Skip to content

fix(ftip): bound the finite-frontier identifiability example [AGENT] - #176

Merged
utensil merged 1 commit into
mainfrom
kepler/ftip-repair-r3-frontier-20260908
Sep 8, 2026
Merged

fix(ftip): bound the finite-frontier identifiability example [AGENT]#176
utensil merged 1 commit into
mainfrom
kepler/ftip-repair-r3-frontier-20260908

Conversation

@utensil

@utensil utensil commented Sep 8, 2026

Copy link
Copy Markdown
Owner

The finite-frontier example previously claimed that every finite set of observations admits a strictly better agreeing frontier. That fails when observations attain a known global bound or the possible class is a singleton.

Replace the universal claim with two explicit bounded monotone frontiers permitted by a declared class. The same two-intervention cost specification realizes both possibilities, with attained finite maxima at every nonnegative budget. State the known-bound and singleton exceptions, and preserve the neighboring conditional certificate's need for upper-bound evidence. Only ftip-00LN.tree changes.

Validation: all-real budget partition and boundary cases; finite analytic controls; 22 prose tests; source/lint, full site build, rendered checks, and inspected HTML/PDF. The focused PDF has 232 pages. A fresh independent adversarial review approved exact head 493e09ff7cf3a401090af9f650cecf3744e191d9 with no unresolved findings.

Replace the universal better-frontier assertion with two explicit finite attainable frontiers that agree at every measured cost. Give a deterministic bounded-utility realization, distinguish the attained global bound and singleton-class exceptions, and retain the conditional certificate and its independent upper-bound premise.
@utensil
utensil merged commit dafcb4e into main Sep 8, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant