act No. 9184 as amended by RA 12009 and related COA circulars. No DPWH District Engineering Office (DEO) was tagged as completely free of potentially anomalous projects. Nevertheless, a substantial majority of DEOs (168 of 198 or 85%) were found to have at least one green flag project, with 3 DEOs having at least half of their projects with no potential anomaly (Table 1). The “greenness” or lack thereof of DEOs throughout the country is shown below (Figure 1). The map presents the spatial distribution of compliance across provinces based on the proportion of Green Flag projects, or those without detected irregularities. Provinces shaded in deep red exhibit higher concentrations of projects with potential anomalies, while yellow and orange tones signify moderate levels, and green identifies provinces where projects are largely tagged by the FAIL Checker as green flags. The pattern unravels the extensive clusters of flagged projects observed in most of Mindanao and many parts of Central Luzon and Northern Luzon. This geographic imbalance underscores the importance of institutionalizing geospatial monitoring, data-driven validation, and risk-based auditing in infrastructure governance. 4 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects A scatter plot analysis was further conducted to unravel DEOs with extreme values. Figure 2 shows that most DEOs with higher Green Flag rates handle a lesser volume of projects (i.e., 50 projects or less), such as Iligan City District Engineering Office with 66.67 percent; Palawan 1st District Engineering Office, 50 percent; and Cebu 2nd District Engineering Office, 50 percent. In contrast, larger DEOs with higher project volumes showed significantly lower compliance rates, including Bulacan 1st District Engineering Office with 452 projects and only 2.88 percent Green Flag rate; Metro Manila 1st District Engineering Office with 186 projects and only 1.30 percent Green Flag rate; and Bulacan 2nd District Engineering Office with 190 projects and only 3.69 Green Flag rate. Fig. 1. Provincial Distribution of “Greenness” of Flood Control Projects among DPWH District Engineering Offices Fig. 2. Relationship Between Green Flag Rate and Total Project Count per DPWH District Engineering Office Province City / Municipality Contract ID Project Description Contractor Start Date Actual Completion Date Contract Cost Bulacan Calumpit 23CC012 3 Construction of Flood Mitigation Structures along Barangay Frances, Calumpit, Bulacan Wawao Builders and Construction Feb 13, 2023 Mar 7, 2023 ₱77,199,999.65 Bulacan Hagonoy 23CC010 2 Construction of Flood Mitigation Structures along Barangay San Agustin, Hagonoy, Bulacan Wawao Builders and Construction Feb 13, 2023 Mar 7, 2023 ₱77,199,989.04 National Capital Region Quezon City 22OF005 8 Rehabilitation of Flood Control Structure, Barangay Sta. Monica, Quezon City E. L. Torres Construction and Trading Inc. Feb 14, 2022 Jun 11, 2022 ₱14,259,000.00 5 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Potentially Ghost Projects It must be emphasized that a green flag classification does not mean a full clearance from possible future findings of irregularity. These projects still have to be validated to uphold a policy of zero tolerance for corruption and inefficiency. The FAIL Checker tool is designed to guide investigators towards which projects should be prioritized for review, serving as a data-based map rather than a certificate of compliance. A green flag is not an assurance of integrity approval, or exemption from scrutiny, it only indicates that, based on available data, no anomaly has been detected for now based on the defined indicators. True accountability demands continuous vigilance, as a green flag remains only a temporary presumption of regularity pending verification. Meaning, hindi ka pa safe! Table 2 identifies projects tagged by the FAIL Checker as a potentially ghost project which were also flagged by external sources. One of the projects tagged as a potentially ghost project was the construction of flood mitigation structures in Barangay Frances, Calumpit, Bulacan which cost ₱77-million. The project, proposed by the DPWH under the 2023 NEP and constructed by Wawao Builders Corporation, was tagged as a completed project despite still being incomplete upon inspection of President Marcos Jr. The model also tagged the same project as a chop-chop project, hinting at the possibility of other projects within the same area. The classification of projects as ghosts is not foreign to the general public with the revelations in the past months. In its common usage, these are projects which are non-existent, plain and simple. However, identifying these ghost projects requires inspection of each of the thousands of projects. Hence, preliminarily identifying what may be potential ghost projects may prove helpful. As mentioned above, ghost projects are identified as those with “abnormally short durations and handled by contractors with excessive project loads.” As such, this is a risk signal rather than a finding, and must be validated on the ground. The FAIL Checker model tags a project as potentially ghost if its execution period is implausibly short or shows a time mismatch between its start and completion dates. Ghost projects summed to 2,934 projects which is 29.9 percent of the total number of flood control projects. Of these, 680 projects were tagged as solely a ghost project, i.e., without additional characterization. As aforementioned, projects may be tagged under two classifications simultaneously. To illustrate, 568 ghost projects may actually turn out as chop-chop projects upon validation, while 338 ghost projects might be doppelganger projects in the end. It can be noted that the DPWH has already identified “at least 421 ghost or non-existent” projects . 1 Manabat, J. (2025, October 9). DPWH uncovers 421 ghost projects out of 8,000 inspected. ABS-CBN News. https://www.abscbn.com/news/nation/2025/10/9/dpwh-uncovers-421-ghost-projects-out-of-8-000-inspected-1030 1 Table 2. Example of projects reported in the news and/or local government units as potentially anomalous and tagged as ‘Ghost Project’ by the FAIL Checker Model District Engineering Potentially Ghost Rate Total Projects Potentially Ghost Projects Total Contract Cost Potentially Ghost Contract Cost Bulacan 1st District Engineering Office 48.67% 452 220 ₱29,183,182,391.00 ₱14,728,359,989.00 Isabela 4th District Engineering Office 66.95% 118 79 ₱8,842,387,407.00 ₱6,252,744,923.00 Pangasinan 3rd District Engineering Office 62.83% 113 71 ₱5,058,138,854.00 ₱3,587,802,534.00 Cebu 7th District Engineering Office 40.00% 170 68 ₱12,663,827,857.00 ₱5,649,896,032.00 Metro Manila 1st District Engineering Office 28.70% 230 66 ₱12,352,162,536.00 ₱3,303,762,640.00 6 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Table 3 further reveals that some District Engineering Offices have alarmingly high numbers of projects flagged as potentially ghost, Bulacan 1st DEO tops the list with 220 out of 452 projects worth ₱14.73billiion, followed by Isabella 4th DEO and Pangasinan 3rd DEO with ₱6.25-billion and ₱3.59-billion respectively, Cebu 7th and Metro Manila 1st DEOs also report billions in potentially non-existent project costs, The presence of these cases across both urban and rural regions indicates that the issue of ghost projects is widespread and persistent, calling for immediate reforms in project validation, site inspection, and digital tracing to esure that public finds result in real infrastructure on the ground. Table 3. Top five District Engineering Offices with the highest number of projects tagged as Potentially Ghost by the FAIL Checker model Chop-chop Projects The identification of chop-chop projects relies on the clustering of projects within the same location. If coordinates or geographic names are imprecise, the grouping may be too broad or too narrow. The model tags a project as a chop-chop project if it appears to be fragmented or duplicated within the same geographic and fiscal boundaries. Based on the FAIL Check model, a substantial majority of the projects (6,290 projects) were tagged as potentially ‘Chop-chop Projects’, comprising 64 percent of the total projects. Of these, various projects were also tagged with another classification on top of it being a chop-chop project. For instance, of the chopchop projects, 1,611 of them are also potentially doppelganger projects, i.e., projects with similar or near identical contract costs, while 696 projects are additionally tagged as siyam-siyam projects. In Barangay San Agustin, Hagonoy, Bulacan, no flood control structure was found by the team of ABS-CBN News despite a supposed ₱77-million flood mitigation structure to be constructed by Wawao Builders Corporation . This project was also tagged as a potentially ghost project by the model. Lastly, the ₱14-million rehabilitation of a flood control structure in Barangay Sta. Monica in Quezon City was also tagged as a potentially ghost project. The findings of the QC Government Inspection Report cited this project as one of the 18 projects which reflected dissimilar project locations between the location indicated in the project title and the location based on the coordinates. These irregularities, deliberate or not, should sufficiently prompt suspicion as they may further reveal a more substantial problem. 2 Taguines, A. (2025, August 19). 'Where is it?' Local officials, residents unaware of P77-million flood control structure in Hagonoy, Bulacan. ABSCBN News. https://www.abs-cbn.com/news/regions/2025/8/19/local-officials-residents-unaware-of-p77-million-flood-control-structure-inhagonoy-0941?utm_source=dlvr.it&utm_medium=facebook&fbclid=IwY2xjawMQ3NFleHRuA2FlbQIxMQABHgCsvIczhwi5nLoWaZ2asnZa_txcPisOPP9tBNzvSu3sHNw6XdKVwI0_-ps_aem_RZ29SaWMm5oJtUjcVw28RQ 2 Province Municipality Contract ID Project Description Contractor Contract Cost La Union BAUANG (LA UNION) 24AF0108 Construction of Flood Control structure along Bauang River Basin (Package 1), Bauang, La Union Silverwolves Construction Corp ₱89,737,118.14 La Union BAUANG (LA UNION) 24AF0109 Construction of Flood Control structure along Bauang River Basin (Package 2), Bauang, La Union Silverwolves Construction Corp ₱89,737,280.52 La Union BAUANG (LA UNION) 24AF0110 Construction of Flood Control structure along Bauang River Basin (Package 3), Bauang, La Union Silverwolves Construction Corp ₱89,736,907.75 La Union BAUANG (LA UNION) 24AF0111 Construction of Flood Control structure along Bauang River Basin (Package 4), Bauang, La Union Silverwolves Construction Corp ₱89,736,445.72 La Union BAUANG (LA UNION) 24AF0112 Construction of Flood Control structure along Bauang River Basin (Package 5), Bauang, La Union Silverwolves Construction Corp ₱89,736,347.57 La Union BAUANG (LA UNION) 24AF0113 Construction of Flood Control structure along Bauang River Basin (Package 6), Bauang, La Union Silverwolves Construction Corp ₱89,737,362.73 La Union BAUANG (LA UNION) 24AF0114 Construction of Flood Control structure along Bauang River Basin (Package 7), Bauang, La Union Silverwolves Construction Corp ₱89,737,260.38 7 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects While the chopping of projects into packages or phases is not illegal or prohibited per se, it can be a telltale sign of steering clear from higher levels of accountability. This is implied in the recent recommendations of the ICI, petitioning for a lower threshold of contract amounts for the different levels of engineering offices. It is being proposed that the district engineering offices be only authorized to allow bids up to ₱75 million, instead of ₱150 million, citing the systemic corruption in the bidding process as the rationale . 3 Table 4 shows that some of the projects tagged by the model as a chop-chop project include the flood control structures along Bauang River Basin in Bauang, La Union . Said project was divided into seven packages, each of the packages similarly costing ₱89.7-million. All of the seven packages, which were not part of the original budget proposal under the 2024 NEP, were awarded to Silverwolves Construction Corp. The project reached media scrutiny upon the inspection of Baguio City Mayor and former Independent Commission for Infrastructure adviser Benjie Magalong, finding, among others, that the project was of poor quality. It can also be easily pinpointed that the contract cost for all of these projects are almost identical and thus additionally tagged under the model as a doppelganger project—raising two red flags in one project. 4 de Villa, K. (2025, October 10). ICI asks DPWH to reduce gov’t engineers’ control over biddings. Inquirer. https://newsinfo.inquirer.net/2122484/ici-asks-dpwh-to-reduce-govt-engineers-control-over-biddings 3 Mangaluz, J. (2025, September 16). In La Union, 'prop' pipes used in P179.4M ghost flood control project. Philstar. https://www.philstar.com/headlines/2025/09/16/2473260/la-union-prop-pipes-used-p1794m-ghost-flood-control-project 4 Table 4. Example of projects reported in the news and/or local government units as potentially anomalous and tagged as ‘Chop-chop Project’ by the FAIL Checker Model City/ Municipality Contract ID Project Description Contractor Contract Cost Quezon City 23OG0063 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XVIII), Quezon City MRB11 Construction Corporation ₱47,907,420.63 Quezon City 23OG0064 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XIX), Quezon City MRB11 Construction Corporation ₱48,963,617.70 Quezon City 23OG0066 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XX), Quezon City MRB11 Construction Corporation ₱48,963,617.70 Quezon City 23OG0065 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XXI), Quezon City Amethyst Horizon Builders and Gen. Contractor and Development Corp ₱48,938,414.23 Quezon City 23OG0067 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XXII), Quezon City Elite General Contractor and Development Corp. ₱48,948,822.01 Quezon City 23OG0068 Rehabilitation of Flood Mitigation Structure along San Juan River (Phase XXIII), Quezon City A. L. Salazar Construction, Inc. ₱48,904,450.12 8 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects The model was also able to identify the Topnotch Catalyst Builders and Beam Team Developer Specialist, Inc.’s ₱98.99-million project in Barangay Bambang, Bocaue, Bulacan as a chop-chop project. The Commission on Audit earlier flagged the joint venture between the two contractors for constructing the structure on a location different from that of the approved site . Moreover, an existing flood control structure was also found in the same site, separate from that of the joint venture. 5 Table 5. Example of projects reported by Quezon City local government units as potentially anomalous and tagged as ‘Chop-chop Project’ by the FAIL Checker Model Relatedly, a series of flood control structures in the Bauang River Basin was also funded in Naguilian, La Union. The project, which was also not part of the original budget proposal under the 2024 NEP, was divided into 10 packages with each package amounting to ₱94.5-million, except for package 10 with a cost of ₱82.40-million. The 10 projects were shared between three contractors: (i) MG Samidan Construction for Packages 1 to 3; (ii) Universal De Leon Construction and Aggregates for Packages 4 to 6; and (iii) Unimax Steel Structure and Construction Corp. for Packages 7 to 10. ABS-CBN News. (2025, September 26). COA submits fraud audit reports on Bulacan flood projects to ICI. https://www.abscbn.com/news/regions/2025/9/26/coa-submits-fraud-audit-reports-on-bulacan-flood-projects-to-ici-1426 5 Funding Year Phases Longitude Latitude Project IDs 2022 II and X 121.019381 14.618373 P00620450LZ - II P00637004LZ - X 2022 III and VIII 121.019617 14.618053 P00620451LZ - III P00637002LZ - VIII 2022 IV, V and IX 121.019873 14.67439 P00620452LZ - IV P00742396LZ - V P00637003LZ - IX 2022 VI and 6 121.020442 14.615012 P00637000LZ - VI P00621437LZ - “6” 2022 VII and 7 121.020928 14.614787 P00637001LZ - VII P00621438LZ - “7” 2022 XI and XIV 121.02038 14.614572 P00637005LZ - XI P00621584LZ - XIV 2023 XVIII, XIX, XX, XXI, XXII, and XXIII 121.0199 14.6117 P00727034LZ - XVII P00727035LZ - XIX P00727036LZXX P00727037LZ - XXI P00727038LZ - XXII P00727039LZ - XXIII 9 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Table 6. San Juan River Flood Mitigation development chopped into multiple phases located in Quezon City Only Phases I to V were found to have been originally proposed under the 2022 NEP and Phases XV to XVII in the 2023 NEP, while the rest of the phases only appeared in the GAA (Table 6). Notably, Phase III of the project was also awarded to Wawao Builders, amounting to ₱43.1-million. Meanwhile, Phases XVIII, XIX and XX were all awarded to MRBII Construction Corporation amounting to ₱145.8-million. with all projects simultaneously started on March 13, 2023 and completed on June 11, 2023. Further spatial analysis revealed 1,739 unique clusters comprising 5,751 projects, which means that more than half of all reported flood control projects have at least one other project located within 500 meters (Table 7). This spatial density is strongly indicative of project fragmentation or artificial subdivision, where a single development site may have been divided into multiple contracts. Moreover, the cluster distribution exposes geographic hotspots of excessive project concentration suggesting potential procurement circumvention and inefficient allocation of public funds within certain localities. Worse still, the cluster distribution indicates highly egregious areas with very high project concentrations. Projects chopped into phases may also be a fertile ground for doubts, employing “continuous phasing or repetitive engagement.” Table 5 shows Phase 18 to Phase 23 of the rehabilitation of flood mitigation structures along San Juan River in Quezon City which were completed in less than 6 months for each phase. It can also be noted that several of these projects exhibit similarities in the contract cost of the various phases. Despite belonging to different project phases, the contract cost all amounted to ₱47.9 or ₱48.9 million. Two of these phases even had identical contract costs down to the last centavo. Based on a recent Quezon City Government Report, this rehabilitation work has had more than 20 project phases. In fact, the same Report also highlighted that the flood mitigation project in San Juan River already reached 92 phases. Interestingly, most phases share exactly the same geo-tagged locations, as follows: District Engineering Office Siyam-siyam Rate Total Projects Siyam-siyam Projects Total Contract Cost Siyam-siyam Contract Cost La Union 2nd District Engineering Office 60.31% 131 79 ₱9,085,915,597 ₱5,601,983,436 Davao City District Engineering Office 71.56% 109 78 ₱5,945,664,660 ₱4,786,102,443 Metro Manila 1st District Engineering Office 31.74% 230 73 ₱12,352,162,536 ₱4,584,602,825 Ilocos Sur 2nd District Engineering Office 69.23% 104 72 ₱2,813,193,329 ₱2,143,400,740 North Manila District Engineering Office 31.35% 185 58 ₱12,768,889,763 ₱3,762,555,009 Number of Projects by Flood Hazard Zone Contract Costs by Flood Hazard Zone Flood Hazard Zone Metro Manila Bulacan Pampanga Metro Manila Bulacan Pampanga 0 (no data/outside zone) 218 (21.00%) 164 (24.77%) 104 (35.14%) ₱9,129,915,647.00 ₱10,212,068,756.00 ₱4,548,824,927.00 1 (low) 95 (9.15%) 32 (4.83%) 30 (10.14%) ₱3,545,822,061.00 ₱1,963,184,722.00 ₱1,799,173,971.00 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Table 15 shows that the La Union 2nd District Engineering Office recorded the highest number of projects tagged as Siyam-siyam, with 79 out of 131 projects totaling ₱5.6-billion. Davao City District Engineering Office followed closely with 78 projects worth ₱4.79-billion, while the Metro Manila 1st District Engineering Office ranked third with 73 projects amounting to ₱4.58-billion. Ilocos Sur 2nd and North Manila DEOs also appeared prominently, registering 72 and 58 projects, respectively. These outputs suggest that several projects have extended far beyond their intended timelines, indicating delays or prolonged implementation periods. The recurring appearance of major urban and provincial DEOs in this category highlights the importance of stronger project monitoring mechanisms and time-bound completion tracking to ensure that contract extensions do not become the norm in infrastructure delivery. 16 Flood control projects, in principle, should be anchored on a science-based hazard assessment. Yet, results from the spatial analysis of three provinces overlaying flood control project, contractor, and hazard data from 2018-2025 reveal that a notable portion of the flood control projects are located in low hazard areas. Table 16 shows that 95 projects in Metro Manila, 32 in Bulacan, and 30 in Pampanga are situated in areas with low flood hazard susceptibility. Potentially Kamote Projects Table 15. Top five District Engineering Offices with the highest number of projects tagged as Siyam-siyam by the FAIL Checker model. Table 16. Projects and Contract Costs by Flood Hazard Zone via Geospatial Layering (2018-2025) 2 (medium) 313 (30.15%) 96 (14.50%) 46 (15.54%) ₱15,901,358,819.00 ₱5,812,670,559.00 ₱2,421,315,766.00 3 (high) 412 (39.69%) 370 (55.89%) 116 (39.19%) ₱22,663,283,040.00 ₱25,426,640,614.00 ₱5,959,247,162.00 TOTAL 1038 (100%) 662 (100%) 296 (100%) ₱51,240,379,567.00 ₱43,414,564,652.00 ₱14,728,561,826.00 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects Projects tagged under the ‘no data/outside zone’ might still carry contract costs because they are possibly located in areas that are not covered by existing flood hazard maps or where data is unavailable. These projects may not directly inform flood-risk interpretation, but they help expose where data coverage ends and where flood control projects are being implemented outside the mapped hazard zones. To provide a snapshot, the figure below shows the distribution of flood control projects in Metro Manila, Bulacan, and Pampanga. With more than a hundred projects located in low hazard zones, what clearly emerges from the map is an illusion of protection that masks a misalignment of purpose and creates a facade of action. The result is a misdirection not only of funds but also of the public trust. 17 Figure 4. Location of flood control projects (black dots) relative to the 100-year flood hazard areas of Metro Manila, Bulacan, and Pampanga (from left to right). Flood hazard data from UP NOAH. The scarcity of green flag projects 1 2A converge of multiple red flags The FAIL Checker Model also detects projects with two or more anomaly characterizations, as the classifications themselves are not mutually exclusive. The implication is doubly troubling. It means that, for instance, there may be projects that are already chopped into smaller projects, thus potentially anomalous, and at the same time have been completed beyond its target completion date, which, in itself, incurs additional cost Key Points Only 9.4% (923 out of 9,826) of all flood-control projects qualified as Green Flag under the FAIL Checker model, meaning that nine of every ten exhibit at least one potential anomaly. This underscores the magnitude of governance gaps in infrastructure implementation and the urgent need for institutionalized, data-driven oversight. 3Non-science-based and misaligned infrastructure 4Institutional risk concentration Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects A geospatial case analysis in three areas (Bulacan, Metro Manila and Pampanga) show that a nonnegligible amount of flood-control projects are located in areas with low flood susceptibility, while some high-risk areas remain underserved. This mismatch warrants an expanded investigation in other areas and avoid the unscientific allocation of public resources moving forward. The concentration of more than half of flood control projects within certain clusters signals potential artificial subdivision of contracts within a small area and inflated project proliferation within specific development sites. Certain district engineering offices merge as consistent hotspots of overlapping contracts and repetitive contractor engagements. 18 5Systemic procurement and audit vulnerabilities Patterns observed in contract costs, project durations, and recurring contractor names expose weaknesses in compliance verification, post-audit reach, and inter-agency data integration. These systemic deficiencies enable the repetition of procurement cycles that escape real-time detection and erode the preventive intent of the laws. 6Imperative of predictive and proactive governance The findings affirm the necessity of predictive auditing where the risk detection precedes disbursement and project completion. The FAIL Checker demonstrates that combining AI, spatial analytics and statutory indicators can help public agencies transition from reactive investigation to preventive accountability ensuring that infrastructure spending is lawful and purposeful. Call to Action The challenge of corruption in flood-control and infrastructure projects must now be addressed not only through punitive post-facto measures, but by building anticipatory mechanisms grounded in law, data and technology. Republic Act No. 12009 also known as the New Government Procurement Act, together with Presidential Decree No. 1445 also known as Government Auditing Code of the Philippines, collectively enjoin public officers to ensure that every peso spent is supported by necessity, economy and transparency. The principles of accountability, competitiveness, and public trust are the cornerstones of fiscal governance. The duty to observe these principles extends not merely to bidding and awards but to the entire lifecycle of public expenditures through monitoring, evaluation and audit. Accordingly it is recommended that: The ICI, DPWH, COA, and the Office of the Ombudsman formally adopt data-driven supervision and investigation, consistent with their constitution and statutory mandates. AI-enabled systems such as the FAIL Checker or similar models should be institutionalized as part of an early-warning mechanism under the principles of proactive auditing in COA Circular 2012-003 and preventive oversight under the 1987 Constitution. 1 The DBM and COA to integrate algorithmic risk tagging in project evaluation and budget preparation to ensure that appropriations reflect not only absorptive capacity but also historical integrity performance of contractors and implementing units. Periodic recalibration through annual reviews and inclusion of new variables such as project length, variation orders or geospatial imagery can be mandated. This action satisfies both COA constitutional duty of continuous audit innovation and the ethical principle of algorithmic accountability under international open-government standards. 2 Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects 19 The DPWH and other agencies involved in infrastructure development to create, in collaboration with the Department of Information, Communication and Technology, a national data pipeline for public works metadata of contract costs, coordinates, timelines and physical progress which allows the continuous machine-assisted audit and to guide citizens’ participation in audit cycle. 3 The private sector and civil society organizations (CSOs), including the media, church, and schools, to make use of the results of AI models like FAIL Checker as they participate in the auditing and validation of projects. This could assist the government to fasttrack the current investigation and preempt possible anomalies in the future. 4 Academic, professional, and other technology groups within and outside government to continue to develop, test, and perform independent replications, issue public commentaries, and share proposals to improve AI-enabled tools. An open data and open science approach must be promoted to enhance transparency and transform proprietary tools into a public good. 5 Lastly, some caveats The FAIL Checker, clustering, and spatial analyses draw inferences only from the fields present in the consolidated flood control dataset and from deterministic rules that we encoded from audit practice. It is neither intended to prove wrongdoing nor replace field inspection. Coverage is limited to projects with records that can be linked by location, year, contractor, and cost. When any of these fields are missing or inconsistently encoded, the either date is miscoded or backfilled after the fact, the measure of delay will be unreliable. For instance, the team has found that eight (8) flood control projects of the Romblon District Engineering Office were tagged under the province of Surigao del Sur. This does not change, however, the validity of the results presented as the province category was not used in any of the project classifications or any other analyses. The findings as laid out above only reiterates a public intuition: that there are more anomalous projects to expose. We hope that by utilizing AI and other tools, the investigation of potentially anomalous projects can be fast-tracked. Corruption and plunder cases are time-sensitive and time-consuming, so any tool that could help identify what project sites need to be checked or validated on the ground should be welcomed. It goes without saying that the work of reform cannot stop at identifying anomalous or ghost projects. It must conform to the deeper design of a governance system that permits inefficiency, fragmentation and collusion to thrive under bureaucratic complexity. The real test of government is not in the number of projects it completes but in its ability to ensure that every peso produces a structure that protects lives rather than endangers them. Each potential anomaly reported in this Policy Note is more than a data point; it is an indictment of the institutional complacency that has turned public works into private ventures. When the rules of procurement are reduced to rituals of compliance, the law becomes ornamental and the data becomes silent. What the FAIL Checker, clustering, and spatial analyses reveal is that corruption leaves traces, small, consistent, and shaped by data. Each repeated contract, recycled description, or hidden project becomes a breadcrumb that leads investigators toward a larger story of neglect and deception. When these trails are followed, technology turns oversight from a blind pursuit into a guided map. It restores accountability to its rightful place where numbers do more than expose wrongdoing, they lead REFORM. The government must therefore institutionalize predictive oversight, automate transparency, and let the data speak where silence once prevailed. Only then can the Philippines rise above the flood, not only of water but of waste, deceit, and unkept promises, and build an infrastructure of trust that stands firm even in the strongest storms. Suggested citation for this Policy Note: Berse, K.B., De Leon, K.R., Marcelo, L., Maranon, J.A., Caballero, R., Garcia, F.M., Tiglao, N.C., Bondad, O.C., Estellena, M., Milante, M. P., and Cabuhat, J. C. (2025). “Tracing Corruption Breadcrumbs: Rapid Detection and Characterization of Potentially Anomalous Flood Control Projects” NCPAGkilatis Policy Note No. 003-102825. University of the Philippines National College of Public Administration and Governance, 24 October.