ProofRoom — Document & script intelligence

PRODUCT // PROOFROOM

The judgment lives in people's heads. ProofRoom makes it yours.

ProofRoom is an AI platform for analyzing scripts and long documents. It runs structured analysis modules against a script or contract, lets the reviewer refine conversationally, and learns your house methodology so its judgment becomes yours.

The value was never the summary. It is the opinionated, house-specific assessment — and the ability to defend it. ProofRoom is built so every analysis is evidenced, versioned, and reproducible.

DOCUMENT-INTELLIGENCEREPRODUCIBLEMETHODOLOGY CAPTURE

FIG. SIGNAL_FROM_NOISE

REF: PROOFROOM-00 Signal pulled from noise: a coherent assessment resolving out of unstructured text

01 // The problem

Reading for quality is slow, expensive, and inconsistent.

A development executive spends the better part of a working day producing coverage on a single screenplay. A legal reviewer reads a long agreement clause by clause. The judgment is real expertise, but it lives in people's heads, varies between readers, and does not scale.

Generic "summarize this PDF" tools miss the point. The value is not the summary — it is the opinionated, house-specific assessment, and the ability to defend it later when someone asks why.

REF: PROOFROOM-01

// COST OF MANUAL REVIEW

~10 HRS OF EXPERT ANALYSIS PER SCRIPT

JUDGMENT VARIES READER-TO-READER

METHOD UNDOCUMENTED // UNTRANSFERABLE

02 // How it works

Ingest. Analyze in modules. Refine. Teach.

ProofRoom ingests a script or document and runs analysis modules against it. Each module produces structured, evidenced findings rather than prose — so the assessment can be inspected, challenged, and defended.

REF: STEP-01

upload_file

Analysis modules

For entertainment work the modules cover story structure, character development, scene efficiency, and dialogue. The module set is configurable per customer and per domain — each one returns evidenced findings tied back to the source.

REF: STEP-02

forum

Conversational refinement

Ask follow-up questions, challenge a finding, or request a rewrite comparison between two versions. The reviewer stays in the loop and in command — the system argues its case and yields to yours.

REF: STEP-03

neurology

Extract insights

Point ProofRoom at examples of your own past assessments and it learns that house methodology. The analysis then reflects how this studio or this firm actually evaluates work — not a generic rubric. Know which knob to turn.

03 // The hard part

Not "just call an LLM."

What makes an assessment defensible is the engineering underneath it: reproducibility, a regression suite for judgment, and a way to encode method without touching prompts.

FIG. GOLDEN_CORPUS_HARNESS

REF: PROOFROOM-EVAL Evaluation ladder: graded exemplars scoring each change before it ships

REF: HARD-01

Versioned, reproducible analysis

Every analysis is pinned to a hash of its inputs — the analysis basis — so re-running is reproducible and any change is attributable to a specific input change, not silent model drift. This is what makes the output defensible.

SPEC_ID: ANALYSIS_BASIS // CONTENT-HASH PINNED

REF: HARD-02

Golden-corpus evaluation harness

A curated set of 10–20 graded reference scripts acts as a regression suite. When prompts or models change, the system is scored against the golden corpus before the change ships — so quality does not silently regress.

SPEC_ID: EVAL_HARNESS // 10–20 GRADED EXEMPLARS

REF: HARD-03

Methodology capture, not fine-tuning

"Extract insights" distils a customer's prior work into reusable analytical guidance the modules apply — letting a non-engineer encode house style without touching prompts. Long documents that blow past context windows are handled with chunking, summarization strategy, and PDF parsing as first-class engineering, not an afterthought.

SPEC_ID: LONG_DOC // CHUNK + SUMMARIZE + OCR FALLBACK

04 // Who it is for

REF: WHO-01

ENTERTAINMENT // ANCHOR VERTICAL

Film & TV script development

Built with a film/TV IP-licensing partner. Development executives and readers get house-consistent coverage in minutes instead of a working day — every finding evidenced and traceable to the page.

Story structure, character, scene efficiency, dialogue modules
House methodology learned from past coverage
Rewrite comparison between script versions

REF: WHO-02

LEGAL // SECOND VERTICAL

Long-document review

The same engine — modules, conversational refinement, methodology capture, reproducible analysis — extends directly to legal document review. Read a long agreement clause by clause with a firm-specific rubric, not a generic one.

Domain-configurable analysis modules
Reproducible, attributable findings for audit
OCR fallback for difficult PDFs

05 // Status & stack

Live, with revenue. The flagship.

ProofRoom is in active daily use by its anchor customer, with ongoing iteration. It is the clearest example of the agency-product flywheel: built for a paying client, now being generalized into a product.

The stack is built for self-hosted and containerized deployment — Next.js, PostgreSQL with Drizzle, multi-provider LLM access with Claude as the default analyst, OCR fallback for difficult PDFs, and a configuration registry that layers settings so behaviour is tunable per deployment. Tune to the right signal.

REF: PROOFROOM-STATUS

// DEPLOYMENT PROFILE

STATUS: LIVE // DAILY USE // REVENUE

STACK: NEXT.JS / POSTGRES + DRIZZLE

LLM: MULTI-PROVIDER // CLAUDE DEFAULT

DEPLOY: SELF-HOSTED / CONTAINERIZED

EVAL CORPUSPROVIDER-AGNOSTICVITEST + PLAYWRIGHT