Project repeats · 2027

Similar line items within NEP and within HGAB 3rd reading

This screen compares NEP with other NEP occurrences, and HGAB 3rd reading with other HGAB 3rd-reading occurrences. It complements the historical GAA screen. Similar titles are candidates for scope review, not established duplicate work. Pair counts and distinct allocation occurrences have different denominators.

FY2027 NEP · first 50 review pairs

Explicit overlap comes first, followed by identical full titles, then the larger individual allocation. All candidate pairs are available in the downloadable JSON. The CSV exports the displayed pairs.

FY2027 HGAB 3rd · first 50 review pairs

Explicit overlap comes first, followed by identical full titles, then the larger individual allocation. All candidate pairs are available in the downloadable JSON. The CSV exports the displayed pairs.

Methodology and exclusions

Compare distinct allocation occurrence IDs within each stage only. FAPs and untrusted NEP primary titles excluded. Same normalized official PAP and either qualified municipality/province or same implementing office required. At least three shared distinctive nonnumeric title tokens required. Same base title after trailing geographic metadata removal, or ≥0.94 SequenceMatcher similarity with qualified municipality/province AND identical distinctive token sets, produces a candidate. Changed route-endpoint or other distinctive scope tokens veto even when text similarity is high. Different explicit bridge numbers identify different structures and veto a pair, even on the same bypass road. Direction, qualified-location and tranche conflicts veto. Explicit disjoint ranges veto. Printed range endpoints are assumed to be ordered min/max. Mixed bare/plus ranges retain an unknown endpoint: a known maximum at or below another range’s known minimum excludes positive overlap, without resolving length. Other mixed ranges remain incomplete; possible overlap is not certified overlap. No amount-equality requirement. Missing chainage never proves scope overlap.

• Matching allocation occurrences does not establish duplicate contracts, duplicate payment, fraud or authorship. Different engineering components and legitimate amendments can share titles.
• Office-only agreement is weaker than site evidence. Unresolved locations are never inferred from similar names.
• Pair counts are not project counts: three linked records produce three pairs. Distinct occurrence counts are shown separately; amounts are not summed as loss.
• Exact official PAP blocking, trusted-title requirements and conservative location rules miss possible cross-program, paraphrased or unresolved duplicates. No flag is not clearance.
• Tranche differences are excluded from this conservative queue, not certified as legitimate. Missing chainage requires source and engineering-scope review.
• NEP baseline includes source records outside the API. Its documented source-title confidence limits remain applicable.

Download every candidate pair, original titles, source IDs, amounts and input hashes

Of the NEP repeat candidates, how many survived into HGAB?

The 737 headline is a NEP-level comparison with old enacted GAA. Tracing the same proposal codes into HGAB 3rd reading gives 2263 clearly retained comparison pairs, representing 1370 distinct proposals. Potential links remain separate.

A proposal matched to two historical years counts twice in the pair column but once in the proposal column. Clear retention means a selected, scope-compatible normalized-title link in our current parser, not a certified work-scope audit. A potential retained link is a provisional selected counterpart. Missing links do not establish that HGAB removed the project.

Export all 686 proposal traces CSV · Full pair and proposal evidence JSON

For retained proposals, investigate whether old appropriations and HGAB fund the same component. For unmatched proposals, check renaming, segmentation, parser fragments and original pages before claiming removal. Amount changes on retained lines are allocation adjustments, not automatically insertion or removal.

Exact ₱5M road and flood allocations cluster in the same offices

35 offices have both categories, compared with 21.7 expected under independent assignment. FY2027 NEP: 96 of 199 DEOs have exact ₱5M access-road/bridge entries and 45 have exact ₱5M flood entries. This is office-level co-occurrence, not evidence of repeated projects or overpriced work.

Exploratory statistical evidence

District offices with exact ₱5M entries in both categories

Counts are budget entries, not independently verified distinct projects. The listed amount counts each included entry once; it is not a loss estimate. All offices satisfying the same rule are included.

Methodology and limits

The road category is BIP access roads/bridges leading to major or strategic public buildings/facilities. Flood categories are construction/maintenance of flood-mitigation structures and drainage systems, and construction/rehabilitation of flood-mitigation facilities within major river basins and principal rivers. Exact ₱5M means a printed allocation of ₱5,000,000, with no rounding band.

The hypergeometric test fixes the number of road and flood offices among the year’s DEOs, then evaluates the probability of at least the observed overlap under independent assignment. The regional permutation test keeps road-office membership fixed and reshuffles flood membership within regions, preserving regional flood-office counts. The odds-ratio interval is an approximate log-Wald interval.

The amount threshold and categories were selected after noticing the pattern. These exploratory p-values and confidence intervals are unadjusted for that selection or multiple comparisons. Office workload, project mix, engineering need and standard designs can explain clustering. Small p-values reject the stated assignment model; they do not establish coordination, procurement splitting, overpricing or intent.

Next steps: pre-specify controls for work type, office workload and historical prices, then inspect quantities and engineering specifications. Group-level clustering is not a finding about any representative or signatory.

Source rows and statistical methods · JSON · CSV buttons export every row of their table, including pages not currently shown.