Short answer
SOPs, work instructions and checklists are different kinds of training data. An SOP describes a whole process: purpose, roles, order and controls. A work instruction tells one person at one station exactly how to do one task, often with photos, tools and parameters. A checklist lists what must be verified, and completed checklists record what was done. Each feeds a different model layer and needs different preparation.
Why the document type matters before you buy
Document labels are inconsistent across businesses. One site calls everything an "SOP"; another files station instructions as "job aids" and checklists as "forms". A corpus described as two thousand SOPs might hold a few hundred process documents, over a thousand station-level instructions and a stack of checklists, and each of those groups yields different labels at a different preparation cost.
Ask for an inventory by type before you price or plan anything. The general limits of procedure text, chiefly that it holds no motion, force or perception data, are set out on what SOPs can and cannot teach a humanoid; this page is about telling the document types apart and buying the right one for each use.
How the three types differ
| Dimension | SOP | Work instruction | Checklist |
|---|---|---|---|
| Scope | A process across roles and stations | One task at one station or machine | The points that must be verified |
| Reader | Supervisors, auditors, every role involved | The person doing the task | The person doing or checking the task |
| Granularity | Stages: receive, inspect, store | Actions with parameters: "torque to the card value" | Yes/no items: "guard closed" |
| Typical content | Purpose, scope, responsibilities, references, records | Numbered steps, photos, tools, parts, settings, quality points, cautions | Item, criterion, sign-off field |
| Change rate | Slow | Often, with each product or equipment change | Moderate |
A fourth category sits next to these: records. Completed checklists, batch records, sign-offs and deviation reports are evidence of execution, not instructions. They are often the most valuable item in a procedures purchase, because they show how often a task runs, how often it fails and whether practice matches the document.
Many quality management systems arrange these in tiers: policies at the top, then procedures, then work instructions, then forms and the records made by filling them in. ISO 9001 is generic and applies to any organization, and its requirements include documented information as part of support (ISO, 2026); how sites map that requirement onto tiers varies, as covered on quality system document hierarchies.
Regulated procedures come with records
In regulated industries the pairing of instruction and record is built in. US drug manufacturing rules require written production and process-control procedures that the quality unit approves, that are followed and documented at the time of performance, and under which any deviation is recorded and justified (FDA via eCFR). A supplier operating under such rules may hold procedures, execution records and deviation logs for the same tasks.
Safety procedures are another case. OSHA's lockout/tagout standard requires employers to develop, document and use energy-control procedures that cover scope, purpose, authorization, rules and techniques, plus specific shutdown, isolation and securing steps, with a narrow documentation exception (OSHA). In documents like these, a missing step is a safety failure, so flag them for the treatment in safety-critical steps in procedures. This is not legal advice; the point is only that regulated settings often keep procedures, records and deviations together.
What each type can feed
| Type | Best model use | Labels you get almost free | Main preparation | Cannot provide |
|---|---|---|---|---|
| SOP | Task decomposition, role and handoff planning, coverage scoping | Task names, stage order, roles, links to instructions | Extract steps from prose; resolve cross-references | Station-level actions; parameters are often absent |
| Work instruction | Step-level grounding, object and tool vocabulary, parameter constraints, image-text pairs | Step text, tool and part lists, photos, settings, quality points | Parse tables and photos; redact confidential values; match revisions | Motion, force, tacit technique |
| Blank checklist | Success criteria, evaluation rubrics, state checks | One criterion per verifiable state | Turn vague items such as "area clean" into observable criteria | How to reach the state |
| Completed record | Task frequency, failure rates, timing, exceptions | Outcomes, timestamps, sign-offs | De-identify signatures; link each record to the revision in force | Proof that the record is accurate |
Work instructions are usually the richest source for grounding, especially when each step carries a photo; see visual work instructions. Checklists convert most directly into success labels, covered on checklists as success criteria, and completed checklists are treated as data on completed checklists as execution records.
Structured sources reduce parsing
Most operating businesses write procedures in word processors and spreadsheets, then publish PDFs. Recovering steps, cautions and tables from those files is the main preparation cost.
Some documentation teams author in structured standards instead. DITA defines topic-oriented document types and a task topic specialized from the general topic, with general and strict task variants (OASIS DITA TC). S1000D stores procedural data modules in a common source database, each identified by a unique data module code (S1000D Steering Committee). Exports from such systems keep steps and their supporting information separate at the source, which removes much of the parsing; structured authoring exports covers what to request.
Illustrative: one task written three ways
# Illustrative only — one synthetic task as it might appear in three document types
task: replace the label roll on case labeler CL-2 (site_009)
sop_excerpt: # SOP-PK-003 "Packaging line changeover", rev F
- "Line operators replenish labels, tape and film at changeover or on a low-supply alarm."
- "Follow WI-PK-031 for label replenishment; the line lead verifies with CL-PK-031."
work_instruction: # WI-PK-031 rev C, 7 steps, 5 photos
- {n: 3, text: "Open guard door B. Lift the dancer arm until it latches.", photo: WI-PK-031_p3.jpg}
- {n: 4, text: "Slide the new roll onto the spindle, labels facing out, until it touches the stop."}
- {n: 5, text: "Thread the web along the path marked in yellow.", photo: WI-PK-031_p5.jpg}
checklist: # CL-PK-031, completed copies kept for 90 days
- "Roll seated against stop (Y/N)"
- "Web follows yellow path with no twist (Y/N)"
- "Test label applied level inside the marked box (Y/N)"
yields:
sop: trigger, roles, links to the other two documents
work_instruction: action sequence, object names, photos for grounding
checklist: three observable success criteria; completed copies give outcome rates
missing_from_all_three: arm motion, threading dexterity, the force needed to seat the roll
The cross-references are as valuable as the content. When an SOP names the instruction it relies on and the checklist that verifies it, you can join the three into one structured task record, such as the one described in the procedure step schema.
Implications for what to buy
- For planners and coverage maps, SOPs plus a task list are often enough.
- For instruction grounding and image-text pairs, prioritize work instructions with photos and current revisions.
- For evaluation and success detection, buy checklists together with completed records, so criteria and outcomes arrive linked.
- For practice-match evidence, ask for completed records and deviation reports alongside the procedures they refer to.
- For physical skills, none of these suffice; budget demonstrations and robot data for every step they name.
Limitations and preparation
Types blur in practice. Work instructions hide in SOP appendices, checklists are printed at the end of work instructions, and some "checklists" are really abbreviated instructions. Vague criteria need rewriting before they can serve as labels. Records can be filled in after the fact or without checking, so treat them as evidence of what was reported. Signatures and initials in records are personal data and need de-identification.
Questions to ask suppliers
- How many documents of each type do you hold: process SOPs, work instructions, checklists and completed records?
- Do SOPs reference the work instructions and checklists they rely on, by ID?
- What share of work instructions include photos or diagrams, and at what resolution?
- Which revision is current, and are superseded revisions kept with their dates?
- Are completed checklists or batch records retained, for how long, and in what format?
- Are deviation or nonconformance reports linked to the procedures they concern?
- Is any documentation authored in a structured format such as DITA or S1000D?
- Which documents were written by equipment makers, consultants or customers rather than your own staff?
The last question matters for licensing; see who owns a procedure.
Discuss your requirements
If you are scoping a procedures purchase, tell SourceX which document types, tasks and sites you need and which model layers they will feed. SourceX looks for US businesses that hold suitable documents, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. It reviews each requirement and follows up about potential sources; a request does not guarantee a match. The procedures hub and the humanoid training data guide cover the rest of the section, and the catalog describes SOPs, playbooks and knowledge bases and licensing SOPs and playbooks.
Explore sourcing and licensing options with SourceX
Sources
- ISO 9001:2026 Quality management systems - Requirements — generic quality requirements including documented information.
- 21 CFR 211.100 Written procedures; deviations — approved written procedures, documentation at time of performance and recorded deviations.
- 29 CFR 1910.147 The control of hazardous energy (lockout/tagout) — required content of documented energy-control procedures.
- DITA (Darwin Information Typing Architecture) - OASIS DITA Technical Committee — topic-oriented authoring with a task topic type.
- S1000D International specification for technical publications using a common source database — procedural data modules identified by code.