Discover · Search · Term Hit Report · Saved Searches

Show the tree, end the argument

A query builder and raw query that are two views of one search, the parsed tree of what actually executed shown under the results, a live 'how many does this find?' count — and a term hit report built for the meet-and-confer table.

Grotabyte — Search

"side letter" w/5 guarantee AND custodian:(smith OR chen)

How many does this find?1,204 documents · 0.03 s

RE: side letter — fee guarantee2024-11-02 · sealed

…we can guarantee the fee via a side letter if…smith

契約書の修正版2025-03-14 · sealed

CJK indexed per character — found exactly, not skipped

Q3 pricing addendumsemantic · sealed

concept match — no shared keyword with the query

phrase · wildcard · fuzzy~ · w/5 · pre/5 · date ranges · 13 fields

Where it starts

Terms are being negotiated, and every number you quote will be quoted back at you. The workflow starts before the first search: agreeing what a term means, and being able to prove what ran.

How it runs

01
Build it, or type it — same search
The builder ('Find documents where…') and the raw query are two views of one search; edit either and the other follows. 'How many does this find?' answers before you commit to reading anything.
02
Read the tree that actually ran
Under the results sits the parsed tree of the executed search, with its features named: proximity w/N, ordered pre/N, phrase, wildcard, fuzzy, NOT, date range, field scoping. When somebody disputes whether the agreed protocol was followed, showing the tree ends the argument.
03
Run the term hit report
One term per line — phrases, prefixes, fuzzy. Per term: hits, threads, incremental contribution in list order, share of corpus, timing. A term adding nothing gets a 'no new docs' chip — the number that moves negotiations.
04
Save the search at the moment it is agreed
A saved search keeps the query text, the parsed tree, and the hit count at agreement. Re-run it later and drift is computed, not remembered: 'forty thousand when we agreed it, sixty thousand now' becomes a citation instead of an argument.
05
Act on the whole set
From the same results bar: place a hold, produce, export, code the set, or record the search against a public-records request. The query travels with each act, so scope stays reproducible.

Why it holds up

Term-report hit counts are exact and uncapped; derived figures state the population they were computed over, per row — a figure without its denominator is the thing this product does not ship.

Where a proximity query compiles to a superset, the result is verified back down to the real count — the figure that is negotiated over has to be the real one.

A tag is searchable the instant it is applied — a privilege call recorded is a privilege call findable.

CJK text is indexed per character and found exactly — a name in 漢字 is as findable as one in ASCII.

A saved search is a query, not a document list — new material matching the description is covered on re-run.

What you hand the regulator

A workflow that ends on a screen isn’t finished. This one ends in a document.

The term hit report

The table you slide across the meet-and-confer table: each side's terms with their real costs, ready to negotiate from evidence.

per term: hits · threads · incremental docs in list order · % of corpus
combined unique documents and corpus share
per-row basis: exact or capped, stated
the executed query tree, disclosable on demand

Adjacent workflows

Early Case Assessment TAR / Continuous Active Learning Productions

See it run on your data scenario

The demo form asks which workflows you want to see — name this one and we’ll stage it.

Book a Demo All workflows

How it runs

Build it, or type it — same search

The builder ('Find documents where…') and the raw query are two views of one search; edit either and the other follows. 'How many does this find?' answers before you commit to reading anything.

Read the tree that actually ran

Under the results sits the parsed tree of the executed search, with its features named: proximity w/N, ordered pre/N, phrase, wildcard, fuzzy, NOT, date range, field scoping. When somebody disputes whether the agreed protocol was followed, showing the tree ends the argument.

Run the term hit report

One term per line — phrases, prefixes, fuzzy. Per term: hits, threads, incremental contribution in list order, share of corpus, timing. A term adding nothing gets a 'no new docs' chip — the number that moves negotiations.

Save the search at the moment it is agreed

A saved search keeps the query text, the parsed tree, and the hit count at agreement. Re-run it later and drift is computed, not remembered: 'forty thousand when we agreed it, sixty thousand now' becomes a citation instead of an argument.

Act on the whole set

From the same results bar: place a hold, produce, export, code the set, or record the search against a public-records request. The query travels with each act, so scope stays reproducible.