Details
Seven steps, three gates, three deployment tiers
The home page keeps to what matters most. This page spells out how each step works, where it runs, and how it differs from the usual approach.
How it works
From a pile of documents to a finding someone can stand behind, in seven steps
Every step's output carries positions and fingerprints, and the next step can only use the previous step's output. The model appears only at steps four and six, and only proposes.
- 1
Documents go in
Any format, any volume; originals are immutable
- 2
OCR and intake
Two OCR engines cross-checked; one fingerprint per document
- 3
Extract facts
Numbers, amounts, dates, parties, each with its exact position
- 4
Rules and candidates
Deterministic rules produce findings; the model only proposes candidates
- 5
Match and recompute
Back to the source line, recomputed on the spot
- 6
Independent replay and review
A second program replays; repeated review in idle time adjusts confidence only, never publication
- 7
Findings · handling · answers
Each with its source, owner, deadline and follow-up
How it differs
Others give you file links. We give you the sentence in the source.
How far a finding can be questioned: to the source, to the recomputation, to an independent replay.
| The question | Usual approach | Fidelog |
|---|---|---|
| What is this finding based on? | A list of links: "these documents were consulted" | Which line, which sentence, recomputed in front of you |
| Is the number right? | Check it yourself | The page recomputes from the source figures and flags mismatches in red |
| Is there a second pair of eyes? | The same model answers again | A separate program replays from the source; if they disagree it is not published |
| Does it work offline? | Needs the internet, or "supports on-prem" with an incomplete chain | The whole chain on your machine, never connected |
Your data
Six rules, three deployment tiers
Security is a set of rules, not a promise. Every document gets a level and a destination when it comes in; pages with real data sit behind the login.
- 01
The whole chain runs offline
Documents, models, evidence and approvals stay on your machine or network, with no outbound connection.
- 02
Classify first, then route
Every document is rated L0–L4; highly sensitive levels never leave, and what may leave is the minimum necessary, de-identified.
- 03
Encrypted in transit and at rest
Encrypted to approved standards, keys kept apart from business data, backups equally encrypted and access-limited.
- 04
Clear roles
Viewing, analysis, recovery and approval are separate roles; every action leaves an auditable record, and logs hold no sensitive text.
- 05
Trails can't be erased
Every document has a fingerprint and every change is chained; alter one thing and everything after it stops matching.
- 06
The model proposes, never decides
Every finding passes a local check; payments, HR and approvals are out of the model's reach.
Fully offline
Finance / highly sensitive sectors
Documents, models, evidence and approvals all inside the network; the whole chain runs offline.
Hosted in your network
Companies with their own data centre or private cloud
The same system installed in your environment; you can stop, audit and export at any time.
Hybrid
Companies that want external models quickly
Sensitive documents and real values stay local; only ordinary text goes to approved external models, on your switch.
Questions
Two more
Will the model make things up?
Yes. That's why every finding passes a local check and is not published if it doesn't match the source; when it can't decide, it says so and hands over to a person.
Do we have to give you our data?
No. The system is installed on your machine or inside your network; whether to use an external model, and up to which level, is your switch.
How soon do we see the first result?
Pick one question, load one batch of existing documents; it runs overnight and we look at it together the next day.
What hardware is needed?
One machine inside your network is enough to start; add more as the volume grows. Sized to your document volume.
Can it replace our ERP / workflow system?
No. It reads your documents, finds problems and pushes handling; it does not replace bookkeeping or workflow systems, and it does not score employees.
Are there format requirements?
PDFs, images, Office files, CSV and archives as they are; scans are OCR'd first. Anything unreadable is marked unreadable, not guessed.
Run one batch of your own documents first
Pick the one question you most want answered; we install it on your machine, run it overnight, and look at the results with you the next day.