RESEARCH

Find it, profile it, monitor it

Microstructure analysis toolkit — scan the universe, profile individual pairs, materialize features in batch, check drift on deployed bundles, and recalibrate.

bash

sequence research <command> [options]

Scan

Find pairs matching criteria across the universe.

Profile

Deep single-pair microstructure analysis.

Materialize

Batch compute profiles for a date range.

Coverage

Data availability across pairs and dates.

Feature-eval

Walk-forward predictive-power evaluation.

Monitor

Drift checks for a deployed bundle.

Recalibrate

Drift check + re-profile + optional promote.

Artifacts

Inspect or list research artifacts.

`sequence research scan`

Telescope: scan all materialized pairs for matching criteria.

bash

sequence research scan \
  --where "maker_edge_bps > 1.5" \
  --where "markout_5s_bps > 0.5" \
  --rank-by markout_5s_bps \
  --top 20 \
  --evidence

Flag	Default	Description
`--where <clause>`	-	Filter predicate (repeatable)
`--rank-by <metric>`	-	Sort results by this metric
`--top <n>`	`10`	Number of results to return
`--evidence`	off	Show detailed evidence for each hit

Returns a ranked table of pairs matching all --where predicates. Evidence includes sample size, walk-forward validation stats, and sensitivity analysis.

`sequence research profile`

Microscope: deep single-pair analysis.

bash

sequence research profile BTC-USD --venue kraken --days 7

Flag	Default	Description
`--venue <name>`	all	Filter by venue
`--days <n>`	`7`	Analysis lookback window

Output includes spread, depth, trade flow, signals (OFI, MLOFI, microprice), execution metrics (markout, toxicity, maker edge), volatility, and regime classification.

`sequence research materialize`

Batch compute profiles for all symbol-venue pairs in a date range.

bash

sequence research materialize \
  --start 2026-03-01 \
  --end 2026-03-08

Flag	Description
`--start <date>`	Start date (YYYY-MM-DD, required)
`--end <date>`	End date (YYYY-MM-DD, required)

Saves materialized profiles to the snapshot store with a unique snapshot ID and formula hash for reproducibility.

`sequence research coverage`

Show data availability.

bash

sequence research coverage
sequence research coverage --symbol BTC-USD

Flag	Description
`--symbol <name>`	Filter to a specific symbol

`sequence research feature-eval`

Evaluate feature predictive power via walk-forward IC (information coefficient).

bash

sequence research feature-eval \
  --feature microprice_ic \
  --feature ofi_1level \
  --label fwd_ret_5s \
  --start 2026-02-01 \
  --end 2026-03-01

Flag	Description
`--feature <name>`	Feature to evaluate (repeatable)
`--label <name>`	Target label (repeatable)
`--start <date>`	Start date
`--end <date>`	End date

`sequence research monitor`

Check performance drift for a deployed bundle without taking action.

bash

sequence research monitor \
  --bundle bun_e4f2a1 \
  --target-kind edge \
  --days 7

Flag	Default	Description
`--bundle <id>`	required	Bundle ID to monitor
`--target-kind <type>`	`edge`	`edge` or `sor`
`--days <n>`	`7`	Monitoring lookback window
`--store-root <path>`	-	Custom data store path

Output: drift report with per-metric status (degraded/stable/improved).

Drift Thresholds

Edge bundles:

Metric	Type	Threshold
`markout_5s_bps`	critical	min 0.0
`realized_sharpe`	critical	min 0.5
`maker_pct`	non-critical	min 0.3
`slippage_bps`	non-critical	max 5.0

SOR bundles:

Metric	Type	Threshold
`implementation_shortfall_bps`	critical	max 2.0
`fill_completeness_pct`	critical	min 0.95
`slippage_bps`	non-critical	max 5.0

A bundle is classified as "degraded" if any critical metric breaches its threshold.

`sequence research recalibrate`

Full recalibration pipeline: drift check, re-profile, quality gates, and optional promotion.

bash

sequence research recalibrate \
  --bundle bun_e4f2a1 \
  --target-kind edge \
  --days 7 \
  --auto-promote-to replay

Flag	Default	Description
`--bundle <id>`	required	Bundle ID
`--target-kind <type>`	`edge`	`edge` or `sor`
`--days <n>`	`7`	Recalibration lookback window
`--store-root <path>`	-	Custom data store path
`--auto-promote-to <stage>`	-	Auto-promote if gates pass (only `replay` allowed)
`--force`	off	Run even if drift is stable
`--baseline-eval <id>`	-	Baseline evaluation to compare against

Load recent outcomes

Reads ExecutionOutcome data for the specified date range.

Compute drift

Compares metrics against baseline or absolute thresholds.

Re-profile

Builds a new artifact with updated microstructure profiles.

Quality gates

Evaluates pass/fail on each metric gate.

Promote (optional)

If --auto-promote-to is set and gates pass, compiles a new bundle and promotes.

`sequence research artifact`

Show details of a specific artifact.

bash

sequence research artifact art_abc123

`sequence research artifacts`

List all persisted artifacts.

bash

sequence research artifacts
sequence research artifacts --type experiment

Flag	Description
`--type <type>`	Filter by type: `alpha`, `execution_model`, `routing_model`, `eval_report`, `policy`, `table`, `experiment`

Typical Workflows

Find promising pairs

bash

# 1. Materialize profiles for the last week
sequence research materialize --start 2026-03-03 --end 2026-03-10
 
# 2. Scan for pairs with strong maker edge
sequence research scan \
  --where "maker_edge_bps > 1.0" \
  --where "daily_volume_usd > 1000000" \
  --rank-by markout_5s_bps \
  --top 10 --evidence

Monitor and recalibrate a deployed bundle

bash

# Check for drift
sequence research monitor --bundle bun_e4f2a1 --days 7
 
# If degraded, recalibrate and auto-promote to replay
sequence research recalibrate \
  --bundle bun_e4f2a1 --days 7 \
  --auto-promote-to replay

Evaluate a new feature

bash

sequence research feature-eval \
  --feature microprice_ic \
  --label fwd_ret_5s \
  --start 2026-01-01 \
  --end 2026-03-01