Kenny Ponders肯玩 AI
← Projects
01 / CaptureCodex + Claude pluginLocal-first

Tracework

Replay why the work took this path.

Git can show changed files. It cannot, by itself, explain the decision, rejected path, or risk boundary that made the work take this shape.

Decision Replay Loopthe report is a view over evidence
sessiongoal · choice · risk · next step
signalslite · standard · deep
raw recordentries · artifacts · decisions
report surfacesDaily · Weekly · Monthly · Roadmap
whyonly when the evidence can carry it

Reports are views over the raw record. Query can replay what local evidence supports; it cannot manufacture history.

Git can tell you what changed. It cannot tell you why the path was chosen.

The missing layer is not another activity stream. It is the decision context around the change: the alternatives that were rejected, the evidence that mattered, the risk boundary that stayed open, and the next step that made the work continue.

Tracework keeps that context close to the session, then lets reports and later queries point back to the raw record instead of smoothing the history into a confident summary.

One record, different moments of return.

The surfaces are views into the same evidence layer, not separate products competing to become the source of truth.

01

capture

session goal, decision, rejected path, risk, artifact, next step

durable raw entry

02

recall

recent decisions, risks, open questions, relevant artifacts

usable session context

03

query

why, alternatives, source refs, answerability

cited answer or explicit evidence gap

04

daily / weekly / monthly

outcomes, work streams, decisions, evidence audit

workplace-facing report

05

roadmap

decision threads, accumulating risks, recurring questions

long-cycle direction with its edges intact

A report earns trust by showing where it stops.

Tracework treats evidence boundaries as product behavior, not a disclaimer added after the answer.

  1. 01Raw first

    Raw entries are the semantic source. Git is only a limited fallback for coverage gaps.

  2. 02Scope before selection

    Reporting groups are separated before headlines are selected, so work and personal context do not blur together.

  3. 03Evidence boundary

    Source refs show provenance; they do not automatically become independent verification.

  4. 04Refusal is a feature

    When the local record cannot support the question, query returns an evidence gap instead of inventing history.

  5. 05Coverage is not outcome

    Activity counts describe coverage metadata. They do not prove that the work produced a result.

The boundary is part of the promise.

Tracework is useful because it keeps the memory inspectable without turning work into surveillance.

It does
  • Keep work choices, risks, artifacts, and next steps close to their evidence.
  • Let daily, weekly, and monthly reports return to the raw record.
  • Give the next session bounded context for continuing the work.
It does not
  • Package activity records as performance outcomes.
  • Use one thin source to support a claim across an entire field.
  • Turn local work memory into an approval or monitoring system.

Keep the why close to the work.

Read the repository for the contracts and workflows, then use the documentation when you are ready to make the loop part of a real project.