Rust Modu Extr Mono

Rust Module Extraction from Monolith main.rs

When to use

• Extracting structs/fns from a giant main.rs (or any monolith .rs) into submodule files
• Shrinking main.rs toward dispatch/orchestration only

Steps

1. Identify extraction candidates by dependency weight

   • Search for struct/fn definitions still inline in main.rs
   • For each, grep which crate-internal types it references
   • Classify: "clean leaf" (only uses types already pub in lib) vs "heavy" (uses pipeline internals, private state)
   • Extract clean leaves first; defer heavy ones

2. Bottom-up ordering

   • If struct A is nested inside struct B's fields, extract A first
   • Example: OptionsHedgingSection before TechnicalPriceSection (which embeds it)

3. Create the new module file

   • Place in the correct subdir (e.g., src/analyze/price_action.rs)
   • All extracted items must be pub
   • Import crate types via crate:: paths (e.g., crate::types::Candle)
   • Use use serde::Serialize; etc. as needed — don't assume parent re-exports

4. Register in mod.rs

   • Add pub mod new_module_name; to the parent mod.rs

5. CRITICAL PITFALL: analyze.rs vs analyze/mod.rs (E0761)

   • If the module was originally a single file src/analyze.rs and you later created src/analyze/mod.rs to hold submodules, BOTH files cannot coexist
   • Rust error: file for module found at both "src/analyze.rs" and "src/analyze/mod.rs"
   • The old src/analyze.rs content should have been migrated into src/analyze/mod.rs (or split into submodules) and the flat file DELETED
   • This error may be HIDDEN by incremental compilation cache — the binary (main.rs) shows a confusing "unresolved import" while cargo check --lib passes fine
   • Fix: rm src/analyze.rs (after confirming mod.rs has its content), then cargo clean and rebuild
   • Detection shortcut: if cargo check --lib works but cargo check --bin fails with unresolved module, run cargo clean to surface the real E0761 error

6. Wire imports in main.rs

   • Replace inline struct/fn with use crate_name::module::submodule::{StructName, fn_name};
   • Delete the old inline code
   • Remove now-unused imports (compiler warns about these)

7. Prefer wiring stable canonical subtrees before exporting more surface

   • If lib.rs / crate exports start drifting while extracting, stop broadening the public surface.
   • Freeze already-stable canonical leaves/subtrees and wire them into real call sites first.
   • In ict-engine, this proved safer for analyze/human_output, analyze/series, and analyze/multi_timeframe_parse than continuing to widen exports around AnalyzeMultiTimeframeInterval.
   • Heuristic:
     • if the extracted module already compiles under src/analyze/mod.rs
     • and main.rs still contains duplicate logic
     • then replace the duplicate logic with imports from the stable subtree
     • and avoid new lib.rs export churn unless a consumer truly needs it

8. Use compile errors as dependency-map hints during wiring

   • After swapping main.rs to canonical modules, run cargo fmt && cargo check immediately.
   • Missing-type errors often reveal exactly which section structs still need importing from the new module files.
   • In this repo, wiring builder fns from canonical analyze modules required also importing sibling section structs such as:
     • AnalyzeMultiTimeframeSection
     • TechnicalPriceSection
     • SmtCorrelationSection
     • OptionsHedgingSection
   • Fix imports first; only then remove now-unused legacy imports.

9. When adding a human-readable surface, append before replacing

• If the repo already emits JSON reports used by tests/workflows, do not immediately replace that output with human text.
• Safer pattern:
  • keep the canonical JSON println!("{}", serde_json::to_string_pretty(&report)?);
  • append a second human-readable rendering built from the canonical report object
• This productizes the extracted human-output module without breaking downstream JSON consumers.
• In ict-engine, a helper like render_human_analyze_output(&report) let analyze and analyze-live print the five-section human view after the JSON block.

12. When rebuilding graphify from repo-local Python fails, report missing module exactly

• Project docs may require:
  • python3 -c "from graphify.watch import _rebuild_code; from pathlib import Path; _rebuild_code(Path('.'))"
• In some environments this fails because the Python package is not importable even though project docs mention graphify.
• Exact observed failure pattern:
  • ModuleNotFoundError: No module named 'graphify'
• Do not claim graph rebuild succeeded if the import fails.
• Finish other required verification (cargo fmt, cargo check, cargo test) and explicitly report the graphify rebuild failure verbatim so the user can fix environment/package wiring.

13. For architecture refactors in drift-prone Rust monoliths, prefer parallel migration over hard replacement

• When introducing a new canonical object (for example a packet/bridge/core struct), do not immediately replace the old consumer-facing fields across the codebase.
• Safer sequence that emerged in ict-engine:
  1. get the repo back to a green baseline first
  2. add the new canonical struct to stable shared types
  3. keep old fields/surfaces intact
  4. add the new canonical field as Option<NewStruct> or as an internal builder output alongside the old surface
  5. make the duplicated builders populate both old and new surfaces from one canonical builder
  6. only after all consumers migrate, delete the old fields
• Reason: hard replacement in main.rs caused dozens of dependent call sites, tests, and report builders to break at once, obscuring the real migration path.
• Heuristic:
  • if changing one field name on a central report struct would force many unrelated consumers to update immediately, stop and switch to additive migration
  • if the repo is already non-green, do not stack architectural replacement on top of existing compile drift
• Phrase the work with positive constraints:
  • "keep old surface stable"
  • "populate new canonical packet in parallel"
  • "migrate consumers in batches"
  • "delete legacy surface only after green verification"

14. When repeated module-extraction attempts fail, stop brute-forcing the headline module and extract dependency substrata first

• In ict-engine, trying to move large top-level clusters (workflow, then factor_pipeline, then full builder bodies) failed repeatedly because the visible target functions still depended on many main.rs-local helpers.
• The robust sequence was:
  1. verify repo is green (cargo fmt && cargo check && cargo test)
  2. identify the repeated call-site duplication around the target surface
  3. extract adapter/re-export layers first into application modules
  4. extract pure helper substrata next (for example belief/pipeline shared helpers)
  5. only then attempt the real builder/body migration
• Practical signs you are still too early to move the headline function:
  • new module compiles only if it references crate::some_helper where that helper exists only in main.rs
  • extraction succeeds only as a facade/re-export while the real implementation remains in main.rs
  • repeated retries keep restoring a green repo but do not reduce the true blocker list
• In that situation, switch to a blocker-driven plan:
  • list the exact remaining local helpers
  • migrate the smallest pure subset first (math/normalization, trace builders, packet/debug adapters)
  • re-run full verification after every slice
• ict-engine specific lesson:
  • moving application/belief/shared.rs, then a debug adapter layer, then pipeline_shared.rs reduced coupling safely
  • attempting to hard-move builder implementations before infer_market_from_symbol, build_frame_features_for_market, pre_bayes_evidence_policy, and build_pre_bayes_evidence_filter were lib-visible still failed
• Rule of thumb:
  • a facade file is not proof of a real migration
  • only count the migration complete when the original main.rs function definitions are gone and search confirms that absence
• Verification standard:
  • after each extraction slice, run cargo fmt --all && cargo check && cargo test
  • after a claimed hard move, also search the original file for the old fn definitions to confirm they were truly removed

14. Do dependency-map discovery before extracting a giant main.rs cluster

• In ict-engine, an attempted first-slice extraction of the workflow/artifact cluster from src/main.rs looked locally contiguous but failed because the real dependency graph sprawled across analyze persistence, artifact diffs/decisions, lineage/trend helpers, prompt injection, and phase snapshot builders.
• Key lesson: textual adjacency in a monolith is not a sufficient extraction boundary.
• Before moving a large function family, first map:
  • direct callees inside the target span
  • helpers called outside the span
  • data types consumed from state::*
  • prompt/report builders that feed back into the same call path
• Practical method:
  1. identify the intended seam from the main call site
  2. search for every callee used by that seam
  3. group the extraction candidate by dependency closure, not by nearby line numbers
  4. if the closure crosses multiple orchestration concerns, stop and pick an earlier / narrower seam
• In this repo, build_workflow_snapshot was not the right first extraction slice even though it sat near related helpers.
• Better-first heuristic for this style of monolith:
  • extract the cluster that already has a partial canonical surface nearby
  • e.g. application/belief/* adjacent factor-pipeline builders before workflow governance
• Warning sign:
  • if a trial extraction requires creating a new module and then importing many unrelated main-local helpers, revert and choose a different slice rather than forcing a mega-move.

14. When extracting duplicated names into sibling modules, avoid parent-level glob re-exports

• In ict-engine, application::belief::debug_report and application::belief::pipeline_types both defined ExpansionLatestSignal, ExpansionProbabilitySupport, and ExpansionBbnSupport.
• A parent mod.rs with:
  • pub use debug_report::*;
  • pub use pipeline_types::*; created ambiguous glob re-exports and confused bin wiring during extraction.
• Safer pattern:
  • keep sibling modules explicit
  • import concrete types from debug_report::... or pipeline_types::...
  • re-export only stable parent API surface, not every sibling symbol
• Good mod.rs pattern:
  • re-export builder fns and top-level report structs explicitly
  • leave same-named helper types namespaced under their submodule
• Example fix:
  • pub use builder::{build_canonical_belief_report, build_canonical_belief_snapshot};
  • pub use debug_report::{build_factor_pipeline_debug_report, FactorPipelineDebugReport};
  • pub use pipeline_types::ExpansionFactorPipelineReport;
• Smell:
  • cargo check passes library but warns about ambiguous glob re-exports
  • main/bin imports need aliases like AppExpansion... / DebugExpansion...
  • tests start failing because one namespace silently changed while another stayed local

15. When a struct family is split into two semantic surfaces, convert call sites by destination type, not by search-replace aliasing

• In ict-engine, one family of names served two different destinations:
  • debug_report::Expansion* for debug-report builder inputs
  • pipeline_types::Expansion* for persisted pipeline report fields
• Naive aliasing in main.rs fixed compile errors in one path but broke tests and struct construction elsewhere.
• Safer sequence:
  1. remove duplicate local structs from main.rs
  2. classify each call site by the destination field/function signature
  3. for builder-call inputs, use the debug-report namespaced types
  4. for stored struct fields, use fully qualified pipeline-types structs
  5. rerun cargo check, then cargo test, because test fixtures often still instantiate the old local type
• Heuristic:
  • if one type name exists in two sibling modules, do not rely on a broad alias rename across the file
  • instead convert each usage based on the receiving API/field type
• Practical shortcut:
  • use fully-qualified paths for the stored/canonical surface in fixtures and constructors to make the distinction visually obvious

16. If the repo rejects a generated-orientation layer, remove both artifacts and workflow obligations

• In ict-engine, the user decided project-local graphify had no value.
• Proper cleanup was not just deleting graphify-out/; it also required deleting the AGENTS.md section that mandated reading/rebuilding graphify artifacts.
• Verification pattern:
  • remove graphify-out/
  • remove project instructions that require graphify use/rebuild
  • verify test ! -e graphify-out
  • verify no remaining project-local graphify obligations via search
• This prevents future agents from resurrecting removed tooling because stale project instructions still mention it.

17. When repeated extraction attempts fail, demote the target cluster and extract the shared substrate first

• In ict-engine, both of these seemingly natural "first slices" failed as direct extractions from main.rs:
  • workflow/artifact cluster
  • factor-pipeline builder cluster
• Root cause was the same: the target cluster still depended on a web of main.rs-private helpers, types, and adapter functions.
• Correct recovery pattern:
  1. try the intended slice once
  2. if compile errors reveal many private helper dependencies, stop forcing that slice
  3. identify the smallest reusable helper substrate shared by multiple future targets
  4. extract that substrate first into an application/library module
  5. keep the repo green after each failed experiment by reverting partial extraction attempts
• Practical heuristic:
  • if the extraction candidate needs more than a few unrelated main.rs helpers imported into the new module, it is not yet the right slice
  • if multiple future slices depend on the same helper family, extract the helper family before any business cluster
• In this repo, a workable first slice was a shared belief-helper substrate under src/application/belief/shared.rs, while direct workflow/factor-pipeline extraction was premature.

18. Use re-export shims to preserve public API while relocating internal ownership

• After moving shared helpers in ict-engine, the cleanest low-drift pattern was:
  • move real implementation into a new canonical file
  • leave old sibling modules as tiny pub use super::shared::{...}; shims
  • update parent mod.rs to declare the new module and export the intended stable API
• This let the repo gain canonical ownership without forcing every downstream call site to change immediately.
• Example pattern:
  • shared.rs owns the implementation
  • builder.rs becomes a re-export shim for canonical belief builders
  • debug_report.rs becomes a re-export shim for debug-report API/types
  • mod.rs declares pub mod shared; and re-exports the stable surface
• Use this when:
  • the new module boundary is correct
  • but you want to avoid a broad call-site rewrite in the same commit
• Verify with:
  1. cargo fmt
  2. cargo check
  3. cargo test

18b. Do not create cyclical type ownership between sibling modules when deduping duplicated Rust structs

• In ict-engine, pipeline_types.rs was made to pub use super::pipeline_shared::{ExpansionLatestSignal, ExpansionProbabilitySupport, ExpansionBbnSupport}; while pipeline_shared.rs still imported those same names from pipeline_types.rs.
• This created an unresolved/circular ownership situation:
  • pipeline_types re-exported from pipeline_shared
  • pipeline_shared imported from pipeline_types
  • cargo check then failed with unresolved imports / private item import errors
• Safe rule:
  • exactly one file owns the duplicated struct family
  • every other sibling only re-exports from that owner
  • the owner file must never import the same structs back through a sibling re-export
• In this repo, the stable ownership was:
  • pipeline_shared.rs owns ExpansionLatestSignal, ExpansionProbabilitySupport, ExpansionBbnSupport
  • pipeline_types.rs owns only ExpansionFactorPipelineReport
  • debug_report.rs re-exports the debug/report-facing surface from pipeline_shared.rs
• Practical migration order:
  1. pick canonical owner for the shared struct family
  2. move struct definitions there
  3. update all builder/internal imports to reference the owner directly
  4. only then add re-export shims in sibling modules
  5. run cargo check
  6. run cargo test
• Smell:
  • unresolved imports appear immediately after replacing local definitions with pub use ...
  • compiler mentions private unresolved item import or suggests importing through another re-export
• Fix:
  • break the cycle by changing the owner module to use local definitions directly, not sibling re-exports

18c. After a type dedupe, test fixtures may need state-type imports that were previously leaked through local duplicates

• In ict-engine, removing duplicate local ExpansionBbnSupport ownership exposed test code in main.rs that instantiated FactorPipelineLabelSource without importing it.
• Before dedupe, that dependency was easy to miss because nearby local structs visually carried the fields.
• Safe check after moving a shared struct family:
  1. run cargo check
  2. run cargo test
  3. inspect test fixtures that manually construct the moved structs
  4. add explicit imports for embedded state/domain types such as ict_engine::state::FactorPipelineLabelSource
• Heuristic:
  • if compile passes for lib but bin tests fail after a struct move, inspect fixture constructors before changing the moved structs again
• This is often just missing fixture imports, not a bad extraction boundary.

19. After the first shared shim works, split mixed-purpose shared.rs into purpose-named layers before larger extraction

20. When extracting a helper from main.rs that depends on monolith-local report structs, move logic behind a tiny trait instead of dragging those structs into the canonical module

• In ict-engine, resolved_multi_timeframe_inputs_for_market looked pure, but its parameter type was still a main.rs-local report struct family:

  • MultiTimeframeCleanFuturesReport
  • nested CleanFuturesReport
  • nested dataset structs

• Directly moving the function into application/multi_timeframe_inputs.rs would have forced one of two bad moves:

  1. copy/paste those report structs into the module, creating ownership drift
  2. widen the extraction scope into a much larger report-struct migration than intended

• Safer pattern:

  1. keep the canonical destination focused on the helper behavior
  2. define a tiny trait in the destination module that exposes only the data shape the helper actually needs
  3. implement that trait for the still-local main.rs report struct
  4. move the helper to operate on T: Trait
  5. rewire call sites, then delete the old local fn

• Concrete shape that worked:

  • trait method returning an iterator of (interval, output_path) pairs
  • helper consumes the trait, not the report struct family

• Why this is better:

  • reduces blast radius
  • avoids promoting temporary/report-only structs to canonical ownership prematurely
  • preserves monolith shrink progress without forcing a big data-model migration

• Common pitfall:

  • the trait method may need the same explicit lifetime on both &self and other borrowed inputs (for example market: &'a str) if the iterator closure captures them together
  • if cargo check reports lifetime may not live long enough, align the trait method parameter lifetime with the returned iterator lifetime

• Verification standard:

  • cargo fmt --all && cargo check
  • confirm the original helper body is gone from main.rs

• In ict-engine, the first workable move was a broad application/belief/shared.rs that temporarily held:

  • canonical belief builders
  • debug-report structs/builders
  • pipeline/debug adapter logic

• That was a useful intermediate state, but not the final architecture.

• Better follow-up pattern:

  1. get a green repo with the first shared shim
  2. identify the stable sub-surface that is really shared by one concern cluster
  3. move that sub-surface into a purpose-named module
  4. leave existing entry modules as thin re-export shims

• In this repo, the next refinement was:

  • create application/belief/pipeline_shared.rs
  • move pipeline/debug helper logic there
  • keep builder.rs and debug_report.rs as re-export surfaces
  • let shared.rs become thinner instead of growing into a junk drawer

• Heuristic:

  • if a new shared.rs starts accumulating unrelated responsibilities, freeze and split it before attempting the next big business-cluster extraction
  • name the second-layer module after the concern (pipeline_shared, workflow_shared, etc.), not after a generic reuse concept

• Benefit:

  • keeps the repo green while converging toward canonical ownership
  • reduces future extraction confusion because helpers are grouped by purpose instead of by "misc shared"

20. When moving a real function out of main.rs, migrate its local helper closure too — or decouple it first

• In ict-engine, moving build_pre_bayes_evidence_filter into src/config.rs still failed at first because it secretly depended on a sibling helper, pre_bayes_distribution, that remained defined only in main.rs.
• The safe checklist for a claimed "real move" is:
  1. search for the target fn body in main.rs
  2. inspect every non-stdlib helper it calls
  3. classify each helper as:
     • already lib-visible
     • must move together
     • must be replaced by a new module-local helper
  4. only then delete the old main.rs body
• Do not assume compile errors will name the full dependency closure up front; the first removed function often exposes a second hidden local helper on the next cargo check.
• Practical sign:
  • after the move, lib compilation fails with cannot find function ... in the crate root or similar visibility errors from the new module.
• Fix pattern:
  • move the small helper into the same destination module if it is only used by the extracted function
  • update internal calls to plain module-local invocation rather than crate::... if the helper is not part of the crate root API
• In this case, pre_bayes_distribution belonged with the extracted pre-Bayes filter logic inside config.rs.

21. When a monolith slice has been successfully re-extracted once, a later emergency git checkout -- file revert can silently reintroduce the old duplicates; re-audit from live files, not from memory

• In ict-engine, a full-file revert used to recover from an unrelated truncation/dirty-state problem restored old main.rs implementations that had already been migrated earlier:
  • left_pad
  • infer_market_from_symbol
  • pre_bayes_evidence_policy
  • pre_bayes_distribution
  • pre_bayes_market_policy_override
  • build_pre_bayes_evidence_filter
• The dangerous trap is assuming the repo still contains the previously completed extraction just because the conversation remembers it.
• Safe recovery sequence after any broad revert:
  1. re-read the destination module and main.rs
  2. search for the old fn definitions in main.rs
  3. confirm which migrations survived and which were rolled back
  4. resume with the smallest previously proven-safe slice, one function at a time
• Practical heuristic:
  • after git checkout -- src/main.rs or any whole-file restore, treat all prior extractions touching that file as untrusted until verified by search
  • conversation history is not proof of repository state
• In this repo, the stable recovery path was:
  • re-establish green baseline
  • re-move left_pad
  • re-move infer_market_from_symbol
  • re-move pre_bayes_evidence_policy
  • re-move pre_bayes_distribution
  • re-move pre_bayes_market_policy_override
  • re-move build_pre_bayes_evidence_filter
  • then re-move FrameFeatures / INDICATOR_PERIOD / build_frame_features
  • then re-move build_frame_features_for_market
• Benefit:
  • avoids mixing recovery from state drift with fresh large-slice extraction
  • each re-applied step is independently validated with cargo fmt && cargo check (and targeted tests where available)

22. When doing scripted block deletion in a huge Rust file, anchor on both start and end targets and immediately verify collateral loss

• In ict-engine, a scripted deletion that removed the old pre-Bayes block from main.rs also erased the adjacent build_frame_features_for_market function because the removal span was anchored too broadly.
• For giant monolith edits, use this discipline:
  1. identify exact start marker for the first function to delete
  2. identify exact end marker for the next function that must remain
  3. after script/edit runs, search for adjacent keeper fns that should still exist
  4. run cargo check immediately; missing adjacent helpers often surface faster than re-reading the whole file
• Heuristic:
  • if using a script or broad text replace on a 10k+ line file, assume collateral deletion is possible
  • verify both:
    • removed target definitions are gone
    • neighboring required definitions still exist
• Recovery pattern:
  • restore the missing neighbor function first
  • then rerun fmt/check/tests before declaring the extraction stable
• In this case, build_frame_features_for_market had to be restored after the scripted deletion removed more than intended.

22. If a nearby function cannot move cleanly because its core type still belongs to main.rs, use a temporary policy/helper extraction instead of forcing the full builder move

• In ict-engine, the next drift target after pre-Bayes was build_frame_features_for_market.
• A full move was initially blocked because the function mutates FrameFeatures, and that struct still lived privately in main.rs alongside build_frame_features.
• The low-drift intermediate pattern was:
  1. keep the thin wrapper fn in main.rs
  2. extract only the market-specific override policy into a lib module (config.rs in this case)
  3. have the wrapper compute/own FrameFeatures, clone the mutable labels, call the extracted helper, then write the labels back
  4. verify with cargo fmt && cargo check
• This is not a "true move" of the wrapper, but it is still useful drift reduction because the branchy market policy leaves main.rs.
• Rule of thumb:
  • if the function's main burden is policy logic and its remaining shell only adapts a local/private type, extract the policy first
  • defer the real move until the underlying struct and base builder are also ready to migrate
• Naming guidance:
  • alias the extracted helper at import site to make the temporary layering explicit, e.g. build_frame_features_for_market as apply_market_frame_overrides
  • this reduces confusion while both a wrapper and a helper temporarily share similar names
• Verification standard:
  • do not claim the original function is gone if a wrapper remains
  • instead report honestly: logic moved, thin wrapper remains, full migration requires moving the owning type next

23. When a rollback wipes out prior successful micro-migrations, re-run the sequence as strict one-function slices and re-check live file state before every patch

• In ict-engine, a broad git checkout -- main.rs config.rs restored green state but also silently erased several already-successful extractions (left_pad, pre-Bayes helpers, frame helpers, trace helpers).
• The robust recovery pattern was:
  1. re-read the live files first instead of assuming prior moves still exist
  2. search for exact remaining fn ... definitions in main.rs
  3. re-apply migrations in the smallest possible slices, one function at a time
  4. after each slice, run cargo fmt && cargo check and at least one focused regression test
• Practical lesson:
  • after any full-file rollback, your earlier session memory is not source of truth; the repo is
  • always inspect current file contents before writing the next patch, even if you "just did" that migration earlier
• Good micro-order from this run:
  • left_pad
  • infer_market_from_symbol
  • pre_bayes_evidence_policy
  • pre_bayes_distribution
  • pre_bayes_market_policy_override
  • build_pre_bayes_evidence_filter
  • build_frame_features
  • build_frame_features_for_market
  • trace helpers
  • multi_timeframe_entry_quality_bias
• Why this worked:
  • each step had tiny blast radius
  • failures were attributable to the current slice, not hidden collateral from previous large edits
  • repeated verification kept the monolith green while shrinking it
• Rule:
  • if a previous rollback or file truncation occurred, stop doing batch migrations; switch to one-definition-at-a-time extraction until stability returns

24. Patch tools may report false-positive lint noise after import-list edits; trust live cargo fmt && cargo check over patch-tool diff formatting complaints

• In ict-engine, small import edits often came back with patch-tool lint output showing wrapped/reflowed import groups as if they were errors.
• These were not semantic compile problems; the authoritative check remained cargo fmt && cargo check.
• Use this discipline:
  1. apply the small textual patch
  2. ignore cosmetic import reflow warnings from the patch tool unless they indicate true parser failure
  3. immediately run cargo fmt && cargo check
  4. only treat the change as bad if the Rust toolchain rejects it
• Especially common triggers:
  • removing a single symbol from a long use ...::{...} line
  • adding one more imported item that forces rustfmt to wrap the group
• Rule of thumb:
  • patch-tool unified diff formatting complaints are advisory
  • Rust compiler + rustfmt are the real gate for extraction work

25. When moving a helper that depends on a tiny local utility, inline the normalization locally if the dependency is otherwise not worth extracting yet

• In ict-engine, multi_timeframe_entry_quality_bias depended on normalize_distribution(&mut bias) which still lived in main.rs.
• Instead of turning that into a blocker or widening the migration scope, the safe move was to inline the tiny normalization logic inside the migrated helper:
  • compute sum
  • divide by sum when non-zero
  • otherwise assign uniform weights
• This preserved behavior while avoiding a premature extra extraction.
• Rule:
  • if the only blocker is a tiny pure helper with obvious local behavior, consider inlining it into the new canonical module rather than expanding the migration surface
• Prefer this only when:
  • the helper is short
  • semantics are obvious
  • duplication cost is lower than pulling another dependency chain across module boundaries
• Verify with a focused unit test covering the moved helper's behavior.

23. When a rollback is required mid-extraction, immediately re-scan live files before continuing — prior migration state may be gone

• In ict-engine, a broad git checkout -- main.rs config.rs used to recover a green repo also erased several already-completed extractions (left_pad, infer_market_from_symbol, pre_bayes_evidence_policy, pre_bayes_distribution, pre_bayes_market_policy_override, build_pre_bayes_evidence_filter, build_frame_features, build_frame_features_for_market, and trace helpers).
• After any rollback or restore command:
  1. re-read the destination module and main.rs
  2. search for the exact fn definitions again
  3. treat the repo as a new baseline, not as if previous migration progress still exists
• Do not stack new deletions/import rewires on top of remembered state; that caused duplicate imports, missing helpers, and confused retry sequences.
• Practical rule:
  • after a rollback, the next step is always search_files/read_file, never assumption-based patching.

24. For drift-prone monolith extraction, use a repeated 'single-function migration loop' instead of batch moves

• The stable loop that worked in ict-engine was:

  1. pick exactly one function/helper
  2. copy it to the target lib module with minimal imports
  3. rewire one import/use site in main.rs
  4. delete only that one original definition
  5. run cargo fmt && cargo check
  6. if relevant, run 1-3 focused tests

• This succeeded repeatedly for:

  • left_pad
  • infer_market_from_symbol
  • pre_bayes_evidence_policy
  • pre_bayes_distribution
  • pre_bayes_market_policy_override
  • build_pre_bayes_evidence_filter
  • build_frame_features
  • build_frame_features_for_market
  • raw_market_regime_trace
  • raw_liquidity_context_trace
  • raw_multi_timeframe_resonance_trace

• Heuristic:

  • if a migration candidate is larger than one function plus trivial import cleanup, split it again
  • prefer many green micro-moves over one ambitious cluster move

• Benefit:

  • failures stay local
  • rollbacks cost less
  • the user gets truthful incremental progress without hidden drift

• Visibility: extracted structs need pub on both the struct and its fields if main.rs was constructing them directly. If fields were private (no pub), the builder fn must live in the same module.

  • Safer recovery rule:
    1. before any broad checkout/reset, note exactly which migrations already landed successfully
    2. prefer reverting only the latest target files or blocks, not the entire working tree slice, when earlier migrations are known-good
    3. after any broad rollback, assume earlier successful extractions may have been undone and re-audit current file state before continuing
  • Robust re-application pattern for a monolith shrink task:
    1. re-read the destination module and source monolith fresh after rollback
    2. search for the exact current inline definitions still present
    3. re-apply extractions one function at a time in ascending-risk order
    4. after each function, run cargo fmt && cargo check plus one focused regression test
  • Practical safe order discovered here:
    • left_pad
    • infer_market_from_symbol
    • pre_bayes_evidence_policy
    • pre_bayes_distribution
    • only then consider heavier functions like pre_bayes_market_policy_override, build_pre_bayes_evidence_filter, build_frame_features, and wrappers
  • Heuristic:
    • move the smallest pure helpers first
    • move already-lib-owned equivalents before policy-heavy builders
    • avoid touching trace helpers or multi-function clusters until the small helper ladder is re-established and green
  • Reporting rule:
    • after a destructive rollback, explicitly tell the user that prior true moves were reverted and that the task is restarting from current file reality, not prior chat history.

23. In drift-heavy monolith extraction, if you must revert to recover green, assume every prior "moved" helper may have been silently undone and re-audit from live files before the next slice

• In ict-engine, a broad git checkout -- src/main.rs src/config.rs was the correct recovery after half-finished trace-helper extraction destabilized the repo.
• But that reset also silently erased earlier successful mini-migrations (left_pad, build_frame_features, pre-Bayes helpers, promoted types/constants) because they were not yet committed separately.
• The practical lesson:
  1. after any full-file rollback, do not trust conversational state or your own prior summary
  2. re-read the live destination/source files
  3. search for the exact fn definitions and imports you believe were moved
  4. only then pick the next slice
• Good recovery checklist after a rollback:
  • read_file destination module to see what actually remains
  • search_files for the old fn ... definitions in main.rs
  • treat the current tree as ground truth, not prior chat memory
• Strategy consequence:
  • when rollback risk is high, reduce extraction to one truly minimal helper per step (e.g. left_pad alone), verify, then proceed
  • avoid stacking several successful-but-uncommitted migrations in working tree state and then doing a broad file reset
• Reporting rule:
  • if a rollback erased prior progress, say so explicitly; do not pretend later steps are building on state that no longer exists.

24. After a rollback, prefer re-entry through the smallest already-proven slice instead of resuming the ambitious target that triggered the reset

• In ict-engine, after reverting main.rs and config.rs to restore green, the correct re-entry was not "continue the trace-helper migration".
• The stable re-entry path was:
  1. confirm green repo (cargo fmt && cargo check)
  2. re-audit actual file contents
  3. restart from the smallest safe helper (left_pad)
  4. re-verify
  5. only then consider the next slice (e.g. infer_market_from_symbol)
• Heuristic:
  • if the previous attempt ended in a broad rollback, your next move should shrink scope, not match the prior scope
  • choose the smallest extraction that has low dependency surface and clear verification value
• Benefit:
  • this re-establishes momentum while reducing the chance of another destabilizing revert cycle.

23. When a rollback restores a monolith file, immediately reapply only the already-validated extractions in dependency order

• In ict-engine, recovering src/main.rs with git checkout -- src/main.rs fixed file corruption but also restored many previously removed inline helpers.
• Blindly replaying the whole migration script reintroduced duplicate definitions and compile failures.
• Safer recovery sequence:
  1. restore the corrupted file to a green baseline
  2. compare current main.rs against already-canonical lib surfaces
  3. reapply only the proven moves, in dependency order
  4. after each reapplication, run cargo fmt && cargo check
• For this repo, the stable reapply order was:
  1. import canonical config::{build_frame_features, build_pre_bayes_evidence_filter, left_pad, FrameFeatures, INDICATOR_PERIOD}
  2. remove local FrameFeatures / INDICATOR_PERIOD / build_frame_features
  3. keep build_frame_features_for_market a

…(truncated)