Give the agent layered context

Cerca de 35 minutos min de leitura, com 9 exemplos em 9 seções.

Edição 2026.11.15 Exemplos aprovados em 19 de nov de 2026 78 de 78 exemplos passam O que mudou

Traduzido com IA. Verificado por pessoas em inglês. Como fazemos os livros

Traduzido do livro em inglês, edição 2026.11.15. Leia o original

Esta parte ainda não foi traduzida, por isso está em inglês.

By the end of this chapter, you can give a model that writes SQL more than the schema, in layers: notes on columns whose names mislead, a glossary of the shop's business terms, and verified queries retrieved for each question. You can fit those layers to a token budget, measure what each one adds on the dev split of the golden questions, and check the best set once on the held-out test split.

A schema can't say what its columns mean. Chapter 1 gave a model only the shop's schema and asked for net revenue in each region in March 2026. The query it wrote ran, looked right and counted cancelled orders. The rule it missed exists, in the finance team's definition of net revenue, but nothing the model was given contains it. Layered context is the most direct fix for what a schema can't say: write the rules down where the model reads them. It isn't the fix this book ends with, since chapters 5 and 6 move the definitions into a semantic layer the model can't misread. But any agent that writes SQL needs context, including the guarded one in chapter 7, and the habit this chapter teaches, adding one layer at a time and measuring each, applies to every change that follows.

O que o esquema diz, e o que não diz

Última alteração na edição 2026.11.12

Um esquema informa ao modelo quais tabelas existem, como se chamam as colunas e que tipo de dado cada coluna armazena. Já é muita coisa, e ele nunca fica desatualizado, porque é o próprio banco de dados que o produz. O que ele não consegue informar ao modelo é o que as colunas significam: quais linhas uma tabela contém, quais linhas entram em um indicador de negócio, em que fuso horário está um timestamp e qual de duas tabelas parecidas é a em que as pessoas confiam. Esses significados determinam se uma consulta devolve o número certo, e eles estão no SQL que construiu os data marts e na cabeça das pessoas que os mantêm.

A medida deste livro são as perguntas de referência, construídas no capítulo 3: perguntas com resposta verificada, avaliadas na divisão de desenvolvimento enquanto o contexto e os prompts são ajustados, e uma única vez na divisão de teste reservada. Toda camada deste capítulo é acrescentada ao esquema, portanto o esquema é a primeira camada, e o ponto de partida é identificar o que ele deixa em aberto. build_context(), em lib/context.py, o módulo compartilhado que constrói todas as camadas deste capítulo, devolve a camada de esquema quando não se pede mais nada.

schema-digest.py Python 3.12.15 · duckdb 2.0.0 Aprovado
"""What the schema tells a model about revenue, and three March 2026 totals that all fit it. The three printed totalscarry the paragraph's point: they differ, and only the one that leaves out cancelled orders follows the finance definition."""from lib.context import build_contextfrom lib.warehouse import connectcon = connect()schema = build_context(con, "", layers=("schema",))      # no question, so only the schema layer is builtprint(f"tables in the schema: {schema.text.count('CREATE TABLE')}")print(f"schema tokens (estimated): {schema.tokens}")for line in schema.text.splitlines():    if "analytics.fct_orders " in line or "analytics.orders_legacy " in line:        print(line)march = "BETWEEN DATE '2026-03-01' AND DATE '2026-03-31'"   # the window of the question, on the business daycandidates = {    "orders_legacy.revenue": f"SELECT sum(revenue) FROM analytics.orders_legacy WHERE order_date {march}",    "fct_orders.net_revenue_eur, every order":        f"SELECT sum(net_revenue_eur) FROM analytics.fct_orders WHERE order_date_local {march}",    "fct_orders.net_revenue_eur, not cancelled":        f"SELECT sum(net_revenue_eur) FROM analytics.fct_orders WHERE order_date_local {march} "        "AND status <> 'cancelled'",          # net revenue leaves out cancelled orders, as the status note says}for name, sql in candidates.items():    (total,), = con.execute(sql).fetchall()    print(f"{name}: {total:,.2f}")
Saídapython schema-digest.py
tables in the schema: 16
schema tokens (estimated): 912
CREATE TABLE analytics.fct_orders (order_id INTEGER, customer_id INTEGER, region VARCHAR, country VARCHAR, channel VARCHAR, status VARCHAR, placed_at TIMESTAMP, order_date_local DATE, currency VARCHAR, line_count BIGINT, gross_revenue_eur DECIMAL(12,2), net_revenue_eur DECIMAL(12,2), discount_eur DECIMAL(12,2), tax_eur DECIMAL(12,2), _loaded_at TIMESTAMP);
CREATE TABLE analytics.orders_legacy (id INTEGER, cust_id INTEGER, order_date DATE, order_status VARCHAR, sales_region VARCHAR, revenue DECIMAL(12,2));
orders_legacy.revenue: 186,635.93
fct_orders.net_revenue_eur, every order: 162,542.79
fct_orders.net_revenue_eur, not cancelled: 150,082.83

A camada de esquema é o armazém inteiro: todas as 16 tabelas, cerca de 912 tokens. Leia as duas linhas impressas como um modelo as leria. analytics.fct_orders tem uma coluna net_revenue_eur e uma coluna status, e analytics.orders_legacy tem uma coluna revenue e uma coluna order_date.

Cada um dos três totais que o exemplo calcula para março de 2026 é uma leitura razoável dessas duas linhas, e eles divergem entre si. A coluna revenue da tabela legada dá 186,635.93 euros, e a coluna net_revenue_eur do data mart, somando todos os pedidos, dá 162,542.79, que é o que o modelo do capítulo 1 calculou. Somente o terceiro, 150,082.83, segue a definição da equipe financeira, e nada no esquema aponta para ele em vez dos outros dois.

O armazém inteiro da loja cabe nesses 912 tokens, então não há motivo para reduzi-lo. Um estudo com modelos fortes no benchmark BIRD constatou que passar o esquema completo, quando ele cabe, pode superar a seleção prévia de tabelas, porque a etapa de seleção pode descartar uma coluna de que a consulta precisa (Maamari et al., 2024). Um armazém com milhares de tabelas é outro problema, que este capítulo não resolve. O problema da loja não é quais tabelas mostrar. É que o que as tabelas mostram não deixa claro qual número está certo.

Regra prática: mantenha o esquema inteiro no contexto quando ele couber, e acrescente significado a ele em vez de retirar tabelas. Antes de escrever qualquer outra camada, tome uma pergunta real e liste todas as leituras do esquema que a respondem. Cada leitura que o esquema não descarta é uma regra que alguma outra camada precisa declarar.

Esta parte ainda não foi traduzida, por isso está em inglês.

Describe columns where names mislead

Última alteração na edição 2026.11.15

The schema leaves open which of several readings is right, and the cheapest place to close some of them is the column itself. A column note is a sentence about one table or column that says what its name doesn't: which rows the table holds, the unit and sign of an amount, the time zone of a timestamp, the values a column takes, or that a table is deprecated. A note isn't documentation of every column. A column called product_name needs none, and a note that repeats the name costs tokens and says nothing. The notes worth writing are on the columns a careful analyst new to the warehouse would misread.

Published evidence points the same way. A study that generated column descriptions with a language model found that adding them to the prompt improved text-to-SQL accuracy and, in an ablation that replaced column names with uninformative ones, that the descriptions helped most there (Wretblad et al., 2024); it is a single preprint on the BIRD benchmark. Databricks' guidance for its Genie Agents says that "quality table and column descriptions in Unity Catalog are critical for Genie accuracy" (Databricks), which is the vendor's advice, not a measurement.

The shop's notes live in examples/data/shop/columns.toml, one entry per table with an optional note per column, and load_column_notes() in lib/context.py turns each into one line, such as analytics.fct_orders.placed_at: When the order was placed, in UTC. The notes cover the traps chapter 2 found: orders_legacy is deprecated, net_revenue_eur still holds a value for cancelled orders, placed_at is UTC while order_date_local is the business day, discount is negative in raw.order_lines and positive in the marts, and the fiscal year starts on February 1.

The recording context-layers shows what the notes do once the examples replay it. The recording holds the pinned model's SQL for every dev question of kind answer, written once with each layer set this chapter measures, using the prompt LAYERED_SQL in lib/prompts.py. The recording was made by recordings/record_context_layers.py, with the same model, sampling and seed as chapter 1's, and the examples replay it by the hash of the exact messages.

column-descriptions.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Add the column notes on columns whose names mislead, and count the dev questions they fix and break under the recordedmodel. The `fixed` and `broken` lines settle the net effect of the notes, and the broken question shows how a note misleads."""from lib.context import LAYER_SETS, build_context, grade_with_context, load_column_notesfrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.warehouse import connectcon = connect()notes = load_column_notes()print(f"column notes: {len(notes)}")print(f"column notes tokens (estimated): {build_context(con, '', layers=('columns',)).tokens}")for note in notes:    if note.startswith(("analytics.fct_orders.status", "analytics.fct_orders.placed_at")):        print(note)model = RecordedModel("context-layers")before, after = LAYER_SETS[0], LAYER_SETS[1]          # the sets are cumulative: [0] is the schema alone, [1] adds the notesdev = [q for q in load_golden() if q.split == "dev" and q.kind == "answer"]fixed, broken = [], []for q in dev:    was, now = grade_with_context(con, model, q, before), grade_with_context(con, model, q, after)    if was != "correct" and now == "correct":          # fixed: wrong or error without the notes, correct with them        fixed.append(q.question_id)    if was == "correct" and now != "correct":        broken.append(q.question_id)print(f"fixed by column notes: {len(fixed)} ({', '.join(fixed)})")print(f"broken by column notes: {len(broken)} ({', '.join(broken)})")
Saídapython column-descriptions.py
column notes: 33
column notes tokens (estimated): 1087
analytics.fct_orders.status: One of placed, shipped, delivered, cancelled, returned. Revenue leaves out cancelled orders; order and customer counts include them. Returned orders count everywhere.
analytics.fct_orders.placed_at: When the order was placed, in UTC. Don't filter or group dates on it; use order_date_local.
fixed by column notes: 7 (gq-001, gq-002, gq-003, gq-004, gq-008, gq-009, gq-011)
broken by column notes: 1 (gq-018)

The 33 notes add about 1087 tokens, more than the schema itself. On the dev questions they fix 7 (gq-001, gq-002, gq-003, gq-004, gq-008, gq-009, gq-011), among them gq-002, the March question from chapter 1, whose query now filters out cancelled orders. They also break 1 (gq-018). That question asks for the refunds on orders placed in March 2026, and with the notes the model filtered on both the order's date and refund_month, so it dropped refunds paid in a later month. The note on refund_month was accurate. The model used it on a question that didn't call for it, which is what a note can't prevent.

Working rule: Write notes only for tables and columns whose names mislead: deprecated tables, amounts with a sign or unit you wouldn't guess, timestamps in a time zone other than the business's, and columns whose values include a placeholder such as (none). Write them as facts about the column, not as instructions for one question. Expect a note to fix many questions and occasionally mislead on one, and keep the questions it broke in the golden set so you see it happen.

Esta parte ainda não foi traduzida, por isso está em inglês.

Add the business glossary

Última alteração na edição 2026.11.15

Column notes describe the tables. Many questions turn on something else: what a business word means at this company. "Revenue" could be gross or net. "Orders" could include cancelled ones or not. Refunds "in March" could mean refunds paid in March or refunds on March's orders. These are choices, made by the teams that own the numbers, and no table states them. A business glossary lists the business words whose meaning a company's teams chose, and writes each meaning in terms of the warehouse's tables and columns.

Evidence for this layer comes from BIRD, a text-to-SQL benchmark whose questions come with "external knowledge evidence": hints that link a question's words to the database, such as what a term means as a formula over columns. In the benchmark's paper, adding the evidence raised GPT-4's execution accuracy on the dev set from 30.90% to 46.35%, with 2023 models and zero-shot prompts (Li et al., 2023). The hints are written per question, though, which is closer to giving the model the answer's ingredients than to giving it a glossary written once.

The shop's glossary is examples/data/shop/glossary.toml. Its entries define revenue as net revenue unless the question says gross, orders as every order including cancelled ones, average order value, fiscal periods, new customers, refunds by the month they were paid, units sold, and the two meanings of "channel". load_glossary() turns each into one line.

The glossary wasn't written blind, and the book's history shows how it changed. In the same change that recorded the model's first dev replies, with the schema alone, the entry for orders was corrected: it had said that order counts leave out cancelled orders, which disagreed with the finance team's golden SQL. The entries for units sold and for customers who ordered were added in that change too, and the column note on status was reworded to match. Later, after the test questions had been recorded with the schema alone, the refund-rate entry was changed to the book-wide definition that chapter 5 uses and that test question gq-028 also follows: refunded amount divided by gross revenue, because a refund pays back tax and only gross revenue includes it. Each change made the glossary agree with golden SQL that it was then scored against. That is how a dev score gets fitted to dev, and the check on test comes back to what it costs.

glossary-context.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Add the business glossary on top of the column notes, and count what it changes on the dev questions. The gq-002lines show the query the prose describes: with the glossary, the March query no longer filters on status."""from lib.context import LAYER_SETS, build_context, grade_with_context, load_glossaryfrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.prompts import layered_sql_messagesfrom lib.warehouse import connectcon = connect()terms = load_glossary()print(f"glossary terms: {len(terms)}")print(f"glossary tokens (estimated): {build_context(con, '', layers=('glossary',)).tokens}")print(next(t for t in terms if t.startswith("revenue:")))model = RecordedModel("context-layers")before, after = LAYER_SETS[1], LAYER_SETS[2]          # schema and column notes; then the glossary toodev = [q for q in load_golden() if q.split == "dev" and q.kind == "answer"]changes = {(was, now): [] for was in ("correct", "other") for now in ("correct", "other")}   # "other" is wrong or errorfor q in dev:    was, now = (grade_with_context(con, model, q, layers) for layers in (before, after))    changes["correct" if was == "correct" else "other", "correct" if now == "correct" else "other"].append(q.question_id)print(f"fixed by the glossary: {len(changes['other', 'correct'])} ({', '.join(changes['other', 'correct'])})")print(f"broken by the glossary: {len(changes['correct', 'other'])} ({', '.join(changes['correct', 'other'])})")q = next(q for q in dev if q.question_id == "gq-002")reply = model.complete(layered_sql_messages(build_context(con, q.question, after).text, q.question))filters_status = "status" in reply     # a query that filters on status has to name the columnassert not filters_status              # the chapter states it: with the glossary, gq-002 no longer filters on statusprint(f"gq-002 with the glossary filters status: {'yes' if filters_status else 'no'}")
Saídapython glossary-context.py
glossary terms: 13
glossary tokens (estimated): 609
revenue: Net revenue unless the question says gross: sum of analytics.fct_orders.net_revenue_eur over orders whose status isn't 'cancelled'. In EUR.
fixed by the glossary: 3 (gq-012, gq-015, gq-018)
broken by the glossary: 3 (gq-002, gq-009, gq-016)
gq-002 with the glossary filters status: no

The glossary's gains and losses cancel out: it fixes 3 (gq-012, gq-015, gq-018) dev questions and breaks 3 (gq-002, gq-009, gq-016). Its 13 terms add about 609 tokens on top of the schema and the column notes. The broken ones are the instructive part. For gq-002, the same March question, the query written with the glossary no longer filters on status, although the revenue entry and the note on status both say to leave cancelled orders out. For gq-016, the model left cancelled orders out of a customer count, applying the revenue rule where the glossary says it doesn't apply. For gq-009, the average order value, it divided the average order's net revenue by the number of orders again, although the glossary defines the measure as net revenue divided by the number of orders.

Each configuration was recorded once, with one sampled reply per question, so a change of one or two questions out of 17 is within what a second sample could undo. Chapter 10 measures that spread with repeated runs. What this result supports is narrower: the glossary didn't make this model better on these questions, and a rule stated twice was still missed. A plausible reason, which we haven't measured here, is that a longer context makes a small model use each part of it less reliably. Studies of long contexts find that models use information in the middle of a prompt worse than at its edges (Liu et al., 2024), and that accuracy falls with length when the question and the passage don't share words (Modarressi et al., 2025).

Working rule: Write a glossary entry for each business word whose meaning is a choice someone made, and state it in terms of tables and columns. Prefer one place for each rule, a column note or a glossary entry; the shop's files state the cancelled-orders rule in both, and the miss on gq-002 shows that saying it twice doesn't make it stick. Measure the glossary on your dev questions before you trust it, and keep it only if it helps there or the next layer depends on it.

Esta parte ainda não foi traduzida, por isso está em inglês.

Retrieve verified queries as examples

Última alteração na edição 2026.11.15

Column notes and a glossary state rules. A verified query shows the shop's rules applied: a question someone asked, with SQL a person checked against the shop's definitions. Given a few such pairs that resemble the question, a model can copy their shape, such as the join to dim_date for a fiscal period or the filter on status, instead of inventing one. A systematic comparison of prompt designs for text-to-SQL found that the best setup chose examples by the similarity of their questions, after masking domain-specific words, and by the similarity of their SQL to a draft query written first, and showed each as a question and its SQL alone (Gao et al., 2023). Warehouse vendors build the same idea into their assistants. Snowflake's Cortex Analyst keeps a verified query repository of "questions and corresponding SQL queries to answer them" (Snowflake), and Databricks says a Genie Agent can use an example SQL query directly or learn from it, and treats parameterized ones as trusted assets (Databricks). Both are the vendors' own descriptions, not measurements.

The shop's verified queries live in examples/data/shop/verified_queries.toml. Their fields follow Snowflake's field names, name, question, sql, verified_at and verified_by, plus the role the query was checked for. Two differ in content. Snowflake's verified_at is optional and counted in seconds since 1970, and its verified_by names the person who checked the query; the shop's files give a date and name a team, never a person. None of the verified queries is a golden question. They are near neighbours of questions in both splits, such as net revenue by fiscal quarter for fiscal 2025 or average order value by region for 2024, which is how a real repository grows: from questions people asked before. A verified query that was itself a golden question would make the score measure copying.

Retrieval in lib/context.py is lexical, with no embeddings. question_terms() lowercases the question, drops common words and masks three kinds of word: numbers, month names and the shop's three region codes. Every other word stays as it is. QueryIndex scores each verified question against the masked question with TF-IDF and cosine similarity, and search() keeps at most three matches that score at least 0.4. This is a cheap stand-in for the selection in Gao et al., not a reproduction of it: it masks far fewer words, and it drops the second step, which compares draft SQL.

retrieve-similar-queries.py Python 3.12.15 · duckdb 2.0.0 Aprovado
"""Retrieve the verified queries most like a question, by TF-IDF over the masked question's words. The matches kept at the0.4 default show that a threshold can leave a question with no example, and the masked phrases show what masking does."""from lib.context import QueryIndex, load_verified_queries, question_termsqueries = load_verified_queries()index = QueryIndex(queries)print(f"verified queries: {len(queries)}")for label, question in (("aov", "What was the average order value in each region in March 2026?"),                        ("parcels", "Which warehouse ships the most parcels?")):    print(f"question: {question}")    print(f"masked terms: {' '.join(question_terms(question))}")    for score, query in index.search(question, k=3, min_score=0.0):     # no threshold here, so the scores show the gap        print(f"  {score:.3f}  {query.name}")    print(f"{label} matches kept: {len(index.search(question))}")         # the default threshold, 0.4, is the one the book uses# Months and numbers are masked; "calendar" and "year" are not common words, and "Q3" is not masked at all.for phrase in ("March 2026", "January 2025", "calendar year 2024", "EMEA in Q3"):    print(f"masked \"{phrase}\": {' '.join(question_terms(phrase))}")
Saídapython retrieve-similar-queries.py
verified queries: 16
question: What was the average order value in each region in March 2026?
masked terms: average order value region <month> <number>
  0.835  average_order_value_by_region
  0.422  gross_revenue_by_region_month
  0.191  orders_of_one_customer
aov matches kept: 2
question: Which warehouse ships the most parcels?
masked terms: warehouse ships most parcels
  0.375  units_sold_by_product
  0.357  net_revenue_by_category
  0.000  average_order_value_by_region
parcels matches kept: 0
masked "March 2026": <month> <number>
masked "January 2025": <month> <number>
masked "calendar year 2024": calendar year <number>
masked "EMEA in Q3": <region> q3

For the average order value in each region in March 2026, the closest of the 16 verified queries is the same measure by region for 2024, and 2 matches pass the threshold. The second question, about which warehouse ships the most parcels, is one the shop's data can't answer at all. Its best scores come from the single shared word "most", and none passes the threshold, so 0 queries are offered. Without the threshold, the model would get three confident-looking examples, two of them about products, for a question about shipping.

The last lines of the output show what masking does and doesn't do. "March 2026" and "January 2025" both become <month> <number>, so a question about one month matches a verified query about another. "Calendar year 2024" becomes calendar year <number>, because "calendar" and "year" are neither common words nor values, and a quarter written as "Q3" isn't masked at all. That is why the March question's best match, a question about calendar year 2024, scores less than a perfect 1.

Working rule: Retrieve a few verified queries per question, chosen by similarity of the masked question, and show each as its question and SQL. Set a minimum score and check it on questions the warehouse can't answer, because lexical retrieval always returns something. Never let a golden question into the verified set. Move to embeddings only when your own measurement shows lexical retrieval missing queries that a person would call similar.

Ajustar o contexto a um orçamento

Última alteração na edição 2026.11.12

Com o esquema, as notas de colunas, o glossário e as consultas verificadas recuperadas, o contexto da loja já ocupa várias vezes mais tokens que o esquema sozinho, e cada camada que você acrescenta o torna mais longo. Um orçamento de tokens é a quantidade máxima de contexto que o agente pode enviar com uma pergunta. Ele existe por três razões. A janela do modelo é finita: o ambiente de execução usado nas gravações do livro dá ao modelo 8,192 tokens, e até 384 deles são reservados para a resposta. Cada token do prompt custa tempo em uma CPU e dinheiro em um modelo hospedado. E um contexto mais longo não é aproveitado de maneira uniforme, pois os modelos usam menos as informações no meio de um prompt longo do que as do início ou do fim (Liu et al., 2024).

Ajustar um contexto a um orçamento exige decidir o que deixar de fora, e essa decisão precisa de uma ordem de prioridade. build_context() percorre as camadas na ordem de prioridade, acrescenta os itens um a um e descarta qualquer item que levaria o contexto além do orçamento, de modo que um item posterior e mais curto ainda pode caber. O esquema nunca é descartado, pela razão apresentada na primeira seção: um esquema que cabe deve ser enviado por inteiro. Qualquer que seja a prioridade, o contexto é renderizado em uma ordem fixa, com as camadas estáveis primeiro, as consultas recuperadas em seguida e a pergunta por último. Isso mantém a pergunta ao lado dos exemplos mais parecidos com ela, e permite que o executor do modelo reutilize a parte do prompt que toda pergunta compartilha, em vez de lê-la novamente a cada chamada. O executor que fez as gravações do livro, llama-cpp-python 0.3.36, faz isso em Llama.generate: ele compara os tokens do novo prompt com os do anterior e avalia apenas os tokens que vêm depois do maior prefixo que eles têm em comum (llama-cpp-python 0.3.36, llama.py).

As contagens de tokens em lib/context.py são estimativas: estimate_tokens() divide o número de caracteres por 3.5, porque os exemplos não carregam o tokenizador do modelo. Mais abaixo, essa estimativa é comparada com a contagem que o tokenizador do modelo informou quando o prompt foi gravado, e o contexto da pergunta sobre o valor médio do pedido, gq-009, é ajustado a um orçamento ilustrativo de 2,400 tokens sob duas prioridades.

context-token-budget.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Fit one question's layered context to a token budget, and check the token estimate against the model's count. Theestimate runs high, the safe side for a budget, and the dropped lines show what each priority gives up."""from lib.context import LAYERS, build_context, estimate_tokensfrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.prompts import layered_sql_messagesfrom lib.warehouse import connectcon = connect()question = next(q for q in load_golden() if q.question_id == "gq-009").questionfull = build_context(con, question)for layer, tokens in full.layer_tokens().items():    print(f"{layer} tokens (estimated): {tokens}")messages = layered_sql_messages(full.text, question)call = RecordedModel("context-layers").call_for(messages)print(f"prompt tokens (estimated): {estimate_tokens(''.join(m['content'] for m in messages))}")   # every message, joinedprint(f"prompt tokens (model's tokenizer): {call['prompt_tokens']}")budget = 2400          # an illustrative budget, well inside the 8,192-token window the recordings usefor name, priority in (("default", LAYERS), ("examples first", ("schema", "columns", "verified", "glossary"))):   # verified ranks above glossary    fitted = build_context(con, question, layers=priority, budget=budget)    kept = {layer: len(items) for layer, items in fitted.items.items()}    print(f"{name} priority: {fitted.tokens} tokens, kept {kept}")    print(f"{name} dropped: {len(fitted.dropped)}")print(f"render order: {', '.join(layer for layer, items in fitted.items.items() if items)}, then the question")
Saídapython context-token-budget.py
schema tokens (estimated): 912
columns tokens (estimated): 1087
glossary tokens (estimated): 609
verified tokens (estimated): 200
prompt tokens (estimated): 2917
prompt tokens (model's tokenizer): 2364
default priority: 2393 tokens, kept {'schema': 1, 'columns': 33, 'glossary': 9, 'verified': 0}
default dropped: 6
examples first priority: 2384 tokens, kept {'schema': 1, 'columns': 33, 'glossary': 5, 'verified': 2}
examples first dropped: 8
render order: schema, columns, glossary, verified, then the question

A estimativa para o prompt inteiro é de 2917 tokens, contra 2364 contados pelo tokenizador do modelo. Uma estimativa que fica acima do real é a direção segura para um orçamento: o contexto admitido sempre cabe. Com a prioridade padrão, que segue a ordem em que as camadas foram apresentadas, o orçamento mantém todas as notas de colunas e parte do glossário, e descarta as duas consultas recuperadas, 6 itens no total. Com as consultas verificadas acima do glossário, o mesmo orçamento mantém as duas consultas e descarta 8 entradas do glossário. Nenhuma das ordens é a correta em geral. A próxima seção mede quais camadas valem os seus tokens nas perguntas da loja, e essa medição é que deve definir a ordem.

Regra prática: Defina um orçamento de tokens bem dentro da janela do modelo e ajuste o contexto de cada pergunta a ele por uma prioridade que você tenha medido, não pela ordem em que escreveu as camadas. Nunca descarte o esquema quando ele couber. Renderize as camadas estáveis primeiro e a pergunta por último, seja qual for o item descartado. Confira sua estimativa de tokens contra a contagem real do modelo uma vez, e mantenha-a do lado alto.

Esta parte ainda não foi traduzida, por isso está em inglês.

Measure what each layer adds

Última alteração na edição 2026.11.15

Each layer so far was judged by the questions it fixed and broke. To choose which layers to keep, and in what order a budget should drop them, you need all of them measured the same way. An ablation measures every layer the same way: it scores a series of configurations that differ by one layer each, on the same questions, under the same comparison rule. The questions are the dev split of the golden questions, because choosing among configurations is tuning, and tuning on the test split would leave nothing held out to check the choice.

The shop's dev split has 24 golden questions. This chapter scores the 17 of kind answer. The other 7 should be declined, because the warehouse can't answer them or the role asking may not see the answer, and a prompt that only writes SQL can't decline, so they would score as wrong under every layer set; chapters 8 and 9 score them. Each layer set adds one layer to the one before: the schema alone, then column notes, then the glossary, then up to three retrieved verified queries. The outcomes come from grade() in lib/golden.py, as in every chapter: correct when the result matches the golden result, wrong when it runs and doesn't, error when it doesn't run. The first layer set, the schema alone, isn't chapter 3's baseline either. That baseline used the schema-only prompt and scored every question in each split, while this run uses the layered prompt LAYERED_SQL with no other layer filled in and scores the answer questions only, so its numbers differ.

layer-ablation-dev.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Score each layer set on the dev split's answer questions, replaying the recorded model's SQL for each, and print thetotals, the average token counts and the outcome grid. The correct totals and the fixed and broken lines are the numbers the prose rests on."""from collections import Counterfrom lib.context import LAYER_SETS, build_context, grade_with_context, layer_set_namefrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.warehouse import connectcon = connect()model = RecordedModel("context-layers")dev_split = [q for q in load_golden() if q.split == "dev"]dev = [q for q in dev_split if q.kind == "answer"]          # the other questions are declined, and are scored in chapters 8 and 9print(f"dev split questions: {len(dev_split)}")print(f"dev questions: {len(dev)}")print(f"dev questions to decline: {len(dev_split) - len(dev)}")grid = {}for layers in LAYER_SETS:    name = layer_set_name(layers)    grid[name] = [grade_with_context(con, model, q, layers) for q in dev]    counts = Counter(grid[name])    tokens = sum(build_context(con, q.question, layers).tokens for q in dev) // len(dev)   # mean over the dev questions, whole tokens    print(f"{name}: correct {counts['correct']}, wrong {counts['wrong']}, error {counts['error']}")    print(f"{name} correct: {counts['correct']}")    print(f"{name} error: {counts['error']}")    print(f"{name} tokens: {tokens}")print("question  " + "  ".join(f"{i + 1}" for i in range(len(LAYER_SETS))))for i, q in enumerate(dev):    print(f"{q.question_id}    " + "  ".join(grid[name][i][0] for name in grid))# What the verified queries changed, against the run with the glossary.before, after = grid[layer_set_name(LAYER_SETS[2])], grid[layer_set_name(LAYER_SETS[3])]   # the glossary run, then the glossary plus verified queriesfixed = [q.question_id for q, was, now in zip(dev, before, after) if was != "correct" and now == "correct"]broken = [q.question_id for q, was, now in zip(dev, before, after) if was == "correct" and now != "correct"]print(f"fixed by verified queries: {len(fixed)}")print(f"fixed by verified queries, IDs: {', '.join(fixed)}")print(f"broken by verified queries: {len(broken)}")
Saídapython layer-ablation-dev.py
dev split questions: 24
dev questions: 17
dev questions to decline: 7
schema: correct 3, wrong 9, error 5
schema correct: 3
schema error: 5
schema tokens: 912
schema+columns: correct 9, wrong 7, error 1
schema+columns correct: 9
schema+columns error: 1
schema+columns tokens: 1999
schema+columns+glossary: correct 9, wrong 8, error 0
schema+columns+glossary correct: 9
schema+columns+glossary error: 0
schema+columns+glossary tokens: 2608
schema+columns+glossary+verified: correct 14, wrong 2, error 1
schema+columns+glossary+verified correct: 14
schema+columns+glossary+verified error: 1
schema+columns+glossary+verified tokens: 2775
question  1  2  3  4
gq-001    e  c  c  c
gq-002    w  c  w  c
gq-003    e  c  c  c
gq-004    w  c  c  c
gq-006    e  e  w  w
gq-007    w  w  w  w
gq-008    w  c  c  c
gq-009    w  c  w  c
gq-010    e  w  w  c
gq-011    w  c  c  c
gq-012    w  w  c  c
gq-013    e  w  w  c
gq-014    c  c  c  c
gq-015    w  w  c  c
gq-016    c  c  w  c
gq-017    w  w  w  e
gq-018    c  w  c  c
fixed by verified queries: 5
fixed by verified queries, IDs: gq-002, gq-009, gq-010, gq-013, gq-016
broken by verified queries: 0

Column notes and verified queries raise the score; the glossary doesn't. The schema alone gets 3 of the dev questions right, column notes take it to 9, and the glossary leaves it at 9. Verified queries on top give 14, the best set on dev.

The grid shows where the change came from. Against the run with the glossary, retrieved examples fixed 5 questions and broke 0. Two of the fixes are new: the fiscal-quarter question gq-010 and the order count by region gq-013, whose nearest verified queries show exactly the join and the count they needed. The other three, gq-002, gq-009 and gq-016, are the questions the glossary run had lost, so there the examples won back what the glossary had cost.

The totals say how much each layer helped. The failures say what the next layer should be, so the next two examples read them, starting with the schema-only queries that don't run.

missing-columns-schema-only.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Why the dev queries written with the schema alone don't run: the column each one names, and the analytics tables that have it.The per-table counts are the evidence the prose rests on: every failure is a column on the wrong table."""import refrom collections import Counterimport duckdbfrom lib.context import LAYER_SETS, build_contextfrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.prompts import layered_sql_messagesfrom lib.tools import extract_sql, run_sqlfrom lib.warehouse import connectcon, model = connect(), RecordedModel("context-layers")dev = [q for q in load_golden() if q.split == "dev" and q.kind == "answer"]where = Counter()for q in dev:    context = build_context(con, q.question, LAYER_SETS[0])          # the schema alone    sql = extract_sql(model.complete(layered_sql_messages(context.text, q.question)))    try:        run_sql(con, sql)        continue    except duckdb.Error as e:        column = re.search(r'column (?:named )?"(\w+)"', str(e)).group(1)   # DuckDB says either form; take the name    (tables,), = con.execute(        "SELECT string_agg(table_schema || '.' || table_name, ', ' ORDER BY 1) FROM information_schema.columns "   # sorted, so a column in two tables lists them in a fixed order        "WHERE table_schema = 'analytics' AND column_name = ?", [column]).fetchall()   # the marts only: where the model should have looked    where[tables] += 1    print(f"{q.question_id}: no column {column}; it is in {tables}")print(f"schema-only queries that don't run: {where.total()}")for tables, n in where.items():    print(f"missing column only {tables} has: {n}")
Saídapython missing-columns-schema-only.py
gq-001: no column fiscal_year; it is in analytics.dim_date
gq-003: no column calendar_year; it is in analytics.dim_date
gq-006: no column quantity; it is in analytics.fct_order_lines
gq-010: no column fiscal_quarter; it is in analytics.dim_date
gq-013: no column order_date; it is in analytics.orders_legacy
schema-only queries that don't run: 5
missing column only analytics.dim_date has: 3
missing column only analytics.fct_order_lines has: 1
missing column only analytics.orders_legacy has: 1

Every failure is a column the model put on the wrong table. Of the 5, 3 ask the orders mart for a fiscal or calendar column that only dim_date has, 1 asks it for quantity, which only the order lines have, and 1 uses the legacy table's order_date.

These are the mistakes column notes are for. With the notes, 1 query in the whole dev split fails to run.

A second script prints the query for each dev question that all four layers still don't get right.

what-stays-wrong.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""The dev questions that stay wrong with all four layers, each with its query, and a check of whether gq-007 fails only onthe spelling of 'tents'. The count line settles how many dev questions stay wrong, and the gq-007 line settles whether the spelling alone causes its failure."""import duckdbfrom lib.context import LAYERS, build_context, grade_with_contextfrom lib.golden import golden_result, grade, load_goldenfrom lib.model import RecordedModelfrom lib.prompts import layered_sql_messagesfrom lib.tools import extract_sql, run_sqlfrom lib.warehouse import connectcon, model = connect(), RecordedModel("context-layers")dev = [q for q in load_golden() if q.split == "dev" and q.kind == "answer"]def sql_with_all_layers(q):    return extract_sql(model.complete(layered_sql_messages(build_context(con, q.question, LAYERS).text, q.question)))unsolved = [(q, outcome) for q in dev if (outcome := grade_with_context(con, model, q, LAYERS)) != "correct"]   # LAYERS is all fourprint(f"not correct with all four layers: {len(unsolved)}")for q, outcome in unsolved:    sql = sql_with_all_layers(q)    print(f"{q.question_id} ({outcome}): {q.question}")    print("\n".join("    " + line for line in sql.strip().splitlines()))    if outcome == "error":        try:            run_sql(con, sql)        except duckdb.Error as e:            print(f"    -- {str(e).splitlines()[0]}")gq007 = next(q for q in dev if q.question_id == "gq-007")respelled = sql_with_all_layers(gq007).replace("'tents'", "'Tents'")   # change the spelling only, to see whether it is the whole failureprint(f"gq-007 with 'Tents' spelled as stored: {grade(gq007, run_sql(con, respelled), golden_result(con, gq007))}")
Saídapython what-stays-wrong.py
not correct with all four layers: 3
gq-006 (wrong): What were net revenue and the number of units sold in each region in 2025?
    SELECT
        o.region,
        SUM(o.net_revenue_eur) AS net_revenue,
        SUM(l.quantity) AS units_sold
    FROM analytics.fct_orders AS o
    JOIN analytics.fct_order_lines AS l USING (order_id)
    WHERE o.order_date_local >= '2025-01-01' AND o.order_date_local < '2026-01-01'
        AND o.status <> 'cancelled'
    GROUP BY o.region
    ORDER BY net_revenue DESC
gq-007 (wrong): How much net revenue came from orders that included a tent in 2025?
    SELECT
        SUM(fol.net_amount_eur) AS net_revenue
    FROM
        analytics.fct_order_lines fol
    JOIN
        analytics.fct_orders fo ON fol.order_id = fo.order_id
    JOIN
        analytics.dim_products dp ON fol.product_id = dp.product_id
    JOIN
        analytics.dim_date dd ON fo.order_date_local = dd.date_day
    WHERE
        dp.category = 'tents'
        AND dd.calendar_year = 2025
        AND fo.status != 'cancelled'
gq-017 (error): What was net revenue in each of our countries in 2025?
    SELECT d.country, round(sum(o.net_revenue_eur), 2) AS net_revenue
    FROM analytics.fct_orders AS o
    JOIN analytics.dim_date AS d ON d.date_day = o.order_date_local
    WHERE o.status <> 'cancelled'
      AND d.calendar_year = 2025
    GROUP BY d.country
    ORDER BY net_revenue DESC
    -- Binder Error: Table "d" does not have a column named "country"
gq-007 with 'Tents' spelled as stored: wrong

With all four layers, 3 questions are still not right. gq-006 asks for net revenue and units sold by region. The model joined the orders mart to its lines and summed net_revenue_eur over the joined rows, so an order's revenue counts once for every line it has. The glossary's entry for units sold says to sum the lines per order first for exactly this reason, so this is a rule the context stated and the model missed. gq-007 filters on the category 'tents' where the data says 'Tents', a value no layer lists. Spelled as stored, the answer is still wrong, because the query sums only the tent lines, where the question asks for the whole revenue of the orders that included one. gq-017 asks for "our countries", which depends on the region of the regional lead asking. No layer says who is asking, and the model's query, which takes country from dim_date, doesn't run.

This is the book's measurement on one synthetic warehouse and one 4-billion-parameter model, with one sampled reply per question, not a general result about layered context. We found no peer-reviewed study from 2025 or 2026 that measures retrieved verified queries separately from other context on business data, which is why the chapter measures it here.

Working rule: Measure context one layer at a time on the dev split, with the same questions, comparison rule and model, and keep the outcomes per question, not only the totals. Treat a difference of one or two questions on a set this size as noise until repeated runs say otherwise. Read the queries that fail, because each one names a layer you don't have yet or a rule the model still skips.

Verifique as camadas escolhidas em teste

Última alteração na edição 2026.11.12

A ablação escolheu as quatro camadas na divisão dev, as perguntas em que o capítulo fez seus ajustes. O conjunto que obteve a melhor pontuação ali ainda precisa ser verificado uma vez na divisão de teste reservada, junto com o esquema sozinho, para que a comparação não seja favorecida por ter sido escolhida com as mesmas perguntas. O registro context-layers-test guarda o SQL do modelo para as 17 perguntas de teste do tipo answer, apenas sob esses dois conjuntos de camadas. Cada pergunta de teste também é pontuada em relação à consulta verificada de melhor correspondência, para os parágrafos que seguem.

best-layers-on-test.py Python 3.12.15 · duckdb 2.0.0 · pyyaml 6.0.3 · sqlglot 30.31.0 Aprovado
"""Choose the best layer set on the dev split, then score it once on the held-out test split beside the schema alone. Thecorrect counts are the numbers the prose rests on, and the retrieval scores measure how close each split sits to the verified queries."""from collections import Counterfrom lib.context import LAYER_SETS, QueryIndex, grade_with_context, layer_set_name, load_verified_queriesfrom lib.golden import load_goldenfrom lib.model import RecordedModelfrom lib.warehouse import connectcon = connect()questions = [q for q in load_golden() if q.kind == "answer"]   # the answer questions only, as in the ablationdev = [q for q in questions if q.split == "dev"]test = [q for q in questions if q.split == "test"]dev_model = RecordedModel("context-layers")          # dev and test replies come from separate recordings; the test one plays no part in the choicedev_correct = {layers: sum(grade_with_context(con, dev_model, q, layers) == "correct" for q in dev)               for layers in LAYER_SETS}best = max(LAYER_SETS, key=lambda layers: (dev_correct[layers], -len(layers)))   # ties go to fewer layersprint(f"best on dev: {layer_set_name(best)}")print(f"best on dev correct: {dev_correct[best]} of {len(dev)}")test_model = RecordedModel("context-layers-test")print(f"test questions: {len(test)}")print(f"test question IDs: {', '.join(q.question_id for q in test)}")for name, layers in (("schema", LAYER_SETS[0]), ("best", best)):    counts = Counter(grade_with_context(con, test_model, q, layers) for q in test)    print(f"test {name}: correct {counts['correct']}, wrong {counts['wrong']}, error {counts['error']}")    print(f"test {name} correct: {counts['correct']}")# How close each split's questions sit to the verified queries, by the retriever's own score.index = QueryIndex(load_verified_queries())for split, qs in (("dev", dev), ("test", test)):    best_match = [(index.search(q.question, k=1, min_score=0.0)[0], q) for q in qs]   # the best match, whatever its score    print(f"{split} mean top retrieval score: {sum(score for (score, _), _ in best_match) / len(qs):.2f}")    print(f"{split} questions with no verified query kept: {sum(not index.search(q.question) for q in qs)}")    (score, query), q = max(best_match, key=lambda pair: pair[0][0])    print(f"{split} closest pair: {q.question_id} and {query.name}")    print(f"{split} closest score: {score:.3f}")
Saídapython best-layers-on-test.py
best on dev: schema+columns+glossary+verified
best on dev correct: 14 of 17
test questions: 17
test question IDs: gq-025, gq-026, gq-027, gq-028, gq-029, gq-030, gq-031, gq-032, gq-033, gq-034, gq-035, gq-036, gq-037, gq-038, gq-039, gq-040, gq-041
test schema: correct 4, wrong 9, error 4
test schema correct: 4
test best: correct 8, wrong 7, error 2
test best correct: 8
dev mean top retrieval score: 0.57
dev questions with no verified query kept: 2
dev closest pair: gq-001 and net_revenue_by_fiscal_quarter
dev closest score: 0.862
test mean top retrieval score: 0.63
test questions with no verified query kept: 1
test closest pair: gq-029 and new_customers_by_region
test closest score: 0.933

As camadas ainda ajudam em perguntas com que ninguém as ajustou, que é o resultado que importa. O melhor conjunto em dev, schema+columns+glossary+verified, acertou 14 of 17 em dev. Em teste, o esquema sozinho acerta 4 e as quatro camadas acertam 8.

A diferença entre as pontuações de dev e de teste é o outro resultado, e este capítulo não mediu sua causa. Uma explicação é que as consultas verificadas se adequam melhor a dev, mas as próprias pontuações do recuperador a contradizem: a pontuação média da melhor correspondência de cada pergunta é 0.57 em dev e 0.63 em teste.

O par mais próximo, uma pergunta de referência e sua correspondência verificada, também está em teste: gq-029 and new_customers_by_region, com 0.933, a mesma pergunta feita sobre o ano civil de 2024. O par mais próximo em dev pontua 0.862.

Duas coisas são conhecidas. Primeiro, dev foi usada para mais do que escolher o conjunto. Como disse a seção do glossário, o glossário foi editado para concordar com o SQL de referência enquanto as primeiras respostas de dev eram registradas, e uma edição desse tipo ajusta o contexto às perguntas em que ele será pontuado, quer alguém pretenda isso, quer não. É assim que dev é superajustada, e por isso uma pontuação em dev não é o resultado a relatar. Segundo, as divisões fazem perguntas diferentes: a divisão de teste pergunta sobre margem bruta, sessões e gastos com marketing, temas que nenhuma pergunta de dev aborda. Distinguir essas causas exigiria mais perguntas e execuções repetidas do que este capítulo tem.

Regra prática: Escolha o conjunto em dev, depois pontue-o uma vez na divisão de teste ao lado da linha de base, e relate o número de teste. Anote cada mudança feita no contexto depois de ler um resultado de dev, porque cada uma é uma forma de a pontuação de dev superestimar o resultado. Quando dev e teste discordam, verifique o que você consegue medir, como a proximidade de cada divisão em relação aos seus exemplos, antes de explicar a diferença.

O que dá errado em produção

Última alteração na edição 2026.11.12

As camadas elevaram a pontuação de desenvolvimento bem acima do que o esquema sozinho permite, e elas continuam corretas apenas enquanto o warehouse e as suas definições permanecerem como estão. Quando as pessoas passam a depender delas, as falhas são silenciosas: uma nota ou uma consulta verificada continua ensinando uma regra que o warehouse já mudou.

As notas se afastam dos marts. Uma nota de coluna é verdadeira sobre o mart como ele era quando alguém a escreveu. Quando os engenheiros de analytics alteram fct_orders, por exemplo para remover pedidos cancelados do mart, a nota que manda filtrar por status fica errada, e nada falha. Você percebe isso mantendo as notas no mesmo repositório do SQL que constrói os marts, revisando-as nos mesmos pull requests e verificando no CI que cada tabela e cada coluna citada em uma nota ainda existe.

Um valor que o contexto nunca lista. O modelo escreveu 'tents' para uma categoria armazenada como 'Tents', e a consulta rodou e não retornou nada. As perguntas citam valores, como categorias, canais, regiões e nomes de campanhas, e o modelo adivinha a grafia deles. Ler as respostas erradas cujo resultado é vazio ou zero ajuda a detectar isso. Para colunas com algumas dezenas de valores distintos, liste-os na nota da coluna. Para colunas com muitos, recupere os valores que correspondem às palavras da pergunta, da mesma forma que você recupera consultas verificadas.

Uma regra declarada no contexto ainda é ignorada. Com o glossário, o modelo deixou de aplicar o filtro de pedidos cancelados em gq-002 e deixou os pedidos cancelados de fora da contagem de clientes de gq-016. Com as notas de coluna, e de novo com o glossário, ele os deixou de fora da contagem de pedidos de gq-013. Ambos os casos contrariam o que a nota sobre status e o glossário dizem. Com as quatro camadas, ele somou a receita de pedidos sobre as linhas de pedido em gq-006, o que a entrada do glossário sobre unidades vendidas adverte contra. O contexto torna uma regra disponível. Ele não faz o modelo segui-la. Só as pontuações golden por pergunta revelam isso. Quando uma regra precisa valer sempre, tire-a do prompt e coloque-a em código que o modelo não consiga contornar, que é o que a camada semântica do capítulo 5 faz.

A recuperação oferece o exemplo errado. A recuperação lexical compara palavras, então uma pergunta que compartilha apenas "most" com uma consulta verificada recebe essa consulta como exemplo, a menos que um limiar a impeça, e uma pergunta que significa o mesmo com outras palavras não recebe nada. Registre qual consulta verificada cada resposta usou, como o Cortex Analyst da Snowflake informa no campo confidence de sua API (Snowflake), e depois leia os registros das respostas que saíram erradas.

Uma consulta verificada é confundida com uma resposta verificada. Uma consulta recuperada orienta o modelo, mas o SQL que ele escreve é um SQL novo, que ninguém conferiu. A Databricks marca uma resposta do Genie como verificada somente quando o texto exato de uma consulta de exemplo parametrizada é usado (Databricks), e essa é a distinção a manter. Nunca apresente uma resposta como verificada porque uma consulta parecida foi verificada. Você percebe isso verificando o que a sua interface diz sobre respostas construídas a partir de exemplos.

As consultas verificadas ficam desatualizadas. Uma consulta conferida em outubro usa as definições de outubro. Quando as finanças mudam uma definição, as consultas antigas ensinam a regra antiga a toda pergunta parecida. Mantenha uma data verified_at em cada consulta, execute novamente cada uma contra a sua contraparte golden quando uma definição mudar e remova as consultas que ninguém conferiu desde a mudança.

O contexto cresce até custar mais do que oferece. Cada camada é barata de acrescentar, e ninguém remove uma que deixou de ajudar. Em uma CPU, cada token do prompt custa tempo, e com um modelo hospedado ele custa dinheiro a cada chamada. Registrar os tokens de cada chamada junto com o seu resultado, como fazem as gravações aqui, e repetir a ablação quando uma camada cresce, mostra qual camada deixou de compensar os seus tokens.

Regra prática: Versione o contexto junto com o warehouse: mantenha-o ao lado dos marts, verifique-o no CI contra as tabelas que ele cita, avalie-o em perguntas golden a cada mudança e registre-o por resposta, com as consultas verificadas que ela usou. Tire do contexto qualquer regra que precise valer sempre e coloque-a em uma definição que o modelo não consiga contornar.

Decisões tomadas neste capítulo

  • O contexto chega em quatro camadas: o esquema completo, notas sobre colunas cujos nomes induzem a erro, um glossário de negócio e consultas verificadas recuperadas para cada pergunta. Cada uma vive em um arquivo em examples/data/shop/, e build_context() em lib/context.py as monta.
  • O esquema nunca é cortado enquanto couber; o orçamento se aplica às outras camadas.
  • As consultas verificadas são recuperadas lexicalmente, por TF-IDF sobre as palavras da pergunta, com números, meses e códigos de região mascarados, no máximo três, cada uma com pontuação de pelo menos 0.4. Nenhuma delas é uma pergunta de referência, e cada uma indica a equipe que a verificou, nunca uma pessoa.
  • O contexto é renderizado com as camadas estáveis primeiro e a pergunta por último, de modo que a pergunta fique ao lado de seus exemplos e o executor possa reutilizar o prefixo compartilhado.
  • Os tokens são estimados a partir de caracteres, à razão de 3.5 caracteres por token. Essa estimativa superestimou a contagem do modelo no prompt verificado, portanto um contexto que cabe na estimativa deve caber no modelo; verifique isso com os seus próprios prompts.
  • As camadas são medidas por ablação nas perguntas de resposta do split de desenvolvimento, uma camada por vez, e o melhor conjunto é verificado uma única vez no split de teste, ao lado do esquema isolado.
  • O livro mantém as quatro camadas nos capítulos que escrevem SQL, porque esse conjunto obteve a melhor pontuação no desenvolvimento. O glossário permanece, embora não tenha melhorado a pontuação por si só: a camada de consultas verificadas foi medida sobre ele, e não medimos consultas verificadas sem o glossário.

Este capítulo, nesta edição

Edição 2026.11.15Exemplos aprovados em 19 de nov de 20269 de 9 exemplos exemplos deste capítulo passam

Traduzido do livro em inglês, edição 2026.11.15.

Citar esta seção

Formato

Cada formato cita o livro em inglês, que esta tradução acompanha.

Todos os formatos citam esta edição. Seu link abre a edição atual, em que o título de cada seção indica quando seu texto foi alterado pela última vez. As datas de acesso e de recuperação são o dia em que você abre esta página.

Nesta página

Capítulo 4 de 11: Give the agent layered context

  1. O que o esquema diz, e o que não diz
  2. Describe columns where names mislead
  3. Add the business glossary
  4. Retrieve verified queries as examples
  5. Ajustar o contexto a um orçamento
  6. Measure what each layer adds
  7. Verifique as camadas escolhidas em teste
  8. O que dá errado em produção
  9. Decisões tomadas neste capítulo

Voltar ao topoTodos os capítulos