 RE:GOV regov.app NBI · nbi.gov.ph Migration report |
Migration report National Bureau of InvestigationAutomated assessment and rebuild of nbi.gov.ph on the BetterGov Kapwa design standard.
|
Migration report · nbi.gov.phRE:GOV regov.app NBI · nbi.gov.ph Migration report |
Migration report National Bureau of InvestigationAutomated assessment and rebuild of nbi.gov.ph on the BetterGov Kapwa design standard.
|
The agent crawled nbi.gov.ph, kept 30 of 30 pages, and rebuilt them as 89 structured content blocks. It recorded 31 accessibility findings on 25 pages, corrected in the rebuild, and 51 items on 30 pages that need a person to confirm before any of this is published.
RE:GOV stops a run when a site’s robots.txt shuts crawlers out or its terms forbid automated access or republishing, and site owners can opt out at any time. How the crawler works.
| Measure | Original | Rebuild | Change |
|---|---|---|---|
| Largest content visible | 1.4 s | 380 ms | 3.6× faster |
| Fully loaded | 1.5 s | 387 ms | 4.0× faster |
| Page weight | 17.6 MB | 410 KB | 44× lighter |
| Requests | 71 | 33 | 2.2× fewer |
| WCAG violations (axe-core) | 4 rules · 37 elements | 0 rules · 0 elements | none left |
Each homepage was opened in a fresh browser with no cache at 1280 × 900, twice, keeping the faster time. Both were measured from the RE:GOV server on October 28, 2026; the rebuild is served through regov.app. Speed on a phone or a slow connection will differ, but the gap in weight and requests holds.
Muse Spark looked at the original homepage with its accessibility findings and picked the ones you can see; SAM 3.1 found each element on the screenshot. The rebuilt site replaces every one of them.
After rebuilding, the agent checked its own work: Muse Spark read the homepage and the 9 pages with the most content (text, and screenshot where available) next to the rebuilt page, and listed anything lost and any statement the original does not support. Rewording is expected; changed facts are not.
Faithful: National Bureau of Investigation (98%) · 521: Web server is down (98%) · Web server is down (100%) · Web server is down (100%) · 521: Web server is down (100%) · Web server is down (100%) · Web server is down (100%) · 521: Web server is down (100%) · 521: Web server is down (100%) · 521: Web server is down (95%)
A model checking a model catches most dropped or invented facts, not all. Pages flagged here go to the agency’s editor first.
Each accessibility finding on the original site is tagged with the WCAG 2.2 success criterion it breaks. The original pages were also scanned with axe-core during the crawl. Every rebuilt page was then tested with axe-core 4.14.0 against the WCAG 2.0, 2.1 and 2.2 Level A and AA rules.
| Criterion | Level | Original site | Rebuild |
|---|---|---|---|
| 1.1.1 Non-text Content | A | 2 findings · 22 elements (axe) | Passes automated check |
| 1.3.1 Info and Relationships | A | 6 findings | Passes automated check |
| 1.4.1 Use of Color | A | 5 findings · 58 elements (axe) | Passes automated check |
| 1.4.3 Contrast (Minimum) | AA | 95 elements (axe) | Passes automated check |
| 1.4.4 Resize Text | AA | 1 elements (axe) | Passes automated check |
| 2.2.2 Pause, Stop, Hide | A | 1 finding | Needs manual check |
| 2.4.4 Link Purpose (In Context) | A | 35 elements (axe) | Passes automated check |
| 2.4.6 Headings and Labels | AA | 2 findings | Needs manual check |
| 3.3.2 Labels or Instructions | A | 3 findings | Passes automated check |
| 4.1.2 Name, Role, Value | A | 3 findings · 35 elements (axe) | Passes automated check |
Automated tools catch roughly a third of WCAG issues. Criteria marked “Needs manual check” (for example keyboard use, focus order, captions and meaningful link text) need a person with a keyboard and a screen reader before the agency can claim conformance. 16 findings are good practice without a specific WCAG criterion.
Noun-phrase prompts (logo, search box, map, infographic, photograph) on the original first screen, before any page was rewritten. The seal was cut out and reused as the new site’s brand mark.
| Region | Confidence | Box (px) |
|---|---|---|
| logo | 88% | 71 × 64 |
| logo | 92% | 61 × 83 |
| logo | 94% | 87 × 114 |
| logo | 53% | 12 × 16 |
| logo | 64% | 12 × 17 |
| search box | 87% | 248 × 37 |
| photograph | 61% | 294 × 184 |
| photograph | 62% | 331 × 244 |
| photograph |
Cut from the SAM 3.1 logo mask. Used in the header, footer and social previews of the rebuilt site.
The old site’s photos stay where they carry facts (officials, events). New artwork is generated only for decoration, from a brief Muse Spark wrote from the site plan: “Morning light bathes the NBI headquarters on U.N. Avenue in Ermita, Manila, Philippine flag flying high on a pole, near Luneta Park's tall palms and Manila Bay skyline. In the foreground diverse Filipinos – a young graduate, an OFW mother with documents, an elderly man – queue orderly at glass clearance booths for fingerprint biometrics, while plainclothes investigators in barong examine a forensic evidence kit and microscope table, red-white-blue-yellow bunting accents brightening the civic plaza, proud and orderly, no text, logos, seals or names.” Muse Image is told never to draw text, seals or logos. The artwork is decoration only (screen readers skip it), and every page of the rebuilt site says it is an unofficial, AI-generated concept.
Issues Muse Spark observed on the original pages. The rebuilt site addresses them with semantic headings, real text instead of images of text, labelled controls and WCAG AA contrast.
| Category | Count | Share |
|---|---|---|
| Images and media | 13 | 42% |
| Heading structure | 6 | 19% |
| Forms and labels | 5 | 16% |
| Other | 5 | 16% |
| Structure and semantics | 2 | 6% |
Content the agent migrated but could not verify, and gaps it could not fill. Each item should be checked by the agency before anything here is treated as current.
| Category | Count | Share |
|---|---|---|
| Other | 29 | 57% |
| Missing or blank content | 19 | 37% |
| Time-sensitive, verify | 2 | 4% |
| Dropped as site chrome | 1 | 2% |
This site has used 249,522 Muse Spark tokens in 64 calls, about $0.56. Cached results cost nothing, so reruns and repeat visitors are free.
| Step | Models | Calls | Tokens |
|---|---|---|---|
| Building the site | |||
| Understand pages | muse-spark-1.3 | 30 | 120,682 |
| Header artwork | muse-spark-1.3, muse-image-1.0 | 3 | 75,813 |
| Self-check | muse-spark-1.3 | 10 | 36,291 |
| Plan the site | muse-spark-1.3 | 1 | 8,185 |
| Evidence | muse-spark-1.3, sam-3.1 | 14 | 5,784 |
| Design check | muse-spark-1.3 | 1 | 2,767 |
| Find the seal | sam-3.1 | 5 | – |
Tokens as reported by the API. SAM 3.1 and Muse Image calls are counted, not tokenized. The dollar figure is an estimate from the average early runs; the Meta developer dashboard has the exact charge.
Prepared automatically by RE:GOV. The rebuilt site is an unofficial, AI-generated concept and is not affiliated with or endorsed by NBI. Findings are machine observations and may be incomplete; verify against the official website before acting on them.
Every site RE:GOV produces is an unofficial, AI-generated concept. It is not operated by, endorsed by, or affiliated with any government agency, local government unit, university, hospital or other public body, and it is not a substitute for their official websites. For services, transactions, forms and current information, always use the official website.
Content is migrated automatically from publicly available pages, or from Wayback Machine captures when a site blocks crawlers, and may be incomplete, outdated or wrong. Illustrations are AI-generated, not photographs. Hotline numbers come from BetterGov.ph’s public dataset and the source site; verify before relying on them. Weather is from Open-Meteo and is not a PAGASA advisory. Every rebuilt page links back to its original. Agencies may request changes or removal through GitHub issues.
© 2026 Ruther Bergonia · RE:GOV is an independent civic-tech project, not a government service. · Privacy · Terms
Muse Spark 1.3 · SAM 3.1 · Muse Image · Kapwa by BetterGov.ph · WikiBayan
| 51% |
| 160 × 195 |
| photograph | 59% | 212 × 184 |
| photograph | 57% | 337 × 244 |