Auditable on purpose. Every score, every read, every draft.

A score you can check, line by line

Every match score is arithmetic, not a vibe. It starts at a baseline and every contribution is a labelled, signed line you can read on the round's page: topic fit, geography, your stated preferences, past delivery echoes. Semantic similarity from AI embeddings joins as one more labelled line, calibrated within your own profile's distribution. And where a round's own published wording restricts who may apply and your recorded entity type cannot enter, the cap appears as a line in the breakdown — never a silent exclusion.

A match score states fit. It is never a funding probability, and we never call it one.

Eligibility: Helm proposes, you decide

Helm never asserts that you are eligible or ineligible for a round. The eligibility panel pulls the guidelines' own who-may-apply requirements out of your uploaded documents for you to verdict, one by one. On Grant Pro, the AI can propose a reading of each row against your recorded facts — looks met, looks not met, or unclear — always naming the fact it leans on and the thing to check. A proposal never sets a verdict; the buttons stay yours. Where Helm holds no deciding fact, the required answer is "unclear", not a guess.

The guideline deep read: quote or omit

One read of the round listing and every guideline document you upload returns the stated funding facts — amounts, co-contribution, dates, eligible and ineligible activities, assessment — each carrying a verbatim quote and its source document. A fact the documents do not state is omitted, never padded. When two documents disagree, both versions are shown with their evidence.

Drafting: cite or ask, never invent

Tender Mate drafts application sections in project-narrative voice from permitted sources only: your organisation's profile, the round's own guidelines, your evidence library, answers you have approved, and — where GrantConnect has published award notices for the same program family — the aggregates. A claim that cannot be traced to a source becomes a visible [CONFIRM: ...] marker. Program aggregates may calibrate scope and budget; they never name a recipient and never state odds.

Verified facts, labelled as such

Your entity type can be checked against the Australian Business Register, and your ABN against a weekly mirror of the ACNC Charity Register. Checked facts are labelled checked, declared facts are labelled declared — and absence from the ACNC snapshot is stated as exactly that, never as "not registered".

The gates that hold this to account

• Hallucination audit: before every release, drafting runs against test profiles including a live grant round, and every number, status and name in the output must trace to a source or the release fails.
• Score integrity audit: every stored score is recomputed from scratch and compared. A drifted score fails the gate.
• Versioned prompts: the instructions the model runs under are operator infrastructure, reviewed and versioned like code.

What the AI never does

• No eligibility rulings. Helm surfaces the guidelines' rules and remembers your verdicts; it never decides them.
• No funding probabilities. Fit is knowable from the outside; your odds are not.
• No apply/don't-apply verdicts. Helm states evidence and gaps; the decision is yours.
• No invented status. A charity registration or incorporation you do not hold cannot appear in a draft as if you did.
• No automated submissions. Helm prepares; you submit.
• No training on your data. AI analysis runs through OpenAI under their API terms; your data does not train their models, and Helm trains no models on it either.

Where your data lives, and what leaves Australia in transit, is on the security page. Ask us anything specific through the contact form; we answer in writing.