SPARKLE

Grant writer with a truth-pack guardrail. The one who makes a proposal sing off the page.

If you're searching this with a deadline two weeks out — proposal voice that doesn't read AI-generated, the SBIR that got rejected for invented stats, a grant writer who'll say do not apply when that's true — the page is for you.

Foundation, federal SBIR/STTR, NSF, NIH, DOE. Voice-locked to your writing, fact-locked to your truth pack, allowed to say do not apply.
An agent that cannot refuse is an agent that must always be supervised.

Why SPARKLE exists

Every AI grant tool that came before SPARKLE failed because it hallucinated numbers. A program officer with three hundred applications in her queue spots manufactured statistics in two paragraphs. One hallucinated fact, and the proposal — and the applicant's reputation with that funder — is dead. Not just for this cycle. For good. Program officers remember.

Grant writing is too high-stakes for a tool that gets the facts approximately right most of the time. It has to be structurally impossible for the tool to invent. The truth pack is not a feature. It is the premise.

Marketing writing persuades by lighting up. Grant writing persuades by proving care. Evidence-based, humble, grounded in a specific beneficiary, falsifiable in the logic model, justified line by line in the budget. SPARKLE is built for that craft, not the easier one.

Four disciplines

Truth-pack as hard guardrail

No sentence ships without a truth-pack citation. The applicant's verified facts are the corpus; SPARKLE writes only from them. Hallucination is structurally blocked at the input layer.

Voice lock

A style fingerprint keyed to the applicant's own writing samples. Every draft is checked against the fingerprint before leaving. The AI-tells that program officers reject — uniform rhythm, parallel structure, equal-bullet cadence — are caught.

Go/no-go veto

SPARKLE is allowed to refuse. A "do not apply" verdict saves 20-60 hours and a bruise on the funder relationship. Refusal is protected behaviour. Override at will; overrides are logged.

Export-everything · data isolation

The applicant owns their data. Every truth pack, voice sample, draft, and citation trace — exportable as plain files, day one. Applicant A's truth pack never improves drafts for Applicant B.

voice

Voice declaration

free

Identity-shape read. SPARKLE introduces herself and what she will and will not do — read before you decide whether the craft fits the engagement.

voice_samples

Voice fingerprint

included

Three to five writing samples at onboarding — blog posts, prior proposals, mission statements. Per-sample fingerprints stored. The drift check that runs on every draft uses this fingerprint as ground truth.

go_no_go

Fit verdict

free

A verdict on whether you should apply to this funder at all. Reasons, score, named gaps. "Do not apply" is a valid answer and is the answer that saves you the most hours.

voice_drift

Drift score

included

Every draft scored against the voice fingerprint before completion. Above-threshold drafts get held until the texture matches your real writing. Funder-voice alignment happens separately at vocabulary + framing.

proposal

Full proposal · success-priced

2% capped $40k

Truth-pack-grounded, voice-locked, theory-of-change aligned. Logic Model. Beneficiary narrative. Budget justification. The applicant reviews and signs. The applicant who loses owes nothing.

cite

Citation trace

included

Every claim in the draft mapped back to its truth-pack source. The audit trail your reviewer expects, on demand. Hand to a foundation program officer; hand to your board.

"Success-based pricing is not a go-to-market tactic. It is the principle made structural."

2% of awarded funds, capped at $40,000. No up-front risk. Both sides bear the outcome. The applicant who loses owes nothing; the applicant who wins pays because the money came in because of the draft. That is why SPARKLE can refuse without cost.