# Independent official-source audit

VERSION HISTORY: 2026-10-14 EDT | chatgpt_codex / source_reviewer parallel pass | created before course integration. DEV_LOG creation covered by parent session codex-2026-10-14-startup, reviews/*.

Observed 2026-10-14. Reviewed only this agent's programs.json and official sources; no peer course consulted. This pass covers 11 providers; other 11 assigned to parent. No data edited here. Findings are field corrections, not eligibility certification. Unknown fields remain unknown unless supported below.

| Provider | Finding and action | Official source |
|---|---|---|
| Microsoft / Azure | Correct stage to pre-seed through Series B. Previously recorded Series C source conflict no longer exists. Application specifies B2B software/AI/technology, MVP trajectory and independent companies. Overview credit ceiling permits exactly $350,000 lifetime free credits; replace 'below' with 'not more than'. Conditional $150k and $200 starter amount remain supported. | [Overview](https://learn.microsoft.com/en-us/startups/microsoft-for-startups/overview), [Application](https://learn.microsoft.com/en-us/startups/microsoft-for-startups/application) |
| AWS / Bedrock | Amounts remain supported: $1k starter, selected Founders up to $5k, Portfolio up to $200k. Landing page now advertises review 5–10 business days; existing application guide may differ, so cite source-specific estimate rather than promise. Paid Tier, pre-Series B, ten-year maximum and Org ID remain supported. | [Credits](https://aws.amazon.com/startups/credits/) |
| Google Cloud / Gemini | AI Scale record supported: $250k first-year maximum, second-year 20% coverage up to $100k, discretionary admission, Gemini core product and funding/age limits. Preserve year-two paid-usage example and third-party model exclusion. Bootstrapped tier should be separately researched rather than infer AI Scale eligibility. | [AI program](https://cloud.google.com/startup/ai) |
| NVIDIA | Existing incorporated/developer/website/under-ten-year eligibility and exclusions supported. Free program and DLI course codes supported; no guaranteed cash-equivalent credits. | [Inception](https://www.nvidia.com/en-us/startups/) |
| OpenAI | Add process: unique VC referral code, product/API-use description, company/key contacts/business email, basic funding details; no pitch deck requested. Add restriction: exhaustion switches to standard billing. Credit amount and expiry remain unknown. VC-dependent additional benefits include rate limits and engineer time, not guaranteed acceptance. | [Startups](https://openai.com/business/why-openai/startups/) |
| Together AI | $15k/$30k/$50k maximums, 3/6/10 engineering hours and funding ranges supported. Preserve $5m boundary ambiguity and Reserved GPU Clusters exclusion. Geography and expiry unknown. | [Accelerator](https://www.together.ai/startup-accelerator) |
| Fireworks | Existing >$500k VC funding, under-five-year, pre-Series D, one-year expiry and on-demand restrictions supported. Amount unknown. Add optional joint marketing/case-study help if desired. | [Startups](https://fireworks.ai/startups) |
| OpenRouter | Replace pending startup-terms audit with linked dedicated terms, updated August 22, 2026. Add no promotion stacking unless permitted in writing; no past balance coverage unless permitted in writing; benefits/fees subject to discretion, caps and revocation. Keep six-month credit / twelve-month fee distinction; award communication controls. | [Program](https://openrouter.ai/startup-program), [Startup terms](https://openrouter.ai/startup-program-terms) |
| Replit | Eligible fund/accelerator backing, up to $25k and onboarding supported by founders page. Startup page yielded title only; amount supported but expiry and detailed award restrictions still unknown. Historical Race to Revenue is separate; no current cohort implied. | [Founders](https://replit.com/usecases/founders), [Startup application](https://replit.com/startups) |
| Vercel | Existing maximum $30k, partner/funding conditions and one-year commitment supported. Preserve explicit ambiguity: FAQ says both total-allocation and monthly AI Gateway 50% cap. Payment method/standard-rate overage/early exhaustion restrictions essential. | [Credits and FAQ](https://vercel.com/startups/credits) |
| Hugging Face | Free LLM course, Python prerequisite and forums confirmed. Bounded official-domain startup search found no current startup grant offer. Do not treat user organization https://huggingface.co/startups as corporate offer. Current official inference billing says free users have no included monthly credits; paid PRO and Team/Enterprise $2 monthly compute credits are subscription benefits, not startup grants. | [Course](https://huggingface.co/learn/llm-course/chapter1/1), [Billing](https://huggingface.co/docs/inference-providers/en/pricing) |

## Access and currentness limits

Microsoft pages displayed an authorization notice but substantive article text was readable; no account sign-in performed. Replit startup application content was not exposed beyond title. Search snippets are discovery only and were not substituted for hidden offer terms. Accessing a page today establishes observation date, not an award or application guarantee. Reverify high-value claims and actual acceptance terms immediately before applying or estimating savings.

## Independent pass conclusion

Assigned records are broadly conservative and consistent with observed pages. Correct Microsoft stage and refine OpenRouter/OpenAI restrictions before release. Record accessed_at for every evidence item; verified_at must reflect actual observed source rather than a mass date refresh on inaccessible fields. Unknown details should remain visible. No claim that all 22 records were independently audited by this pass.