Observe

W03 · 5 h 10 min

Discovery, indexing, and retrieval

Essential question

Can the system obtain the source at all?

What you should be able to do

  1. Separate crawl/index eligibility from retrieval.
  2. Compute a first-stage retrieval baseline.
  3. Name the denominator at every candidate-loss point.
PrerequisitesBasic Python and command-line file inspection.Bridge note on inverted indexes, BM25, and Recall@k.
Builds on

Basic Python and command-line file inspection. · Bridge note on inverted indexes, BM25, and Recall@k.

Investigates

Can the system obtain the source at all?

Feeds

L03 retrieval sandbox I.

Arrive with a prepared artifact

Reading route
Core PAPER-06/PAPER-10; Extend PAPER-23; protocol/platform route PLAT-01/PLAT-05/PLAT-07/PLAT-09/PLAT-11/PLAT-12.
Viewing route
Open the complete text-first lecture packageSeven-minute bridge: inverted index, document score, candidate recall. The current equivalent is notes, slide script, worked case, and no-video transcript; no recording is claimed.
Readiness check
Calculate Recall@5 from one supplied qrel and ranked list.
Bring
A one-page pipeline sketch with measurable and unknown denominators.

Concepts, assumptions, and boundary

Eligibility is not exposure

Robots rules, sitemaps, HTTP responses, canonicalization, and indexing policies affect whether a source can enter a system’s available corpus. None guarantees that a particular query retrieves it. Retrieval then maps a query to candidates using sparse lexical matching, dense representations, or hybrid combinations. A useful audit records counts at each observable boundary—submitted URLs, successful fetches, indexed documents, retrieved candidates—and does not replace an unknown closed-platform denominator with a convenient proxy.

Baselines make failure visible

A sparse baseline such as BM25 is not included because it is always best. It provides an interpretable reference: term overlap, document length, and query wording can be inspected directly. Dense retrieval can improve semantic matching but introduces model, pooling, domain, and embedding-version dependencies. Hybrid retrieval can inherit strengths and failure modes from both. Topic-level error analysis matters more than a single mean because a system may gain on paraphrases while losing exact entities, dates, or rare terminology.

Inspect the mechanism or evidence structure

MechanismCandidate funnel with explicit loss points
Candidate funnel with explicit loss pointsDocuments narrow through four stages. Beside every narrowing is a count pair and a possible failure class. Unknown closed-system counts remain visibly blank rather than being estimated from the final response.Crawl eligiblen / NIndexedn / NRetrieved top-kn / NContext candidaten / N
Figure design. A four-stage funnel for crawl, index, retrieve, and context candidacy; each stage has numerator/denominator placeholders and a named “unknown” state. Each narrowing needs its own numerator, denominator, and unknown state.
Long description

Documents narrow through four stages. Beside every narrowing is a count pair and a possible failure class. Unknown closed-system counts remain visibly blank rather than being estimated from the final response.

One action, one feedback state

Action

Compare frozen BM25, dense, and hybrid top-k run files by topic.

Feedback

Selecting a topic reveals false negatives, vocabulary mismatch, and whether the difference occurred before or after candidate generation.

Accessible alternative

All ranked lists and metrics are duplicated in a sortable text table.

Open the visibility-chain diagnostic

Produce a reviewable intermediate file

Task
Run or inspect sparse, dense, and hybrid retrieval on the frozen mini-corpus.
Inputs
Corpus, topics, qrels, precomputed runs, CPU script, environment lock.
Timebox
75 minutes
Intermediate file
three TREC-format run files and one topic-level error table
Stop condition
Do not download a model if the environment or license differs from the declared route; switch to supplied runs.

Why each controlled source is here

Complete, retrieve, and revise

Checkpoint
L03 retrieval sandbox I.
Reflection
Identify one failure caused by vocabulary and one caused by corpus absence.
Revision
Attach the correct candidate-set denominator to every Recall@k statement.
Low-compute route
Use supplied run files and the metric script; no model or index download is required.

Five retrieval questions

01Why can a crawlable page still have zero retrieval exposure?

Eligibility is only an upstream condition; indexing, query matching, ranking, freshness, and context selection can each exclude it.

02What does Recall@k require?

A declared relevant set, a ranked candidate list, k, topic aggregation rule, and treatment of unjudged documents.

03What is the purpose of the BM25 baseline?

It provides an interpretable lexical reference and error surface, not a universal claim of superiority.

04What is the evidence boundary for W03?

Platform and protocol documents describe named surfaces and rules; they do not establish retrieval effectiveness or a visibility guarantee.

05What must you submit or revise after this week?

L03 retrieval sandbox I. Attach the correct candidate-set denominator to every Recall@k statement.

Continue in the practice package