Re-Ranking Context with Explicit Feedback, Without Letting It Take Over

●1 ●11 ●39
calendar_today ago • schedule1 min read

tokencap ask "<task>" --adaptive re-ranks TokenCap's candidate files using your explicit feedback.

The guardrails:

  • Only feedback from tasks with the same subject counts. Words like "fix" and "bug" are ignored.
  • The boost per file is capped at 2.
  • It only re-orders files retrieval already found. Nothing is added, nothing is removed.
  • Each result keeps its baseline rank, so you can see what moved.

Inspect the evidence with tokencap analyze --learning and tokencap analyze --patterns. Patterns are advisory and act on nothing.

npm run eval:adaptive exits non-zero if adaptive ranking loses recall on its controlled corpus.

Details at tokencap.vansharora.app

Want more deep dives?

Join 5,951+ developers sharing real benchmarks, code breakdowns, and projects. Free forever.

2 Comments

0 votes
0

More Posts

Tuesday Coding Tip 06 - Explicit template instantiation

Jakub Neruda - Apr 7

I’m a Senior Dev and I’ve Forgotten How to Think Without a Prompt

Karol Modelski - Mar 19

SE Ranking Review (2026): Pros, Cons, and Verdict

floxolab - Oct 22

Cisco's Amy Chang: A Model's "Passport" Doesn't Tell You Where It Actually Came From

Tom Smithverified - Aug 27

From Subjective Narratives to Objective Data: Re-engineering the Elderly Care Communication Loop

Huifer - Jan 28
chevron_left
1.7k Points • 51 Badges
Jaipur, India • t.co/BRmkFprjYq
17Posts
9Comments
15Connections
tech freak

Related Jobs

View all jobs →

Commenters (This Week)

2 comments
2 comments
1 comment

Contribute meaningful comments to climb the leaderboard and earn badges!