Tracking Outcomes

Recording an outcome

When something happens — a rejection, a stage advance, an interview invite, an offer — paste the signal into the agent and say what happened:

I got rejected from Thornfield Labs they want to schedule a screen I got an offer

The agent routes to track-outcomes, which is the only writer of status transitions in workspace/tracker.json. No other skill promotes a role through the pipeline. This keeps the dashboard's funnel and stage ladder accurate.

What track-outcomes does

• Classifies the incoming signal (rejection, advance, offer, withdrawal, etc.)
• Executes the status transition on the application row
• Updates the stage in the tracker and re-renders the dashboard
• Appends a durable learning entry to the role-family learning file when there is a lesson worth keeping (rejection signal, comp intel heard, what worked)
• Checks reevaluation thresholds and hands off to reevaluate-strategy if they trip

Withdrawing

If you decide to pull out of a role, say:

I withdrew from [Company] / I'm pulling out of this one

track-outcomes sets status: withdrawn, logs the withdrawal, and records any exit reason in the learning file. The role moves off the active board.

Strategy review (reevaluate-strategy)

reevaluate-strategy is triggered automatically when outcome thresholds trip. The defaults (tuneable in your config):

• ~5–8 total rejections, or 3 in one role family or fit band
• A cluster of advances or offers in one family or channel

You can also trigger it manually:

review my strategy / why am I getting filtered

The skill reads the full funnel, rejection notes, interview transcripts, and wins. It recommends concrete tuning — targeting cut/keep signals, comp anchoring, fit calibration, channel mix, writing-style — and logs the review so the dashboard's strategy-review nudge stays quiet until enough new outcomes accumulate.

Learning memory

Every role family gets a private learning file at candidate/learnings/<family>.md. Interview debriefs, rejection patterns, and win signals accumulate here. The next time the agent evaluates or tailors for that same role family, it reads the file and incorporates the lessons.

Learning files are managed by the rolester learnings helper. Skills append via the helper; it validates entries and refuses anything with placeholder residue or a private comp input.