iatro-clintrace 1.0.1

ClinTrace clinical-course compression, grounding, and longitudinal decision modeling

Iatro ClinTrace

EN | 中文

Gated model weights: Hugging Face | ModelScope. Both repositories contain the same ClinTrace model artifact; clinical assets and patient-level outputs are not distributed.

ClinTrace is a longitudinal clinical-course model for HCC notes. Its job is to turn free-text clinical notes into a reusable course representation that can support treatment-pathway reasoning, evidence-grounded decision completion, and downstream model integration.

Research use only. ClinTrace is not a medical device and must not be used for clinical diagnosis, treatment selection, triage, or patient management.

The authoritative design record is docs/clintrace_total_design.md. This README is the open-source entrypoint: it describes the package boundary, public command surface, and controlled-asset contract.

Table of Contents

• Scientific Contract
• Frozen Model
• Release Boundary
• Installation
• Quickstart: Local Inference
• Training Workflow
  • 1. Build Source IAC Assets
  • 2. Build or Import Compressed Text Assets
  • 3. Build and Check Paired IAC Assets
  • 4. Train, Select, and Export the Compressor
  • 5. Build Frozen Note Features
  • 6. Build Decision Supervision
  • 7. Build Direct-Token and Decoder Planes
  • 8. Build History-State and Decision-Contrast Assets
  • 9. Train the ClinTrace Decision Expert
  • 10. Evaluate and Freeze Release Candidates
• Configuration
• Optional LLM Client Utilities

Scientific Contract

ClinTrace is released as one final clinical-course model, not as separate curriculum checkpoints. It does not consume manually supplied decision-chain labels at inference time.

input
  ordered preceding-admission documents -> H
  current clinical evidence package E

supervision and audit
  LLM-extracted E/P/X/O chains
  timeline-derived labels
  history-state summaries

output
  ranked PX decision candidates
  candidate-conditioned observed O distributions
  clinician-facing trusted source-evidence references

E/P/X/O labels, chain metadata and timeline-derived fields define the supervision, evaluation, and audit assets. Completion is conditioned on H + E, while demonstration callers keep each supplied evidence item linked to its trusted source record for review. Displayable evidence comes from controlled source spans, not token-offset fragments chosen by the model.

ClinTrace also treats document inclusion and admission ordering as an upstream asset contract. Users decide which notes belong in the training/release asset before building IAC packs. The framework does not apply private note-type rules to filter longitudinal history. Within the same admission, users must provide a stable clinical order through admission/episode metadata, timestamps, and stable document keys; ClinTrace consumes that order rather than correcting ambiguous hospital-system timestamps.

The central modeling target is:

history state + observed clinical evidence -> plausible P/X/O decision candidates

This is intentionally open-world. Missing labels in the record are treated as unobserved, not as clinical negatives. The model is evaluated primarily by whether documented decisions are ranked highly. Review inspects the supplied evidence and its trusted source reference; the final model does not generate evidence spans from a full current note.

Frozen Model

The frozen ClinTrace model accepts ordered admission history H and current evidence E:

(H,E) \longrightarrow \mathrm{candidate}\ PX\ \mathrm{slot}
\\
(H,E,PX\ \mathrm{slot}) \longrightarrow \mathrm{observed}\ O\ \mathrm{distribution}

Hungarian matching uses the observed P and/or X facets available for an O bundle to assign it to a generated candidate slot. Conditional-O metrics score observed O labels within their native facet; unobserved labels remain unknown, not negatives. The conditional-outcome branch does not alter completion-ranking parameters.

The selected final checkpoint SHA-256 is 665fcd8779fa884101edee532060bd00bca35f659647c739bad6ed9b947ff709. Its frozen decision-backbone source SHA-256 is 59134e4ab5e96fb73e5c35220f8641dad1d09228acadda7b542429bfaf17873c. The recorded final configuration SHA-256 is 1d6907e6f319102d8e492c1f92c3531b7d51aa807af36cb922b241619a6ea98b.

Selection used validation conditional-O loss. Final evaluations wrote no patient-level prediction exports.

Split	PXO R@5	O facet targets	O R@1	O R@3	O R@5	O MRR	O observed-set NLL
Validation	0.880236	3,551	0.615883	0.901436	0.957477	0.763672	1.141718
Internal test	0.879324	2,365	0.611416	0.897252	0.957294	0.759065	1.089216
External	0.808669	17,608	0.548046	0.877101	0.946047	0.719123	1.505285

PXO ranking equals the frozen decision-backbone evaluation, confirming that the conditional-outcome branch did not alter the primary completion task.

Release Boundary

The GitHub repository contains source code, reproducible configurations, packaged prompt contracts, tests, and documentation. The gated Hugging Face and ModelScope repositories distribute the corresponding final model artifact.

Installation

Default installation includes the released inference runtime, IAC tooling, and the local LLM client utilities.

pip install iatro-clintrace

Install the single extension only when rebuilding, training, evaluating, or packaging a model:

pip install "iatro-clintrace[full]"

Development checkout:

pip install -e .
pip install -e ".[full]"

Quickstart: Local Inference

Start from a local clone so that the command-line interface and documentation stay together:

git clone https://github.com/iatrode/iatro-hcc-clintrace.git
cd iatro-hcc-clintrace
pip install .

Request access to the gated release through either Hugging Face or ModelScope. Once access is approved, download the complete release once:

clintrace download

The command chooses ModelScope for a China public IP and Hugging Face otherwise, then reuses the selected hub's existing local login. It also uses HF_TOKEN or MODELSCOPE_API_TOKEN when that environment variable is already configured.

If no local login or configured token is available, either authenticate with an official client first:

hf auth login
# or
modelscope login

or provide the gated token for this download explicitly:

clintrace download --token "$HF_TOKEN"
# or
clintrace download --token "$MODELSCOPE_API_TOKEN"

For explicit tokens, ClinTrace selects the hub from the token format (hf_ for Hugging Face and ms- for ModelScope), rather than from public-IP location.

The official clients can also download the gated repository directly. In that case pass the resulting directory explicitly with --release-dir when running clintrace infer or clintrace compress. ClinTrace stores its own downloaded bundle under ~/.cache/iatro-clintrace/releases/{hf|modelscope}/; infer and compress only load local weights and never initiate a download themselves.

Prepare ordered preceding-admission documents as separate UTF-8 files, and put the current evidence package E in a separate UTF-8 file. The current evidence file is not a full current note and must not contain P/X/O target annotations. Then run:

clintrace infer \
  --history-note examples/toy/inference/prior_admission_note_01.txt \
  --history-note examples/toy/inference/prior_admission_note_02.txt \
  --evidence-note examples/toy/inference/current_evidence.txt \
  --evidence-source "toy-current-evidence" \
  --top-k 5

The terminal report presents ranked PX candidates and candidate-conditioned O estimates. --evidence-source is displayed for review and never enters the model. Use --json --output result.json only when a machine-readable result is needed in an approved local workflow.

examples/toy/inference/ contains the example input layout used above.

Training Workflow

The section below documents the full reproduction path used to rebuild release artifacts. It is not the normal runtime path for future users. Commands use the local data/, runs/, and artifacts/ layout. Aggregate strategy selection is recorded in docs/training_strategy_record.md.

Before the first training command, create local configuration copies. Repository configs use demo_* asset identifiers and are templates, not ready-to-run datasets. In each local copy, replace the paths for decision samples, extracted chains, timeline index, compressor features, direct-token features, admission-history states, and decoder input plane. Set the local train and external institution identifiers at the same time. The history and navigation configurations use the label_vocab.json written by the completion run that supplies their initialization checkpoint.

cp configs/decision/default.yaml configs/decision/local_grounding_completion.yaml
cp configs/decision/stage3_final_earlystop.yaml configs/decision/local_history.yaml
cp configs/decision/stage4_navigation_final.yaml configs/decision/local_navigation.yaml

1. Build Source IAC Assets

Start from a de-identified document table in JSONL, CSV, or TSV. Required columns are:

patient_id      stable de-identified patient identifier
doc_id          stable de-identified document identifier, unique within dataset
text            clinical note text

Optional columns are:

doc_type        observable note title/type
doc_timestamp   sortable clinical timestamp string
episode_index   integer visit/admission index, -1 if unknown
encounter       optional encounter identifier
doc_category    optional source category, defaults to clinical_document
source          optional source-system label

Before packing, finalize the source asset scope. Remove documents that should not be part of longitudinal history or supervision, and normalize admission or encounter boundaries. For same-admission records, provide timestamps and stable document identifiers that encode the intended clinical order. The source packer performs deterministic sorting; it does not infer clinical sequence from damaged timestamps or apply built-in note-type filtering.

Pack the table into a clinical_text IAC:

clintrace build source-iac \
  --input data/source_documents/demo_train.jsonl \
  --output data/00_source/demo_train.iac \
  --institution demo_train \
  --overwrite

The source pack builder sorts documents within each patient by episode_index, doc_timestamp, and doc_id. This is a deterministic serialization rule over user-supplied metadata, not a clinical reordering algorithm. It skips rows with missing required fields, duplicate doc_id, or empty text. It reports:

input_rows=<input table rows>
patients=<packed patient count>
docs=<packed document count>
duplicate_doc_ids=<skipped duplicate document ids>
empty_text=<skipped empty documents>
missing_required=<missing required-field counts>
text_raw_mb=<UTF-8 text payload size>
output=<source IAC path> size_mb=<written pack size>

Verify the source pack:

clintrace build source-iac \
  --input data/source_documents/demo_train.jsonl \
  --output data/00_source/demo_train.iac \
  --institution demo_train \
  --verify

Verification must report:

verify_docs == verify_expected_docs
verify_unique_doc_ids == verify_expected_docs
verify_mismatches == 0

The resulting asset is:

data/00_source/{site}.iac
  clinical_text pack containing original de-identified notes.

2. Build or Import Compressed Text Assets

Create or import compressed note text before compressor training. This asset is a normal JSONL, CSV, or TSV table, not an IAC pack. Required columns are:

doc_id             same de-identified document identifier as source IAC
compressed_text    compressed clinical note text

Accepted aliases for compressed_text are text and output. Optional columns are doc_type, doc_timestamp, and episode_index; when omitted, the source IAC metadata is used.

data/01_compressed/{site}.jsonl
  compressed-note table keyed by doc_id.

The compression method is outside this public contract. It may be produced by a human-reviewed pipeline, a local model, or an external system; ClinTrace only requires aligned doc_id values and clinically readable compressed text. If a dataset requires de-identification or identifier alignment, complete that data preparation before building ClinTrace pairs. The open-source package does not define or distribute private identifier-mapping protocols. For the optional LLM client wrapper that uses the active packaged profile prompt and writes a paired-iac-ready JSONL, see Optional LLM Client Utilities.

Extract EPXO supervision from the same compressed-note rows before building decision supervision. Configure an OpenAI-compatible endpoint locally; the examples use environment variables for endpoint configuration.

export CLINTRACE_LLM_API_BASE="https://your-local-or-approved-endpoint/v1"
export CLINTRACE_LLM_API_KEY="..."
export CLINTRACE_LLM_MODEL="..."

clintrace llmc expo \
  --institution demo_train \
  --input data/01_compressed/demo_train.jsonl \
  --out data/05_decision_chains/results.jsonl \
  --api-base "$CLINTRACE_LLM_API_BASE" \
  --api-key "$CLINTRACE_LLM_API_KEY" \
  --model "$CLINTRACE_LLM_MODEL"

clintrace llmc expo \
  --institution demo_external \
  --input data/01_compressed/demo_external.jsonl \
  --out data/05_decision_chains/results.jsonl \
  --api-base "$CLINTRACE_LLM_API_BASE" \
  --api-key "$CLINTRACE_LLM_API_KEY" \
  --model "$CLINTRACE_LLM_MODEL"

3. Build and Check Paired IAC Assets

Once the source IAC and compressed table share de-identified document identifiers, build the source/compressed training pairs:

data/02_paired/{site}.iac
  source/compressed clinical_text_pair pack used by the compressor and router.

Build a paired source/compressed pack:

clintrace build paired-iac \
  --source-iac data/00_source/demo_train.iac \
  --compressed-input data/01_compressed/demo_train.jsonl \
  --output data/02_paired/demo_train.iac \
  --institution demo_train \
  --overwrite

paired-iac is the only command that writes compressed text into an IAC asset. There is no standalone compressed IAC in the public workflow. If a dataset needs de-identification or identifier mapping to make source and compressed IDs align, complete that preparation before paired-iac. The paired IAC itself is source text paired with compressed text; it is not a de-identified/non-de-identified pair.

Record the following IAC-level acceptance metrics before model training:

patients=<paired patient count>
paired_docs=<paired document count>
source_docs=<source document count>
compressed_rows=<compressed input table rows>
compressed_docs=<unique usable compressed document count>
duplicate_doc_ids=<skipped duplicate compressed document ids>
missing_compressed=<source docs without compressed counterpart>
empty_source=<skipped empty source docs>
empty_compressed=<skipped empty compressed docs>
missing_required=<missing required-field counts>
source_raw_mb=<UTF-8 source text payload size>
compressed_raw_mb=<UTF-8 compressed text payload size>
output=<paired IAC path> size_mb=<written pack size>

Verify the paired pack:

clintrace build paired-iac \
  --source-iac data/00_source/demo_train.iac \
  --compressed-input data/01_compressed/demo_train.jsonl \
  --output data/02_paired/demo_train.iac \
  --institution demo_train \
  --verify

Verification must report:

verify_docs == verify_expected_docs
verify_unique_doc_ids == verify_expected_docs
verify_mismatches == 0

Then inspect both the raw pack and paired pack before starting a long run:

clintrace inspect iac data/00_source/demo_train.iac --head 5
clintrace inspect pairs -i demo_train --head 5

The IAC acceptance gate is: all expected in-scope documents are paired exactly once, source/compressed text round-trips without mismatch, admission boundaries and within-admission order are already resolved by the user-provided metadata, and sampled records are clinically readable after de-identification.

4. Train, Select, and Export the Compressor

The compressor maps a clinical note to fixed note tokens. Train it from the paired source/compressed IAC:

clintrace train compressor --config configs/compressor/train.yaml

Training writes checkpoints and metrics under runs/ or the configured compressor output directory. Select the checkpoint by the predeclared validation rule, then export the release-sized compressor artifact into artifacts/compressor/:

clintrace build compressor-artifact \
  --checkpoint runs/compressor/checkpoints/course_encoder_step5500.pt \
  --output artifacts/compressor/clintrace_compressor_qwen35_2b_n32_d768_step5500.pt

The compressor training validation rule and the exported-artifact metadata are the training-stage acceptance record. clintrace compress intentionally loads only a complete released model bundle, so it is not a checker for this local intermediate .pt. Continue with feature generation after selecting and exporting the compressor artifact.

5. Build Frozen Note Features

Create longitudinal indexes and frozen compressor note-token caches. These caches are the model input substrate for downstream decision training.

The timeline index records document order and audit metadata. It does not decide which documents enter longitudinal history. Downstream history is consumed from fixed assets such as history_document_keys or admission-state rows.

clintrace build timeline
clintrace build compressor-features --checkpoint artifacts/compressor/clintrace_compressor_qwen35_2b_n32_d768_step5500.pt

6. Build Decision Supervision

Compile the extracted EPXO chains into supervision assets. The chain labels and spans are used only as training/evaluation targets, not as inference inputs. History context in the compiled supervision is written explicitly as history_document_keys; training consumes those fixed keys instead of reconstructing history with note-type filters.

clintrace build grounding-supervision \
  --chains data/05_decision_chains/results.jsonl \
  --timeline-dir data/03_timeline_index \
  --features-dir data/04_clintrace_features \
  --output-dir data/06_decision_supervision \
  --train-institutions demo_train \
  --external-institutions demo_external \
  --overwrite

The institution names above are placeholders. Use the local de-identified train and external-site identifiers from your own data/03_timeline_index/*.jsonl manifests. If no institution split is supplied, the first discovered institution is treated as train and the rest as external.

Within each train institution, patients are assigned deterministically to 75% train, 15% validation and 10% test. The assignment keeps patients intact while balancing document-level EPXO bundle signatures, primary supervision roles, chain-count bins and document roles; inspect the generated report.json before training.

7. Build Direct-Token and Decoder Planes

Build the direct-token feature plane used for compact or already-compressed text, then cache per-note grounding features.

clintrace build decoder-plane
clintrace build grounding-features \
  --documents data/06_decision_supervision/document_samples.jsonl \
  --chains data/05_decision_chains/results.jsonl \
  --checkpoint artifacts/compressor/clintrace_compressor_qwen35_2b_n32_d768_step5500.pt \
  --output-dir data/07_grounding_features \
  --overwrite

grounding-features does not read configs/decision/default.yaml; pass its document table, EPXO chain table, compressor checkpoint, and output directory explicitly.

8. Build History-State and Decision-Contrast Assets

Admission-history-state summaries provide longitudinal patient context for training. Decision-contrast pairs are conservative strict-far EPXO bundle comparisons used to reduce near-neighbor confusion without converting unrecorded decisions into negatives.

The same locally configured endpoint is used for admission-history state construction:

clintrace llmc history \
  --institution demo_train \
  --compressed-input data/01_compressed/demo_train.jsonl \
  --timeline-dir data/03_timeline_index \
  --output-dir data/08_patient_state_history \
  --api-base "$CLINTRACE_LLM_API_BASE" \
  --api-key "$CLINTRACE_LLM_API_KEY" \
  --model "$CLINTRACE_LLM_MODEL"

clintrace llmc history \
  --institution demo_external \
  --compressed-input data/01_compressed/demo_external.jsonl \
  --timeline-dir data/03_timeline_index \
  --output-dir data/08_patient_state_history \
  --api-base "$CLINTRACE_LLM_API_BASE" \
  --api-key "$CLINTRACE_LLM_API_KEY" \
  --model "$CLINTRACE_LLM_MODEL"

clintrace build decision-contrast \
  --samples data/06_decision_supervision/samples.jsonl \
  --output-dir data/06_decision_supervision/expo_distance_analysis

9. Train the ClinTrace Decision Expert

The decision expert is trained through a four-part curriculum. The command names describe the capability being learned, not separate deployable experts.

clintrace train grounding \
  --config configs/decision/local_grounding_completion.yaml \
  --output-dir runs/clintrace_grounding \
  --overwrite

clintrace train completion \
  --config configs/decision/local_grounding_completion.yaml \
  --initialization grounding_init \
  --grounding-checkpoint runs/clintrace_grounding/clintrace_grounding.pt \
  --output-dir runs/clintrace_completion \
  --overwrite

clintrace train longitudinal \
  --config configs/decision/local_history.yaml \
  --initialization completion_init \
  --completion-checkpoint runs/clintrace_completion/clintrace_decision_expert.pt \
  --output-dir runs/clintrace_longitudinal \
  --eval-splits validation test external \
  --overwrite

clintrace train navigation \
  --config configs/decision/local_navigation.yaml \
  --checkpoint runs/clintrace_longitudinal/clintrace_longitudinal.pt \
  --output-dir runs/clintrace_navigation \
  --overwrite

Grounding, completion, and longitudinal completion accept the same runtime override flags: --output-dir, --device, --batch-size, --token-budget, --num-workers, --max-epochs, and --overwrite. Completion additionally accepts --grounding-checkpoint; longitudinal accepts both --grounding-checkpoint and --completion-checkpoint.

grounding trains the span-grounding auxiliary. completion learns observed-E to P/X/O completion. The first longitudinal run adds admission history state H and strict-far ranking regularization. navigation accepts only the history-conditioned completion checkpoint, freezes that pathway, and trains only the candidate-conditioned observed-O branch.

Formal training defaults to data.feature_loading: preload for the >=32 GB training target. Use data.feature_loading: mmap only as the lower-memory alternative: workers retain first-touch shards as read-only memory mappings, while each batch coalesces direct/history shard reads and prefetches subsequent complete batches without LRU tensor caches or dense-bank copies.

grounding writes a content-light training sample order plan to data.sample_order_plan. The plan contains sample identifiers only, not tensors or clinical text. completion and longitudinal reuse that sample order when available, then repack batches under their own token-budget rules so the three phases share the same auditable order without forcing identical batch shapes.

When only EPXO chains are refreshed after the compressor note features have already been fixed, keep data/00_source through data/04_clintrace_features and the existing data/08_patient_state_history, then rebuild:

data/05_decision_chains
data/06_decision_supervision
data/07_grounding_features
runs/clintrace_grounding*
runs/clintrace_completion*
runs/clintrace_longitudinal*

10. Evaluate and Freeze Release Candidates

Evaluate every selected checkpoint on validation, internal test, and external splits. The gated release candidate is chosen from the complete validation chain, not from an ad hoc artifact bundle.

clintrace evaluate grounding --split validation
clintrace evaluate completion --checkpoint runs/clintrace_completion/clintrace_decision_expert.pt --split test
clintrace evaluate longitudinal --checkpoint runs/clintrace_longitudinal/clintrace_longitudinal.pt --split test
clintrace evaluate navigation \
  --config configs/decision/local_navigation.yaml \
  --checkpoint runs/clintrace_navigation/clintrace_navigation.pt \
  --split test \
  --output-dir runs/clintrace_navigation/evaluation_test
clintrace evaluate decision-contrast --help

After the complete validation chain is frozen, export the final release artifact.

Configuration

Configs are grouped by component:

configs/compressor/              compressor training and local smoke configs
configs/decision/default.yaml     development defaults
configs/decision/stage2_*.yaml    recorded evidence-conditioned experiments
configs/decision/stage3_*.yaml    recorded history-conditioned experiments
configs/decision/stage4_*.yaml    recorded conditional-outcome experiments

Copy recorded configurations into local_*.yaml files before adapting them to a new corpus. They document the frozen research run rather than a portable dataset configuration.

Optional LLM Client Utilities

clintrace llmc commands are wrappers around OpenAI-compatible chat endpoints. They are client-side data-preparation utilities, not ClinTrace build steps. The three frozen Chinese prompts, the zh_cn profile, and the EPXO label vocabulary are bundled inside the installed package under iatro.clintrace.assets; --profile, --prompt, and --label-vocab are optional overrides for controlled reruns. The default profile is zh_cn; non-Chinese deployments should add a separate profile asset rather than editing core code.

clintrace llmc compress    profile compressor prompt -> compressed-note JSONL
clintrace llmc expo        profile EPXO prompt + label_vocab.json -> EPXO chain JSONL
clintrace llmc history     profile history-state prompt -> admission-history-state JSONL

llmc compress reads a source IAC, applies the versioned compression prompt, and appends compressed-note rows to a JSONL file that can be consumed directly by clintrace build paired-iac --compressed-input.

The output JSONL contains at least:

doc_id
patient_id
institution
doc_type
doc_timestamp
episode_index
compressed_text
usage

Default execution targets a local OpenAI-compatible server so clinical text stays on the host.

export OPENAI_API_BASE=http://127.0.0.1:8000/v1
export OPENAI_API_KEY=local
export OPENAI_MODEL="your-local-model-id"

clintrace llmc compress \
  --institution demo_train \
  --source-iac data/00_source/demo_train.iac \
  --out data/01_compressed/demo_train.jsonl \
  --model "$OPENAI_MODEL" \
  --workers 16 \
  --thinking disabled

Useful preparation checks:

clintrace llmc compress \
  --institution demo_train \
  --source-iac data/00_source/demo_train.iac \
  --out data/01_compressed/demo_train.jsonl \
  --dry-run

clintrace build paired-iac \
  --source-iac data/00_source/demo_train.iac \
  --compressed-input data/01_compressed/demo_train.jsonl \
  --output data/02_paired/demo_train.iac \
  --institution demo_train \
  --overwrite

Rows are resumable by doc_id: rerunning the wrapper skips documents that already have a non-empty compressed_text in the output file. Non-local API endpoints require --allow-external-endpoint; only use that with an approved clinical-data processing endpoint.

EPXO and history-state preparation use the same endpoint policy:

clintrace llmc expo \
  --input data/01_compressed/demo_train.jsonl \
  --out data/05_decision_chains/results.jsonl \
  --model "$OPENAI_MODEL" \
  --workers 16

clintrace llmc history \
  --institution demo_train \
  --compressed-input data/01_compressed/demo_train.jsonl \
  --timeline-dir data/03_timeline_index \
  --output data/08_patient_state_history/demo_train.jsonl \
  --model "$OPENAI_MODEL" \
  --workers 16