transaction-trace-analysis

by benchflow-ai

Install this Skill

curl -fsSL -o transaction-trace-analysis.zip https://skill.seembox.com/api/v1/skills/transaction-trace-analysis/download && unzip transaction-trace-analysis.zip

Download transaction-trace-analysis.zip

SKILL.md About

---
name: transaction-trace-analysis
description: Teaches evidence-based analysis of transaction logs and traces. Use when reconstructing transaction behavior from tables, TSV/CSV files, timestamps, counters, object histories, validation records, abort records, retries, partial logs, or when producing classified transaction outputs from trace evidence.
---

# Transaction Trace Analysis

Use this skill when the task provides concrete logs, tables, counters, timestamps, or partial records and asks you to infer transaction behavior.

This skill is about *reconstruction under partial evidence*:

- normalize columns into a stable event vocabulary
- reconstruct per-transaction intent and per-object version timelines
- join them through explicit fields (not through row order guesses)
- produce a minimal, auditable classification for the required output

## Quick Reference (pick the next file to read)

| If you need… | Read |
|---|---|
| Mapping weird columns into tx/key/op/timestamp/counter/status/reason | [trace-schema-normalization.md](references/trace-schema-normalization.md) |
| Grouping events per transaction and avoiding “attempt vs request” mixups | [transaction-history-reconstruction.md](references/transaction-history-reconstruction.md) |
| Reconstructing a per-key version timeline from writes/validation | [object-version-history.md](references/object-version-history.md) |
| Interpreting timestamp/counter semantics and scope correctly | [timestamp-and-ordering-fields.md](references/timestamp-and-ordering-fields.md) |
| Understanding validation failures, abort reasons, and conservative checks | [validation-and-abort-records.md](references/validation-and-abort-records.md) |
| Avoiding overclaims when the trace is incomplete | [evidence-quality-and-uncertainty.md](references/evidence-quality-and-uncertainty.md) |
| Emitting deduped/sorted ids and minimal justifications | [output-classification-patterns.md](references/output-classification-patterns.md) |

## Trace Ledger (what you build first)

Build two tiny ledgers before doing any deep inference:

### Transaction ledger

```text
txn_id -> {
  seen_keys: [...],
  read-evidence: (key, observed_version_fields...),
  write-evidence: (key, installed_version_fields...),
  validation/abort-evidence: [...],
  candidate_commit_fields: ...,
}
```

### Object ledger

```text
key -> timeline [
  { event: write/abort/validate, txn_id, version_fields, counter_fields },
  ...
]
```

Only after both exist should you attempt joins like “did T’s read become stale?” or “could an aborted T have committed?”

## Workflow (operational)

1. **Schema normalization**: name every column; identify scope for each counter/timestamp.
2. **Per-txn reconstruction**: summarize each txn’s evidence and candidate serialization point fields.
3. **Per-key reconstruction**: build a timeline per key; validate what orders it (wts? counter?).
4. **Decision modeling**: re-express the protocol’s decision rule in terms of trace fields.
5. **Classification**: prove “must abort” vs “could commit” vs “unknown from evidence.”
6. **Output**: dedupe ids, sort if required, write exact format.

## Common Pitfalls

- **Treating per-key counters as global order.**
- **Using file row order as time order without justification.**
- **Calling something ‘unnecessary’ without exhibiting a safe order consistent with evidence.**