Cite, don’t infer. Then human-review every claim.

TenderPulse content draws from four classes of primary source. We do not paraphrase secondary blogs, AI summaries, or industry chat as if they were authority.

• BPPA · bppa.portal.gov.bd — Bangladesh Public Procurement Authority circulars, gazetted rules, official guidance.
• CPTU · cptu.gov.bd — Central Procurement Technical Unit forms, model documents, e-GP user guides.
• e-GP · eprocure.gov.bd — the live tender notification + submission system. Tender ZIPs analysed by TenderPulse originate here.
• Multilateral docs — World Bank Procurement Regulations and ADB Procurement Policy + Procurement Regulations, the canonical references for WB/ADB-funded projects in Bangladesh.

1. Topic selection — questions procurement teams actually ask (e-GP help-desk patterns, BD-procurement search demand, customer support tickets), not invented SEO bait.
2. Primary-source pull — the writer pulls the most recent BPPA circular / CPTU model document / WB or ADB regulation that governs the topic. The post is structured around what those documents actually say.
3. AI-assisted draft — Anthropic Claude on a managed inference layer drafts the explanatory layer in EN and বাংলা, grounded in the primary-source pull. The model is configured to emit rule numbers verbatim and refuse to invent BDT figures.
4. Specialist review — a procurement compliance specialist on the team validates every rule citation, every BDT figure, every regulatory interpretation against the primary source. Posts that touch legal interpretation are gated additionally on external advisor review.
5. Publish with byline + dated review tag — the byline names the team specialty (Editorial, Procurement Compliance, Engineering, or Research). When a specialist reviewer signs off, the byline shows “Reviewed by TenderPulse SME panel on [date]”. Individual reviewer names are kept off the public surface by design.
6. Corrections-watch — every page has a dated lastUpdated; if a BPPA circular changes the relevant rule, we re-pull, re-draft, re-review, and append a dated correction notice.

TenderPulse content is researched and drafted with AI assistance, then reviewed by human procurement specialists before publication. The model is grounded in the primary-source pull (BPPA + CPTU + WB/ADB) so it cites instead of inventing. When the model flags a low-confidence claim — e.g. a BDT figure or a rule applicability question — it is escalated to specialist review before it ever reaches the page.

We disclose AI assistance explicitly in our editorial standards. The honest framing: AI broadens our research and bilingual capacity, but the human specialist owns the final word on every regulatory claim.

Every public-facing page ships in English and বাংলা. Bangla is native-authored, not translated — voice and tone are validated against the way Bangladeshi procurement teams actually speak about tendering, not Google-translated from English.