Week 10 · lesson

Lab: Audit Entity Spans

Mission

Run the local Entity Span Audit Lab. Compare three supplied prediction sets against one gold BIO sequence, reproduce exact-span metrics, and export an Entity Span Audit Record.

Guided start

Select Exact match and run the audit.
Confirm three true positives and F1 1.000.
Select Missed date and identify the false negative.
Select Boundary error and find both the false positive and false negative.
Explain why the location error counts twice in the span comparison.

Procedure

Preserve the token sequence, entity schema, and gold tags.
Predict TP, FP, and FN for all three supplied outputs.
Run each audit and preserve decoded spans and metric operands.
Compare missed-entity and boundary-error consequences.
Preserve your first valid-output claim and write a bounded correction.
Complete Observation, Measurement, Decision, and Limitation.
Export the record and confirm it names ai-for-beginners-entity-span-audit-lab.

Success criteria and boundary

gold and predicted BIO tags remain visible;
decoded spans preserve boundaries and types;
every precision, recall, and F1 value is reproducible;
boundary errors are not hidden by mostly correct tokens; and
the limitation rejects claims about fluency, factuality, privacy, safety, or performance on other text.

Use only supplied fictional text. The Lab sends no text to a model or service.

Robotnix Lab · contract v1

Entity Span Audit Lab

Decode BIO tags, audit exact entity spans, and calculate extraction precision, recall, and F1.

Help and boundaries

A deterministic browser simulation uses supplied fictional text and prediction sets and contacts no model or service.

No personal data, private prompts, or API keys. This Lab makes no hidden remote writes and does not persist your work after the page session.

No result processes personal data, invokes a model, or authorizes action.

Lab draft is active for this page session only.

Objectives

Decode BIO labels into typed spans.
Calculate exact-span precision, recall, and F1.
Separate valid formatted output from factuality, privacy, safety, and general performance.

Procedure

Inspect schemaRecord tokens, PER/LOC/DATE types, and gold BIO tags.
Decode spansPreserve gold and predicted span boundaries and types.
Calculate metricsCount TP, FP, and FN and reproduce precision, recall, and F1.
Bound extraction claimSeparate supplied audit results from fluency, factuality, privacy, safety, and other-text performance.

Workbench

Decode exact spans before trusting extraction output.

Prediction set

Select a supplied prediction set and run the exact-span audit.

Boundary: This fictional audit does not invoke a model. Its results do not establish factuality, privacy, safety, or performance on other text, schemas, prompts, or systems.

Text and static fallback

Every token, tag, decoded span, count, metric, and boundary appears in accessible text and tables.

Exact F1 1.000; missed-date F1 0.800; location-boundary-error F1 0.667.

Evidence record

Record only the evidence requested. Do not enter names, personal information, private prompts, or credentials. This draft stays in memory only unless you explicitly export it.

Schema and tags *Record tokens, declared types, gold tags, and each predicted tag set.Span metrics *Record decoded spans, TP, FP, FN, precision, recall, and F1 for all presets.Extraction claim repair *Preserve and repair one valid-output-means-correct claim.Limitation *State what the fictional audit cannot establish about a live language model or other text.

Success criteria

The entity types, token sequence, and gold labels are explicit.
All spans and metrics are reproducible from displayed evidence.
The correction follows exact-span evidence only.
The limitation rejects unsupported factuality, privacy, safety, and general-performance claims.

Complete every procedure step and required evidence field before export.

Import evidence JSON

Vocabulary lab

Flip the idea, not just the card

Explain the term before you reveal the back. Then compare your explanation with the definition, example, and warning.

Read all terms without animation

true positive: A predicted entity whose boundary and type exactly match one gold entity. Example: Rin/PER matches in all presets. Do not confuse it with: A partial span is not a true positive.
false positive: A predicted entity without an exact gold match. Example: Harbor/LOC is a false positive when the gold span is Harbor City. Do not confuse it with: It can coexist with a false negative for the intended span.
false negative: A gold entity without an exact predicted match. Example: Monday/DATE is missed in one preset. Do not confuse it with: It measures omission, not extra output.
F1 score: The harmonic mean of precision and recall. Example: Two TP, one FP, and one FN produce F1 0.667. Do not confuse it with: It does not describe every error consequence.
Entity Span Audit Record: Evidence preserving schema, tags, spans, metrics, claim repair, and limitation. Example: The Lab export. Do not confuse it with: It is not evidence from a live language model.