IB-WP-0016-T04: trading-literature profile

Ship a specialized profile for trading memoirs and market-structure
texts. The profile names eight entity categories (trader, market,
strategy, error, psychological_pattern, institution, instrument,
evidence_bearing_claim), five relation types (cause_effect,
lesson_evidence, risk_mitigation, actor_venue, strategy_outcome), and
four evaluation criteria (groundedness, lesson_clarity,
historical_context, overgeneralization_risk). Each is reflected in the
prompts and contracts so the LLM is steered toward operator-level
findings rather than biographical detail or moralising.

The generic profile remains the default. A 2-chapter Lefevre smoke run
with --profile trading-literature completes end-to-end with viable
metrics; 93 tests pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/infospace_bench/profiles/trading-literature/contracts/entity.contract.md

@@ -0,0 +1,15 @@
+# Entity Contract — Trading Literature
+
+Each generated entity must be a Markdown artifact with:
+
+- one top-level heading containing the entity title
+- a `## Category` line containing exactly one of: `trader`, `market`,
+  `strategy`, `error`, `psychological_pattern`, `institution`,
+  `instrument`, `evidence_bearing_claim`
+- a `## Definition` section
+- optional `## Context`, `## Source Evidence`, and `## Review Notes`
+  sections
+
+Entity titles should be stable, short, and reusable across chapters of
+the same source. Do not include the chapter number in the title; that
+provenance belongs in the source artifact, not the entity.

src/infospace_bench/profiles/trading-literature/contracts/evaluation.contract.md

@@ -0,0 +1,19 @@
+# Evaluation Contract — Trading Literature
+
+Each evaluation must be Markdown with YAML frontmatter containing:
+
+- `artifact_id`
+- `evaluator`
+- `evaluated_at`
+- `scores`
+
+Scores must include all four criteria on a 0 to 5 scale, with 5 best:
+
+- `groundedness`
+- `lesson_clarity`
+- `historical_context`
+- `overgeneralization_risk` (higher = lower risk; an entity that
+  silently universalises a chapter-local claim scores low)
+
+Optional `## Review Notes` should quote any specific lines from the
+entity body that drove a low score on any criterion.

src/infospace_bench/profiles/trading-literature/contracts/relation.contract.md

@@ -0,0 +1,16 @@
+# Relation Contract — Trading Literature
+
+Each generated relation must be a Markdown artifact with:
+
+- one top-level heading containing the relation title
+- `## Subject`
+- `## Predicate`
+- `## Object`
+- `## Relation Type` — exactly one of: `cause_effect`, `lesson_evidence`,
+  `risk_mitigation`, `actor_venue`, `strategy_outcome`
+- optional `## Evidence` and `## Feedback Role`
+
+Subject and object values should match generated entity titles whenever
+possible. A relation whose subject or object does not correspond to any
+extracted entity must include an `## Evidence` section that quotes the
+phrase from the source supporting the link.

src/infospace_bench/profiles/trading-literature/contracts/summary.contract.md

@@ -0,0 +1,11 @@
+# Summary Contract — Trading Literature
+
+Each source summary should preserve:
+
+- the narrator's actions and the market events they reacted to
+- named strategies, instruments, venues, and institutions present in
+  the chunk
+- explicit lessons or rules of thumb the chunk states
+- evidence phrases (dollar figures, dates, counter-party names, tape
+  behaviour) useful for later extraction
+- unresolved ambiguities or anachronisms a reviewer should flag

src/infospace_bench/profiles/trading-literature/profile.yaml

@@ -0,0 +1,35 @@
+id: trading-literature
+name: Trading Literature
+description: |
+  Infospace generation profile for trading memoirs, market-structure texts,
+  and operator narratives. Tunes entity, relation, and evaluation prompts
+  for traders, markets, strategies, errors, psychological patterns,
+  institutions, instruments, and the lessons drawn from them.
+terminology:
+  source_chunk: Chapter or chapter-part of a trading memoir or market-structure text
+  entity: Trader, market, strategy, error pattern, psychological habit, institution, instrument, or evidence-bearing claim
+  relation: Typed link between two trading-literature entities (cause/effect, lesson/evidence, risk/mitigation, actor/venue, strategy/outcome)
+entity_categories:
+  - traders
+  - markets
+  - strategies
+  - errors
+  - psychological_patterns
+  - institutions
+  - instruments
+  - evidence_bearing_claims
+relation_categories:
+  - cause_effect
+  - lesson_evidence
+  - risk_mitigation
+  - actor_venue
+  - strategy_outcome
+granularity:
+  default: |
+    Prefer durable trading concepts and operator-level lessons over biographical
+    detail or stock-price trivia. Each entity should be reusable across chapters.
+evaluation_criteria:
+  - groundedness
+  - lesson_clarity
+  - historical_context
+  - overgeneralization_risk

src/infospace_bench/profiles/trading-literature/templates/evaluate-entity.md

@@ -0,0 +1,34 @@
+# Evaluate Trading-Literature Entity
+
+Profile: {{ macros.profile }}
+
+Evaluate the generated entity as Markdown with YAML frontmatter. Include
+`artifact_id`, `evaluator`, `evaluated_at`, and a `scores` list. Score
+each criterion on a 0 to 5 scale where 5 is best.
+
+Required score names:
+
+- `groundedness` — does the entity stay anchored to the source chunk,
+  with no invented dates, dollar figures, or quotes?
+- `lesson_clarity` — for `strategy`, `error`, `psychological_pattern`,
+  and `evidence_bearing_claim` entities, is the operator-level lesson
+  stated crisply enough to be reused in later chapters? For purely
+  factual entities (trader, market, institution, instrument), score
+  this on the clarity of the definition.
+- `historical_context` — is the entity placed correctly in the era and
+  venue of the source (e.g. early-1900s American equities) without
+  importing modern terminology or instruments?
+- `overgeneralization_risk` — is the entity scoped narrowly enough to
+  resist becoming a vague universal claim? Higher score means lower
+  risk. Flag entities that quietly claim to apply to all markets or
+  all operators when the source restricts the claim.
+
+Add a short `## Review Notes` section listing any specific lines from
+the entity body that drove a low score on any criterion.
+
+Entity artifact: {{ input.artifact_id }}
+Entity title: {{ input.title }}
+
+## Entity
+
+{{ input.content }}

src/infospace_bench/profiles/trading-literature/templates/extract-entities.md

@@ -0,0 +1,37 @@
+# Extract Trading-Literature Entities
+
+Profile: {{ macros.profile }}
+
+Extract reusable infospace entities from the source chunk. Return one
+Markdown bundle where each entity starts with `# Entity Title` and has a
+`## Definition` section, plus a `## Category` line drawn from the list
+below. Add `## Context` and `## Source Evidence` when the chunk gives
+enough material; leave them out rather than inventing detail.
+
+Allowed categories (use exactly one per entity):
+
+- `trader` — a named operator, broker, manipulator, or counter-party
+- `market` — a market, exchange, pit, or named instrument family
+  (e.g. the New York Stock Exchange, the cotton market, the bucket-shop
+  circuit)
+- `strategy` — a named tactic, system, or recurring playbook
+  (e.g. pyramiding, scale buying, tape reading)
+- `error` — a recurring mistake, anti-pattern, or losing habit
+- `psychological_pattern` — a named cognitive or emotional habit that
+  drives decisions (e.g. tip-following, hope-against-evidence)
+- `institution` — a firm, regulator, news organisation, or social venue
+- `instrument` — a specific security, commodity, or contract
+- `evidence_bearing_claim` — a concrete operator-level claim the text
+  asserts and partially supports (e.g. "amateurs buy on tips, pros buy
+  on tape"); preserve the supporting evidence in the body
+
+Prefer entities that will recur across chapters. Avoid fictionalised
+people whose role is purely narrative colour. Avoid wrapping a single
+trade as an entity unless the trade is itself a teachable case.
+
+Source title: {{ input.title }}
+Source artifact: {{ input.artifact_id }}
+
+## Source
+
+{{ input.content }}

src/infospace_bench/profiles/trading-literature/templates/extract-relations.md

@@ -0,0 +1,32 @@
+# Extract Trading-Literature Relations
+
+Profile: {{ macros.profile }}
+
+Extract a small set of important relations from the source chunk. Return
+one Markdown relation artifact per relation. Each artifact uses sections
+`## Subject`, `## Predicate`, `## Object`, and `## Relation Type`. Add
+`## Evidence` whenever the chunk supplies a concrete supporting phrase.
+
+Use exactly one of these relation types per relation:
+
+- `cause_effect` — one entity drives a measurable market or operator
+  outcome (e.g. a strategy causing a loss; a market event causing a
+  policy change)
+- `lesson_evidence` — an `evidence_bearing_claim` is supported (or
+  undercut) by a concrete trade, event, or quote in the source
+- `risk_mitigation` — a strategy, rule, or habit reduces a named risk
+- `actor_venue` — a trader operates in a market, institution, or pit
+- `strategy_outcome` — a named strategy is applied to a specific trade
+  or campaign and produces a labelled outcome (win, loss, scratch)
+
+Subject and object values should match entity titles you would (or did)
+extract in the entities stage. Skip relations whose subject or object
+would be a one-off fictional flourish. Skip implicit moralising; prefer
+relations the chunk actually evidences.
+
+Source title: {{ input.title }}
+Source artifact: {{ input.artifact_id }}
+
+## Source
+
+{{ input.content }}

src/infospace_bench/profiles/trading-literature/templates/summarize-source.md

@@ -0,0 +1,23 @@
+# Summarize Trading-Literature Source
+
+Profile: {{ macros.profile }}
+
+Summarize the source chunk as Markdown for a trading-literature infospace.
+Preserve in this order:
+
+- the narrator's actions and the market events they reacted to
+- named strategies, instruments, venues, and institutions
+- explicit lessons, rules of thumb, or warnings the text states
+- evidence phrases (dollar figures, dates, tape behaviour, counter-party
+  names) that should guide later entity and relation extraction
+- ambiguities or anachronisms that a reviewer should flag
+
+Keep the summary to a single page; do not paraphrase the moral of the
+chapter, only the material a downstream extractor needs.
+
+Source title: {{ input.title }}
+Source artifact: {{ input.artifact_id }}
+
+## Source
+
+{{ input.content }}

src/infospace_bench/profiles/trading-literature/templates/synthesize-report.md

@@ -0,0 +1,13 @@
+# Synthesize Trading-Literature Report
+
+Profile: {{ macros.profile }}
+
+Synthesize a concise review report from the generated source summaries,
+entities, relations, evaluations, and collection metrics. Group entities
+by category (trader, market, strategy, error, psychological pattern,
+institution, instrument, evidence-bearing claim). Surface the relations
+whose `relation_type` is `lesson_evidence` or `strategy_outcome` first —
+those are the operator-level findings a reviewer will want to read
+before anything else. End the report with an explicit "Overgeneralization
+risks" section that quotes any entities whose evaluation flagged that
+score below 3.