Messy documents.
Training-grade truth.
This is a deep document intelligence engine, not a single extractor and not a single summary pass. It preserves what matters, exposes what is weak, and gates what can safely move downstream.
One run view from upload to readiness decision.
Phase-6 NKU Checkpoint
Procedure-grade source with strong sequence retention and clear operational truth.
Critical procedure steps and triggers are consistently preserved across extraction, recovery, and NKU refinement.
Accepted core is strong; weak units are contained and do not threaten downstream planning.
ML improves judgment quality.
Source truth stays in control.
Noisy source signals enter. Trust-calibrated judgments come out.
Scormy does not use ML as a black-box replacement for source truth. It uses ML to make the document engine more reliable on noisy, real-world material and to improve the quality of decisions before training generation begins.
In practice, ML is used where it creates leverage: evaluating extractor behavior, surfacing likely missing critical truths, strengthening coverage decisions, and improving NKU quality judgment over time. The output is not blind generation. The output is better trust calibration.
Scoring extractor behavior across noisy structures
Predicting likely missing critical truths
Improving coverage and recovery decisions
Evaluating NKU quality signals
Optimizing signal weighting over time
How it works.
Click each stage to inspect the full workflow detail and release logic.
Upload and Tenant Admission
The document enters as a controlled source asset, not just a file dropped into storage.
At this point Scormy anchors identity, ownership, and provenance. The system establishes the trust envelope that everything else depends on.
- Tenant scope and run context are fixed at intake
- Source fingerprint and version identity are recorded
- Every later decision can be traced back to this source record
A trustworthy document identity is established, so extraction can be judged against the real source and not a drifting copy.
Normalized knowledge, validated and tightened before release.
From extracted text to operational knowledge primitives
Once critical content has been identified, coverage-checked, and sharpened, Scormy compiles it into Normalized Knowledge Units: a structured internal layer built for planning, validation, and training design. These are not loose chunks. They are typed, inspectable units that preserve what the document is actually telling the learner to know, do, decide, or verify.
This is the point where the system stops reading the document as pages and sections, and starts operating on teachable truth: enforceable rules, executable procedures, accountable responsibilities, explicit purpose statements, and visible support gaps. Internally, this typed NKU surface behaves like an RP2S frame: Rules, Procedures, Responsibilities, Purposes, and Support Gaps held in one inspectable knowledge layer.
Rules
Non-negotiable constraints, obligations, and decision boundaries that must survive transformation.
Procedures
Ordered operational flows the learner must execute correctly, not just recognize.
Responsibilities
Named actors, ownership lines, and accountability handoffs that shape real-world action.
Purposes
The explicit reason a step, control, or requirement exists so training keeps intent, not just wording.
Support Gaps
Visible weak spots where the source does not support safe downstream confidence yet.
Downgrade
Low-value generic units are reduced in influence
Merge
Duplicate or overlapping units are consolidated
Split
Overloaded multi-action units are decomposed
Route to Review
Uncertain units are isolated from automatic downstream use
Useful material is preserved, weak material is visible, and no unit is treated as equally trustworthy by default.
Source truth comes first. It shows which arguments the document supports before training content is approved.
Prompt-based systems can produce polished training language from partial context. Scormy is built to do the opposite: preserve source truth, constrain output to what the document really supports, and expose weakness before it becomes training.
| Axis | Source truth | Prompt-first |
|---|---|---|
| Truth of source | If the source does not support the claim, the claim does not ship. | Infers likely meaning from nearby text and still turns it into training output. |
| Duration as a constraint | Training length is earned from source value, not inflated to hit an arbitrary runtime target. | Expands thin material into lesson-shaped content so the output feels complete. |
| Real knowledge | Preserves operational truth: what to do, who does it, when it happens, and what blocks progression. | Replaces procedure logic with polished summary language and generic explanation. |
| Hallucination tolerance | Unsupported statements are rejected, recovered, or held back from downstream use. | Unsupported claims can survive if they sound coherent and relevant to the prompt. |
| Failure visibility | Missing truths stay visible as weak, review-required, or unsupported states. | Fluent output hides where extraction failed or where evidence disappeared. |
| Procedure integrity | Keeps ordered action flow, release conditions, and stop states intact before planning begins. | Breaks step order and procedural dependencies into summary-style teaching language. |
| Exception handling | Preserves edge cases as first-class operational truth because they change what the learner must do. | Treats edge cases like optional footnotes and often loses them in the main narrative. |
| Actor accountability | Tracks actor, responsibility, and trigger relationships as explicit knowledge units. | Blurs who owns a step, who approves it, and who escalates it. |
| Release discipline | Uses readiness and confidence gates to stop weak documents before they become training. | Lets coherent-looking output move forward even when extraction quality is weak. |
The output is source truth turned into training: evidence-backed, no filler, no padding, no hallucination, only real knowledge.
Scormy ONE is the Orchestration Narrative Engine behind training generation.
Nothing moves downstream on fluency alone.
Every accepted training claim stays attached to source evidence before it is allowed into narrative planning.
Weak or partially preserved truths are repaired before they are allowed to shape modules, scenes, or assessments.
NKUs do not pass as one flat output dump; they are separated into confidence-aware layers the system can reason about.
Readiness is an operational gate, not cosmetic labeling, so weak documents can be stopped before they create polished but wrong training.
- Narrative-ready blueprint and module progression
- Scene and script candidates tied to source evidence
- Assessment intent linked to must-have truths
- Confidence-aware unit inventory
- Readiness report with review paths
- Evidence ledger snapshot for governance and compliance
Scormy One is the orchestration narrative engine that turns source documents into source-grounded, quality-gated, training-ready narrative structure.