Case Study 02: Following the Thread

How to build a narrative from a single line of inquiry.

1 keyword, 355 signals, 10-year timeline: the Euston cost escalation.

All evidence excerpts in this case study are unaltered Accounter output, validated at the character level against source documents.


The Starting Point: A Thread Worth Pulling

In Case Study 01: Navigating Complexity, an investigator used Accounter to navigate a 2,220-page document set, turning an overwhelming volume of information into a clear map of the key issues. From this map, one name kept appearing: Euston station. The budget blowout, the construction pause, the £6.5 billion in cancelled scope—all tied to this single location.

This case study demonstrates the next step: following that thread. The goal is to move from high-level discovery to a detailed, evidence-grounded timeline of a single, critical issue.

The Challenge: Connecting the Dots Across Time

The Euston story isn't in one document; it's scattered across dozens of reports, committee transcripts, and audits published over nearly a decade. Manually piecing together this timeline is a painstaking process of searching, cross-referencing, and reading, with a high risk of missing crucial context. How can an investigator efficiently build a complete and verifiable narrative?


The Keyword Run: Focused Extraction

Accounter supports different run modes. For example, a Discovery Run (used in Case Study 01) scans the full corpus for substantive findings. A Keyword Run narrows the focus: before processing, the investigator specifies one or more search terms, and Accounter extracts only passages where those terms appear.

Here, the investigator configures a Keyword Run with a single term: euston. The same 60 documents are processed, but this time Accounter is looking for a specific thread. The results are added to the same permanent Signal Ledger—the evidence base grows with each run.

Step Action Outcome
01 Configure Run Set run mode to Keyword. Enter search term: euston.
02 Run Extraction Accounter processes all 60 documents, extracting only passages mentioning Euston. This run: 90 seconds.
03 Review the Ledger The Signal Ledger now contains 355 Euston signals—tagged and rated by salience.
04 Sort Chronologically Switch the ledger view to sort by document date, creating a timeline spanning 2013–2025.

Salience Breakdown (355 signals)

  • Exceptional: 6 signals — project cancellations, major budget blowouts
  • High: 133 signals — cost pressures, governance failures, delivery warnings
  • Medium: 148 signals — procurement updates, design changes, stakeholder concerns
  • Low: 68 signals — routine mentions, administrative references

Building the Thread

With 355 signals sorted chronologically, the investigator begins thread-building: selecting the signals that mark turning points in the Euston story. Scrolling through the timeline, they flag signals for inclusion—a 2016 warning here, a 2023 budget figure there. Each selection is a judgment call, but every option is grounded in verbatim evidence, not memory or notes.

This is the investigator's craft: curation. Accounter doesn't decide what's important. It ensures that whatever the investigator chooses to include is accurate, traceable, and ready to publish. Every quote can be copied directly into a story, with the source citation already attached. No transcription errors. No misattributed dates. No need to re-check the original document.

The selected signals form a curated thread—a focused narrative assembled from across the corpus. Below is the Euston thread, built from 10 key signals spanning a decade.

A note on synthesis

The 355 Euston signals are validated verbatim excerpts. But deciding which 10 signals tell the story? That required judgment — and I used an LLM to help identify narrative threads.

I gave Claude the full set of 355 signals and asked: "What's the through-line here?" It pointed to the budget escalation arc. I then manually selected the 10 signals that best illustrated that arc, verifying each one against the source documents.

The validation is mechanical. The curation is collaborative — human judgment guided by AI pattern recognition. This is probably where Accounter needs the most refinement: making the journey from "here are 355 true things" to "here's the story" less manual.

If you have ideas on how to improve this, I want to hear them.

The Euston Timeline: 10 Signals, 10 Years

From the curated thread, the investigator generates a Timeline Report—a structured, exportable document that presents the selected signals in chronological order. Each point below appears exactly as it does in the generated report: the signal label, the verbatim evidence excerpt, and the source citation—ready to copy, cite, and publish.


The Story Emerges

Reading the timeline chronologically, the narrative writes itself:

Every point in this narrative is grounded in verbatim evidence. The investigator didn't have to infer the story—they followed it through the documents.

Explore the Full Euston Corpus

The 10 signals above are the curated thread—but the investigator has access to all 355 Euston-related signals in the Findings Ledger. Below, explore the complete keyword search results: filter by source document, by theme, or by salience to discover additional context or find signals for follow-up investigation.

Explore the evidence for yourself (355 signals)

Browse the complete ledger from both runs. Toggle between All Findings, Discovery Run only, and Keyword Run only to see how the second pass enriched the investigation.


The Outcome: A Publish-Ready Timeline

With 90 seconds of processing time, the investigator has moved from a mountain of documents to a detailed, evidence-backed story.

Accounter is an amplifier, not a replacement.

Good investigation still requires judgment: which threads to follow, which evidence matters most, how to frame the story. Accounter doesn't make those decisions—it makes them possible by ensuring the investigator has seen everything relevant and can trust every quote they use. The craft remains human. The drudgery doesn't have to be.

Accounter didn't "write" the story of Euston. It made the evidence so accessible and navigable that the story became impossible to miss. It changes what is humanly possible to review, allowing an investigator to focus on the highest form of analysis—connecting the dots and holding power to account—rather than the manual labour of reading and searching.


Also see

Case Study 01: Navigating Complexity — 60 documents, 2,220 pages, evidence-first navigation.

Previous case study

Get in touch

I'm looking for 3-5 pilot collaborators. Tell me about your document corpus and what you're trying to understand.

I'm interested in