Scans come back empty
The pages are images. There is no text layer to extract, so the document is recorded as read and contains nothing. Nobody is told which pages went missing.
Case Analysis & Argument
Lodge the bundle as it exists — scanned FIR, charge sheet, statements, orders. Litora AI reads every page, tells you where your file is strong and where it will be attacked, then puts your advocate and theirs in front of a bench and lets them argue it out. You read the exchange, and you read the bench’s reasoning at the end.
£99 per matter Scanned bundles read page by page No subscription Credits do not expire
The question before the court Whether the identification evidence is sustainable in the absence of a test identification parade.
The sole sighting was at night, at distance. No TIP was held at any stage. Identification made for the first time from the dock is worth very little.
The witness knew the accused by sight beforehand. Where the witness is previously acquainted, a parade adds nothing and its absence is immaterial.
Where on this record is prior acquaintance established? The statement says “recognised him” and goes no further. Who proves it, and when?
Prior acquaintance is asserted and not proved. On this record the identification does not carry the weight placed on it.
An FIR typed on an old machine, stamped, annotated by hand and scanned crooked. A charge sheet that is forty pages of images with no text in the file at all. Any tool that cannot read that bundle cannot help you with the matter, whatever else it does well.
The pages are images. There is no text layer to extract, so the document is recorded as read and contains nothing. Nobody is told which pages went missing.
A long scan is sent to a model in one piece. The reply runs out of room around page ten and stops. The first ten pages read perfectly, so the document looks complete.
Pages are transcribed in small batches, several at a time. You watch the count climb — page 12 of 48. If a batch fails, that gap is printed in the transcript where the pages should have been.
Nothing needs retyping, renaming or reordering. Upload the file in whatever state you hold it, and put each document in the pile it belongs to — that part matters, and § 03 explains why.
A local pass works out what the file is, how many pages it has, and whether its text layer is usable. No model is called, so this is instant and costs nothing.
A scanned bundle is cut into batches of a few pages each. Each batch is transcribed on its own, several running at once, so a long file moves steadily instead of all at the end.
FIR numbers, section numbers, dates, times and amounts are transcribed as written, in the original script — Hindi, Punjabi, Urdu or English. Nothing is translated or tidied.
Illegible handwriting is marked [illegible] rather than guessed at. A batch that fails leaves a visible gap in the transcript, not a silent join from page 8 to page 13.
If a batch of very dense pages was cut short, one button re-reads that document in smaller batches. Retrying at the same size would truncate in the same place.
Once read, each document is catalogued: what it is, who it names, what dates it fixes, and which provisions it cites. That index is what the analysis reasons over.
A charge sheet records what the investigating officer says. That is not the same as what is proved. Any system that treats the prosecution’s bundle and your client’s affidavit as the same kind of material will read assertions against your client as established facts.
The first information report itself, with its endorsement or GD entry. The document every date in the matter is measured against.
Charge sheet, statements under sections 161 and 164, panchnamas, recovery memos, FSL and medico-legal reports, seizure lists.
For an appeal or revision, the judgment being challenged. Without it there is nothing to argue error against, and the system says so plainly.
Defence documents, alibi material, your witness statements, replies and applications — kept separate from the prosecution’s throughout.
The workspace also asks which court you are in and at what stage. A High Court appeal is argued against the findings of the court below — error, not evidence afresh. A trial is the opposite. The same file supports very different arguments depending on the answer.
One reader takes in the whole file at once and reports on it. Not a summary — a working note of the kind a good junior would put in front of you before a conference.
Every point names the paper it came from and how heavily it weighs. A weakness carries the way the other side will use it, and what would blunt it.
Documents pleaded but not produced, dates asserted but not evidenced. A gap in the record is stated as a gap, never filled in by inference.
Each section is marked clearly engaged, arguably engaged, or asserted without any factual basis on the record.
Limitation, sanction, jurisdiction, custody, investigation defects — each returned as satisfied, defective, or impossible to determine on this record.
The gaps the papers cannot close. Your answers count as instructions, and they move the assessment when you re-run it.
| Item | Finding | Status |
|---|---|---|
| Identification | No test identification parade held at any stage. | Strong for us |
| Recovery | Seizure memo produced nine days after the alleged recovery. | Exposed |
| Malkhana register | Not produced; continuity of custody cannot be traced. | Not on record |
| Sanction | Nothing on the file establishes whether sanction was obtained. | Undetermined |
| Authority cited | Supporting case name produced by the model, not checked against a reporter. | Unverified |
Not a committee. A hearing has two sides and someone deciding, so that is what runs. Twelve turns, strictly in order, each one reading everything said before it.
Opens on the theory of the case, answers what the other side puts, meets the bench's questions directly, and closes on the point most likely to carry. Argues from the papers, and says so plainly where the papers are silent rather than filling the gap.
Deliberately hostile and genuinely competent. Takes the weakest link in our position and works it. Its value is entirely in being difficult before someone else is difficult in a room where it counts.
Neutral. Presses whichever side is being loose, marks what the record does not support, and at the end sets out how the argument stands and what would change it. Its view is a rehearsal view, never a ruling.
An earlier version ran sixteen specialists over the same file. They largely agreed, because they were one model reading one record from overlapping angles — and sixteen agreeing outputs look like corroboration while being nothing of the sort. The specialist work did not disappear; it moved into the assessment, which does it once, properly, before anyone argues.
It is instructed to argue the other side’s case as they would actually argue it — competently, at its strongest — and to work the weakest link in your position. Its entire value is in being difficult here, rather than in a room where it counts.
At the end it sets out which submissions carried, which failed, what this record simply does not allow anyone to decide, and what would change the answer. It has no authority of any kind and does not write as though a court has decided anything.
You asked how strong the case is, so you get a number. But a bare number on a screen reads as a measurement, and it is nothing of the sort — so it never travels alone.
A whole-number estimate of the realistic prospect of success on the papers presently on file. Never 0, never 100 — nothing in litigation is certain.
Worked out from how much of the file was readable, whether the documents this kind of matter turns on are present, and how many critical questions are still unanswered.
Why it is not ten points either side. Stated in words you can disagree with, which is the only form of estimate worth having.
The specific document, admission, witness or authority that would raise or lower it, and roughly by how much. This is the part you can actually do something about.
When confidence is low, the figure is deliberately pulled toward the middle and the screen says so. A thin file cannot support a confident-looking number, and dressing a guess up to look like a judgement would be the most harmful thing this product could do.
The order is enforced, not suggested. Nothing is assessed before the papers have been read, and nothing is argued before the assessment has been made.
Name it, say which court and which stage, and record the FIR number and sections if you have them. Which court you are in changes how everything downstream reasons.
FIR, police and prosecution papers, the order under challenge, your own material. Drop them in as they are — scans, photographs, mixed formats, inconsistent numbering.
Strengths, weaknesses, gaps, provisions, procedure, authority and the estimate. Answer the questions the papers could not, then run it again and watch the estimate move.
Both advocates argue, the bench questions them and gives its reasons. Export the whole thing to PDF — the professional judgement, and the signature, remain yours.
Client material is held on the footing a practitioner would expect: restricted, logged, and removable on instruction.
Documents are transmitted over TLS and stored encrypted. Access is restricted to the matter they were lodged against.
Case material is processed to answer your instructions and is not used to train or fine-tune any model.
Partners, associates and clerks see only the matters they are assigned to. Permissions are set by the account holder.
Uploads, reads, assessments, arguments, exports and deletions are recorded with user and timestamp, so the chain of handling is reconstructible.
A matter and its documents can be deleted from the workspace at any time, subject only to retention we are required by law to observe.
Time-based one-time codes are available on every account and can be made mandatory across a firm.
There is no subscription and no free tier. You buy matter credits; one credit admits one matter to the workspace — the full bundle read however many pages it runs to, the assessment, and the argument. Credits do not expire.
One matter, taken start to finish.
£49
£49.00 per matter
One-off charge. No subscription, no renewal.
For a practice with a steady contested list.
£449
£44.90 per matter
One-off charge. Credits do not expire.
For chambers and firms running volume.
£999
£39.96 per matter
One-off charge. Best rate per matter.
For corporate legal teams, insurers and volume users.
On application
Rate agreed on volume
Quoted per engagement, invoiced on agreed terms.
A credit is spent when the first document is lodged. At that moment the reading begins and the cost is incurred on our side, so the credit is consumed and is not returned afterwards. Opening a matter you never upload to costs nothing.
If yours is not answered here, put it to us directly and you will get a straight reply.
Yes, and that is the part we have worked hardest on. Scanned pages are transcribed in small batches, several at a time, so a forty-page bundle moves steadily rather than timing out. You see the page count climb as it goes. Handwriting that is genuinely illegible is marked as illegible rather than guessed at, and any batch that fails leaves a visible gap in the transcript so you know exactly which pages are missing.
A typical forty-page scanned bundle reads in around a minute. It depends on how dense the pages are and how busy the model is. You do not have to sit and watch — a background worker continues the reading after you close the tab, and the progress is there when you come back.
No, and it is not designed to. It prepares material for a qualified professional to review, test and settle. Nothing it produces is advice, and no output should be filed, served or relied upon without that review.
It is an estimate on the papers you supplied, and it is only ever as good as those papers. That is why it is never shown alone: it carries a confidence level worked out from how much of the file was actually readable and whether the documents that matter are present at all. When confidence is low the figure is deliberately pulled toward the middle and the screen says so. It should not be repeated to a client as a number in its own right.
No, and every one of them is labelled unverified for that reason. Nothing in this system reads a law report. Case names produced by a language model can be plausible and non-existent, and a fabricated citation carried into a court is the worst outcome this product could cause. Verify each one against SCC Online, Manupatra or the court’s own site before it goes anywhere near a filing.
We ran sixteen. They largely agreed with each other, because they were the same model reading the same file from overlapping angles — and agreement between instances of one model is not corroboration, however much it looks like it. A hearing has two sides and someone deciding. The specialist analysis still happens; it happens once, structurally, in the assessment, before anyone argues.
It is a rehearsal and nothing more. It has no bearing on any real proceeding, no connection to any court or tribunal, and carries no authority of any kind. The bench’s figure is an estimate for your own preparation, not a forecast.
When the first document is lodged against a matter. Opening an empty matter costs nothing, and you can name it, frame the issues and change your mind freely. The moment a bundle is uploaded the reading begins, the cost is incurred on our side, and the credit is gone. It is not returned if you later decide not to proceed.
Neither. There is no subscription and nothing renews. Credits sit on the account until you use them, and you buy more when you need more.
These are not buried in the terms. They are the terms, stated where you can read them.
Nothing produced by this platform constitutes legal advice, and use of it creates no advocate–client relationship. A qualified professional must review every output before it is relied upon, served or filed.
Language models make errors, including confident and well-worded ones. The workspace marks what it could not verify; it cannot mark what it does not know it got wrong.
Nothing here checks a case name against a law report. All authority produced by the system is labelled unverified and must be confirmed against the primary source before use.
Machine transcription of a photocopied or handwritten page can misread a digit or a name. Check any FIR number, section, date or amount you intend to rely on against the original page.
The argument is a preparation tool with no bearing on any real proceeding. No judiciary, government department, police service, bar council, tribunal or regulator has endorsed, approved or certified this platform.
The percentage is a reasoned estimate on the papers supplied, with its confidence and its missing inputs disclosed. It is not a forecast of what any court will do, and it must not be presented to a client as one.
One credit admits one matter. It is consumed when the first document is lodged, because the analysis cost is incurred at that moment. It is not refundable afterwards, whether or not you use the output.
Lodge the bundle as it is. Find out where you are exposed before the other side does, and hear the argument against you while there is still time to answer it. Ninety-nine pounds for the matter, nothing monthly, nothing to cancel.
Litora AI is operated by Rank First Technologies Private Limited, a private limited company incorporated in India under the Companies Act, 2013. It is not a law firm and does not provide legal advice. All outputs must be verified by a qualified advocate before use.
Prepare your next contested matter
The whole bundle read page by page, an assessment of where you are exposed, and both sides argued in front of a bench. No subscription, nothing to cancel.