# Source validation log Every figure published on this site is traced back to a document here. This log records what was read, how, and when, so a claim can be re-checked independently rather than taken on trust. "First-hand" means the file itself was downloaded and parsed, or the page was read directly, not summarised from someone else's reporting. Last validated: 18 September 2026. --- ## 1. US DOL, PERM Disclosure Data FY2020 to FY2026 (first-hand) Eight workbooks across seven fiscal years, 797,465 real case rows once empty rows are excluded. **Correction, 21 September 2026.** Until this date the figure here was 775,173 and FY2024 was given as 92,258. Both were wrong. FY2024 is the ETA-9089 transition year and DOL publishes it as two files: the 155 column old form and a separate 135 column `PERM_Disclosure_Data_New_Form_FY2024_Q4`. Only the first was being read, so FY2024 was short by 22,292 cases, a quarter of that year. The two files were checked for overlap before merging: zero shared case numbers, and the formats differ (`A-23141-45393` against `G-200-23212-225785`), so they are distinct cases and not duplicates. The Bangladeshi figures are unaffected. The new form dropped COUNTRY_OF_CITIZENSHIP, so that file contains no identifiable Bangladeshi cases and the FY2024 counts of 1,612 / 477 / 883 / 242 stand. It does follow that a Bangladeshi who filed on the new form in FY2024 cannot be seen in this data at all, which is a limit of the source, not of the parsing. Every other year was re-derived on this date by a second parser written independently of the builder, and all six matched exactly, status counts included, as did the Bangladeshi counts for all five citizenship years. Parsed directly out of the workbook XML. Bangladeshi rows isolated on `COUNTRY_OF_CITIZENSHIP`, giving 1,612 cases. Everything below is counted from those rows: - Outcomes: 477 certified, 883 certified-expired, 242 denied, 10 withdrawn. - National comparison: 35,505 / 50,814 / 4,892 / 1,047 of 92,258. - Denial among decided cases: 15.1% Bangladesh against 5.4% all countries. - Decision time: median 392 days, p75 471, p90 627, max 1,099. - Status at filing, from `CLASS_OF_ADMISSION`: 731 not in the USA, 337 F-1, 126 H-1B, 38 F-2, 38 B-2. - Prevailing wages, annualised from `PW_WAGE` and `PW_UNIT_OF_PAY`. - `EMP_RECEIVED_PAYMENT` is "N" on all 1,612 rows. - Attorney and employer tables counted on `AGENT_ATTORNEY_FIRM_NAME` and `EMPLOYER_NAME`, normalised for punctuation and entity suffixes. Limitation: country of citizenship appears in FY2020 through FY2024. DOL's revised ETA-9089 dropped it from FY2025, along with birth country and class of admission, taking the file from 155 columns to 137. So Bangladeshi cases can be isolated for those five years and not after; employer totals continue. Two parsing traps found while extending the range, both of which would have published wrong figures silently: - The FY2026 workbook carries about 813,000 empty padding rows. Counting XML rows gives 925,430 where the real case count is 112,550. FY2025 has 1,603 such rows. - FY2025 and FY2026 write the status as "Certified - Expired" with spaces, where every earlier year writes "Certified-Expired". Matching the literal string dropped 57,073 expired cases in FY2025 alone, which made employers in that year look cleaner than they are. Status is now normalised before counting. ## 2. US Dept of State, Report of the Visa Office, Table VI Part II (first-hand) FY2021, FY2022, FY2023, FY2024 PDFs, parsed with `pdftotext -layout`. Bangladesh row, immigrant visas issued: EB-3 Other Workers 32 / 44 / 104 / 78, total 258. EB-2 10 / 49 / 69 / 107. EB-3 skilled 11 / 15 / 19 / 12. Philippines FY2024 Other Workers 1,022. Comparators for Pakistan, Nepal, India, Vietnam and Mexico read from the same tables. ## 3. US Dept of State, Visa Bulletin, October 2025 to September 2026 (first-hand) All twelve issues read. EB-3 Other Workers final action, worldwide: 15JUL21 rising to 01APR22 over eleven months, 8.5 months of queue cleared, a rate of 0.78. EB-3 skilled and professional: 01APR23 to 01SEP24, 17 months cleared, rate 1.55. Bangladesh is charged to "All Chargeability Areas Except Those Listed", so the worldwide column is the one that applies. Correction this log exists to prevent: an earlier draft compared a Dates for Filing figure against a Final Action figure and reported the queue moving at half speed. Reading all twelve issues corrected it to 0.78. ## 4. USCIS, Form G-1055 Fee Schedule, edition 09/09/26 (first-hand) PDF downloaded and parsed. I-140 $715 paper / $665 online. Asylum Program Fee $600 regular employer, $300 small employer or self-petitioner, $0 nonprofit. I-485 $1,440 / $1,390. I-765 $520 / $470, or $260 alongside a pending I-485 filed on or after 1 April 2024. I-907 premium processing on an EB-1/EB-2/EB-3 I-140, $2,965. ## 5. US Dept of State, Fees for Visa Services (first-hand) Employment-based immigrant visa application (DS-260) $345. Family-based $325. I-130 $675. Affidavit of Support review $120. DV registration $1, DV application $330. ## 6. US Dept of State, Directory of Visa Categories (first-hand) The complete directory contains two B classes, B-1 and B-2. There is no B-3. This is the basis for the article correcting the term as it circulates in Bangladesh. ## 7. eCFR, 20 CFR 656.12, Improper commerce and payment (first-hand, verbatim) 656.12(a): labor certifications "are not articles of commerce. They shall not be offered for sale, barter or purchase by individuals or entities", with investigation, denial under § 656.24, revocation under § 656.32 and debarment under § 656.31(f) as consequences. 656.12(b): the employer "must not seek or receive payment of any kind for any activity related to obtaining permanent labor certification, including payment of the employer's attorneys' fees". The worker may pay their own costs including their own attorney, "except that where the same attorney represents both the alien and the employer, such costs shall be borne by the employer". ## 8. USCIS, I-140 receipts, approvals and denials by preference (first-hand) Table by class of admission, FY2013 to FY2023, parsed from the published PDF. EB-3 FY2022: 53,096 approved against 1,487 denied, a 97.3% approval rate. EW3 approvals by year: 403 in the first year of the series rising to 9,272 in FY2022 and 8,631 in FY2023. ## 9. CLINIC v. Rubio, SDNY (secondary, and labelled as such) The 75-country immigrant visa pause took effect 21 January 2026. On 21 August 2026 the US District Court for the Southern District of New York vacated it. The government filed a notice of appeal on 10 September 2026. Bangladesh is named in the list of 75 countries. The docket itself was not read. This rests on the case page maintained by a party to the litigation and on concurring accounts from several immigration practices. It is the weakest source on this list and the only claim on the site not traced to a primary document. ## 9b. Rights to republish the PERM data The PERM disclosure workbooks are works of the US federal government. Under 17 U.S.C. § 105 such works are not subject to copyright in the United States. DOL publishes them on its own site as part of what it calls "an open and transparent government", and attaches no terms of use, licence or redistribution condition to them. What DOL withholds is narrower than it first appears, and it is worth being exact. DOL does not publish the name of the sponsored worker, so no beneficiary is named anywhere in the file. It does publish a great deal that identifies other people: employer addresses, phone numbers, email addresses, FEINs and attorney names are all present. And where the employer is a private household sponsoring a caregiver, a nanny or a home health aide, the employer name is itself a natural person's name. So the file is not scrubbed of personal data, and an earlier version of this log said otherwise. What this site republishes is therefore narrower than what the files contain: employer name, and case counts and outcomes by year. Employer addresses, phone numbers, email addresses, FEINs and attorney names are not published. Employers that are private individuals are excluded from the index entirely. Most common occupation, main worksite state and median prevailing wage describe the sponsored worker rather than the employer, so they are published only where an employer filed five or more Bangladeshi cases, which is the usual floor for releasing a cell of a statistical table. ## 10. YouTube oEmbed (first-hand) Every embedded video ID on the site resolved against `https://www.youtube.com/oembed` to confirm it exists, is public, and to read its real title. 53 IDs checked. One, `jMJZ3Ot3zs4`, no longer resolves and its embed was removed. Three scraped IDs pointed at videos merely linked in the 2021 text rather than embedded, and were corrected against the original HTML. --- ## What is enforced automatically `npm run check` builds the site and runs `scripts/verify.mjs`, which fails on: legacy URLs that are neither built nor redirected; articles marked as researched that carry no sourced citation; duplicate titles or descriptions; missing canonical, description, structured data or `html lang`; structured data that does not parse; a bn/en pair where one side is missing from the build; em dashes; the author's name or the branding email appearing in a body; and key figures that appear in one language but not the other.