docs: multi-option discernment — Pareto frontier extension (README +1 link only)#9
Merged
Merged
Conversation
Extends the bench's three objectives (effectiveness · harm · hassle) from grading a SINGLE action (proceed/ask/verify, harm-vs-hassle matrix) to SELECTING among competing alternatives, using the Pareto frontier: dominated = stupid, frontier = non-stupid, discernment chooses among the non-stupid. Golden label = the frontier + rationale, not one answer. Medical amendments (skeptical): uncertainty-aware dominance, option value / sequencing, graded feasibility, a temporal immediate-vs-durable effectiveness split (GLP-1 regain), and a "keep the vector, never collapse to a scalar utility" rule. Worked sketch included. README: one line added to the Repository map only — no existing content changed.
There was a problem hiding this comment.
Pull request overview
Adds a new research note that extends τ-discernment’s existing three-objective framing (effectiveness · harm · hassle) from evaluating a single action to selecting among multiple alternatives using a Pareto-frontier lens, and surfaces it via a single new link in the README’s repository map.
Changes:
- Added a new markdown note describing multi-option discernment via Pareto frontier selection, including a pipeline sketch and worked example.
- Added one README “Repository map” bullet linking to the new note (no other README prose changes).
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| README.md | Adds a single repository-map bullet linking to the new Pareto-frontier extension note. |
| docs/multi-option-discernment-pareto-frontier.md | New research note documenting the multi-option / Pareto-frontier extension and how it relates back to belief tracking. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a research note and one link to the Repository map. No existing README prose changed (verified: the README diff is a single added bullet).
New doc
docs/multi-option-discernment-pareto-frontier.md— extends the bench's three objectives (effectiveness · harm · hassle) from grading a single action (proceed/ask/verify, the harm-vs-hassle matrix) to selecting among competing alternatives, via the Pareto frontier:Golden label for a selection = the frontier + rationale, not one answer (richer supervision; localizes disagreement to the genuinely contestable trade-offs).
Skeptical medical amendments (so naive pruning doesn't over-simplify): uncertainty-aware dominance · option value / sequencing (the reversible first move you'd otherwise prune is the n-of-1 signal) · graded feasibility · an immediate-vs-durable effectiveness split (GLP-1 regain) · and a "keep the vector, never collapse to a weighted scalar" rule. Includes a worked "durable weight loss" sketch and ties back to belief tracking as the enabling layer.
README change
+1 bullet under Repository map only. Opened as a PR (not pushed) so you approve the exact README line before it's live.