I AGREE Submission packet · Exhibits A–E Open the prototype

In the matter of the terms nobody reads

I AGREE

The fine print decoder

Five things were asked for. All five are below, each one filed as an exhibit you can open and check. Numbers run down the left margin because the point of this project is that every claim should cite exactly where it came from.

Index of exhibits Filed 6 August 2026
ExhibitA

Project description

The concept and purpose, inside the 200-word limit.

Every day people click “I agree” on documents nobody reads. Reading the terms you accept in a year would take weeks, so the agreements deciding what happens to your data, your money and your work go unread — and they are written knowing that.

I AGREE is a fine print decoder. Paste a terms of service, privacy policy or rental agreement, or give it a link, and in about ten seconds it returns a plain-English verdict plus the specific risks across three categories that cost people real money: what happens to your personal data, how hard the service is to cancel, and what rights you sign away over content you create.

The difference is verification. An AI asked to summarise a contract can quietly invent a clause that was never in it. Here, every flag must quote the document word for word, and each quote is checked character by character against the source before anyone sees it. Quotes that cannot be matched are withheld for human review instead of presented as fact. Each warning links to the exact highlighted line that caused it, so you never take the tool’s word for anything.

No account. No upload. Free.

197 words of 200

ExhibitB

Presentation

Three slides: the idea, how it works, and what it changes. Scroll the deck sideways, or take the PDF.

01 / 03

The idea

Contracts are written to be skipped. This one gets read.

Consent is the legal basis for most of what companies do with our data and our money — and it is collected through documents built to be too long and too dense to read. The asymmetry is the product, not an accident.

Unreadable
Terms run to tens of thousands of words, in language written for courts rather than for the person clicking accept.
Unavoidable
Refusing means losing the bank, the flat, or the job application. The consent is real in law and hollow in practice.
Unequal
One side has lawyers. The other has thirty seconds and a checkbox. A decoder puts a reader on the second side.
02 / 03

The workflow

Four steps, and one of them is the whole point.

These are numbered because the order decides the outcome: nothing reaches the reader until it has survived step three.

Step 1
Take the document
Pasted text or a link. Long documents split into overlapping sections so no clause is lost at a boundary.
Step 2
Find the risks
A language model returns strictly structured findings in three categories, each carrying a severity and a quote.
Step 3 — the point
Verify every quote
Each quote is matched character by character against the source. No match, no claim: it is withheld and marked for human review.
Step 4
Show the evidence
A verdict, plain-English warnings, and the document itself alongside them with every cited clause highlighted in place.
03 / 03

Expected impact

Make the unreadable checkable, for the cost of a rounding error.

The claim worth testing is not “AI reads contracts” — plenty of tools say that. It is that a reader can check the tool, clause by clause, in the source text itself.

10 seconds
In place of a document most people would need an hour and a law degree to work through, assuming they tried at all.
Every flag cited
Each warning shown is anchored to verbatim text. Findings that cannot be verified are separated out, never quietly dropped or dressed up as fact.
Under $0.002
Measured cost per document, which is what lets it stay free for the people who need it most.

Scroll sideways for all three slides

Download the slides
ExhibitC

Working prototype

Deployed and usable now. Paste any terms of service and it runs the real analysis.

The prototype: risk flags on the left, the source document on the right with every cited clause highlighted.
Open the prototype Check live status
ExhibitD

Demonstration video

38 seconds, inside the two-minute limit. The screen footage is the deployed application, recorded in one unedited session.

Download the video
ExhibitE

Files and source

Everything behind the packet. One repository holds the application, this page, the slides, and the scripts that produced the video and the model benchmark.

Item What it is Where
Source code The whole application: analysis pipeline, quote verification, interface. github.com/satyandeveloper/VASANT_VALLEY_innovvate
Live prototype The deployed application, running against the live model. innovvate.devjindal.me
Demonstration video 1080p MP4, composed in Remotion from a real recorded session. assets/demo-film.mp4
Slides The three slides above, as a printable PDF. assets/i-agree-slides.pdf
Verification The quote matcher that decides whether a finding may be shown at all. lib/verification.ts
Model benchmark Scores candidate models on cost and on verbatim-quote accuracy, using the app’s own verifier as the measure. scripts/bench-models.ts
Video source The Remotion composition and the Playwright script that captured the footage. supporting/
Sample documents The public documents used for the demos, with their sources credited. scripts/samples/

Certificate

I AGREE is an information tool, not a law firm. It points at terms worth a second look; it does not give legal advice, and it will not catch everything. For anything that matters, talk to a qualified lawyer. The prototype is at innovvate.devjindal.me.