Explore broadly
Surface risks, tensions, and deviations across entire document sets — even when they are buried or euphemistically phrased.
How it works
From documents to a permanent, navigable evidence base.
Accounter reads your full document corpus, extracts substantive claims with verbatim evidence, validates every excerpt character-by-character, and produces a structured ledger you can filter, search, and return to indefinitely.
How a run works
You upload your documents. Accounter does the reading. Every signal it surfaces is backed by a verbatim excerpt that has been validated against the source text. Here is what happens at each step.
| Step | What you do | What happens |
|---|---|---|
| 01 | Build your library | Upload PDFs. Metadata is extracted automatically — document title, date, authoring body. You review and correct labels before processing begins. |
| 02 | Run extraction | Every page is read: body text, appendices, footnotes, extracted tables. Substantive claims are captured with verbatim source text. Nothing is skipped or weighted — appendices receive the same scrutiny as main sections. |
| 03 | Validation | Every excerpt is checked character-by-character against the extracted text. If a match fails, the signal is discarded. No exceptions, no "close enough" — either it is verbatim or it does not surface. |
| 04 | Review the ledger | Validated signals are auto-tagged by theme and rated by salience (Low, Medium, High, Exceptional). Filter by source, tag, date, or salience. Search, sort, export. The ledger is yours to explore. |
The ledger is permanent — not a one-time report. Every signal, every tag, every validation result is stored and searchable indefinitely. Return months later, add new documents, run new queries, or follow a new thread — using the same source material and the same evidence rules.
Ways to discover
Accounter lets you explore large document sets from multiple angles — without changing methodology or losing traceability.
Focus on a topic
Examine how a specific issue (e.g. finance, safety, governance) appears across documents, committees, or time.
Follow a thread
Take a handful of notable excerpts and ask: "Where else does this kind of thing appear?" using similarity of language, context, and reference patterns.
Check explicit references
Find every direct mention of a name, project, or phrase — with page-accurate citations.
Rediscover at any point
Run a fresh discovery, narrow the scope, or follow a new line of inquiry — using the same source material and the same evidence rules.
The approach
Designed to increase the chance of surfacing important signals while keeping human judgment explicit.
Full documents
Body text, appendices, footnotes, extracted tables. No content weighting — appendices and footnotes receive the same scrutiny as main sections.
Optimised for coverage
Designed not to miss signals. You decide what matters; the system ensures you see the options.
Character-level validation
Every excerpt is validated against extracted text. No signal surfaces without a validated verbatim match.
Why it complements manual review
Accounter does not replace human judgment — it changes what is humanly possible to review. Different failure modes. Use both.
Reads exhaustively
- Footnotes, appendices, extracted tables
- No fatigue, no time pressure
- Does not skip what looks routine
Surfaces what you didn't search for
- Keyword search won't find what you don't know
- Surfaces related issues even across documents
- No need to guess the right terms upfront
Preserves the evidence
- Every signal linked to verbatim text
- No inference, no judgment, no hallucination
- Structure and validation that raw AI lacks
Fast and repeatable
- Processes years of documents in minutes
- Consistent, structured outputs across runs
- Built to support longitudinal comparison
See it in action
Two case studies using real UK government documents from the HS2 rail project demonstrate Accounter on a 60-document, 2,220-page corpus.
Case Study 01: Navigating Complexity
60 documents, 2,220 pages, 2,253 signals. Demonstrates broad discovery across a large corpus — navigating by source, tag, and salience to find the story.
Case Study 02: Following the Thread
1 keyword, 355 signals, 10-year timeline. Demonstrates focused thread-following — from a single keyword run to a publish-ready narrative about the Euston cost escalation.