Source note2 min read
Line-Anchored Feedback Cuts Token Costs and Improves Correctness in AI Code Editing
Source
arXiv
arxiv.org/abs/2607.12713v1
Source
- Source name: arXiv
- Source kind: paper
- Authors / organization: William Franz Lamberti
- Venue: arXiv: cs.SE
- Published: 2026-07-14
- Collected: 2026-07-15
- URL: http://arxiv.org/abs/2607.12713v1
- PDF: https://arxiv.org/pdf/2607.12713v1
- DOI: UNKNOWN
Plain-Language Brief
A study shows that using structured, line-anchored feedback for generative AI (GAI) code editing can reduce the number of generated tokens, improve correctness, and is especially beneficial for weaker models.
What the Document Says
- The cost, latency, and energy of generative AI (GAI) code editing are driven by generated tokens.
- Two deliveries of the same requested changes were compared: holistic prompt (control) and structured, line-anchored feedback from FileMark (treatment).
- FileMark is a VSCodium extension for inline comments on any file.
- In a paired experiment, line anchoring reduced generated tokens by varying percentages for different models, with some models generating significantly fewer tokens after multiple-testing correction.
- Correctness improved for models with headroom (enough capacity) under line-anchored feedback.
- An exploratory experiment showed that the correctness benefit grows further when the edit-application burden is lifted.
- The study finds that line-anchored feedback reduces what stronger models spend and improves what weaker models get right.
Key Concepts Mentioned
- generative AI
- GAI code editing
- generated tokens
- feedback
- FileMark
- VSCodium extension
- inline comments
- cost
- latency
- energy
- holistic prompt
- line-anchoring
- Claude Opus
- Claude Sonnet
- correctness
- stronger models
- weaker models
Claims Made by the Document
- Line anchoring reduces generated tokens in GAI code editing.
- Correctness improves under line-anchored feedback for models with headroom.
- The correctness benefit grows further when the edit-application burden is lifted.
Evidence Used
- Reduction of generated tokens by varying percentages for different models
- Improvement in pooled and specific local model correctness scores
Methods / Approach
UNKNOWN
Limitations Stated by the Authors
UNKNOWN
Limitations Noted by the Agent
Agent interpretation — not from the source.
UNKNOWN
Open Questions
UNKNOWN
Notes for Future Q&A
May include interpretation. Labeled where it does.
UNKNOWN
Raw Abstract / Excerpt
Generated tokens are a direct driver of the cost, latency, and energy of generative AI (GAI) code editing. We show the format of feedback is a lever on all three. We compare two deliveries of the same requested changes: a holistic prompt (control) versus the structured, line-anchored export of FileMark (treatment). FileMark is a VSCodium extension for inline comments on any file. In a paired experiment line anchoring cut generated tokens by 22% (Claude Opus) and 58% (Claude Sonnet), reaching 24%-80% on files of 100 lines or more, with four of seven models generating significantly fewer tokens after multiple-testing correction. Correctness rose where models had headroom: +2.0 points pooled and +5 to +7 points for three of five local models. An exploratory experiment in which the harness, not the GAI model, applies function-level patches shows the correctness benefit grows further when the edit-application burden is lifted: local-model correctness on 100+ line files roughly triples under anchoring. Line-anchored feedback reduces what stronger models spend and improves what weaker models get right.
