▪
Forecast
Predict expected audience reception band before production — given screenplay strength and planned cast, director, release window, and distribution. Forecasts validated within ±5pt of actual outcome on retrospective tests.
ReelPen runs on a wholly proprietary algorithmic system, designed and built in-house. No open-source models, no third-party LLM wrappers, no off-the-shelf scoring components. What we disclose: the inputs, the outputs, the validation results, and how to read them. What we don't disclose: the internal weights, formulas, or calibration parameters — those are exclusive ReelPen IP.
Every analysis follows the same five-stage pipeline. Each stage's intermediate output is preserved so the final verdict can be re-derived from raw inputs at any time. Re-running the same submission next quarter produces a diff, not a fresh start.
PDF, FDX, Fountain, or DOCX parsed into a structured representation. Scene boundaries, character cues, dialogue, and action are isolated.
PDFFDXFOUNTAINDOCXPosition-anchored beat detection across 11 industry storytelling frameworks. Identifies primary structural canon, deviations, and whether deviations are intentional genre conventions or weaknesses. Source-cited academic diagnostic.
ReelPen Structure Engine11-framework canonSix dimensions scored by ReelPen's proprietary algorithm, then anchored against canonical industry works. Tier-classified into Canonical / Distinguished / Certified / Verified / Develop / Reject. Independently blind-validated on a sealed holdout corpus — mean prediction error sub-1-point.
ReelPen Anchor Systemproprietary weightingEvery flag, every score, every recommendation is attached to a specific page, beat, or scene reference. Audit log signed cryptographically.
page-citedSHA-256Signed PDF report, structured JSON, and direct API hand-off. Diffs against prior drafts available from the dashboard.
PDFJSONREST APIEach module is an independently-trained component of ReelPen's proprietary system. Below are the capabilities most often deployed by acquisition teams, lit agents, and development executives.
Predict expected audience reception band before production — given screenplay strength and planned cast, director, release window, and distribution. Forecasts validated within ±5pt of actual outcome on retrospective tests.
Identify structural weaknesses, pacing risks, character-arc completion, and dialogue craft issues — page-cited to the exact moment that triggered each flag. The basis for prioritized rewrite recommendations.
Calibrate every score against canonical industry excellence. Tier-classified into six bands from Canonical (the cinematic standard) to Reject (developmental). Resists evaluator drift; reproducible to ±2pt across runs.
Independent blind-validation across an unseen corpus. Mean prediction error sub-1-point. Every score, every flag, every comparison signed cryptographically and defendable in greenlight committee.
The Greenlight Score is the output of ReelPen's proprietary scoring algorithm — a six-dimension assessment combined under our internally-developed weighting and calibration system. The exact formula, weights, and calibration parameters are proprietary intellectual property and are not disclosed externally.
What you can rely on: every component score is reproducible, every flag is page-cited, and the entire system is blind-validated against unseen scripts (Phase Q: 407 sealed holdout, mean error 0.72pt). Re-running the same submission next quarter produces a result within ±2pt of the original.
ReelPen's analytical engine is a wholly proprietary algorithmic system, designed and built in-house by our research team.
| Dimension | Score | Verdict |
|---|---|---|
| Concept | 78 | Top quintile |
| Structure | 68 | Late midpoint flagged |
| Character | 82 | Strong arcs |
| Dialogue | 74 | Distinct voice |
| Theme | 71 | Consistent through act 3 |
| Marketability | 65 | Theatrical-first viable |
| Composite (G) | 73 | Verified tier |
Every score, citation, and recommendation in your report is hashed, signed, and replayable. Diff your draft against last quarter's version. Defend any flag in a greenlight committee meeting with the exact artifact that triggered it.
Submitted IP is encrypted in transit and at rest, never used for model training, and purged on request. Enterprise contracts include explicit data-residency, on-prem analysis, and DPA terms. NDAs countersigned within 24 hours.
28 pages. Phase Q blind validation results, statistical performance, calibration discipline, and benchmark composition (without disclosing proprietary internal parameters). Available under enterprise NDA.
Request validation report (NDA) →