Skip to content

notebook: converge on the shared oracle/ package - #3

Open
karthikreddy-02 wants to merge 1 commit into
mainfrom
notebook-use-oracle-package
Open

notebook: converge on the shared oracle/ package#3
karthikreddy-02 wants to merge 1 commit into
mainfrom
notebook-use-oracle-package

Conversation

@karthikreddy-02

Copy link
Copy Markdown
Collaborator

Removes the notebook's embedded oracle; it now clones the repo and imports oracle/ directly. The 1 GB SQLite DB is replaced by the package's 22 MB matrix (Drive-cached). Brute-force fallback stays in-notebook, refute-only. Dataset statuses now use the package vocabulary and include the oracle's evidence string.

Verified locally against every label branch; fresh Colab run pending. Known gap: prose-wrapped outputs now land as parse-failure (follow-up PR to strip_noise).

…tp_oracle.py

Replaces the two %%writefile modules (find_equation_id.py, etp_oracle.py) with
a Colab-side clone of this repo plus oracle/ on sys.path, and replaces the 1 GB
SQLite edge-list database with oracle/'s 22 MB matrix (built via
build_matrix.py, cached to Drive as before). The brute-force finite-magma
fallback stays in-notebook, rebuilt on normalizer's AST; refute-only label
semantics unchanged (incomparable/unknown only). Dataset rows now carry the
oracle's evidence string, and statuses converge on the package vocabulary
(mapped / outside-fragment / parse-failure). Removes the stale commented-out
loop cell that still used the pre-fix weaker/stronger-from-refutation labels.

Known gap vs the embedded copy: no prose-extraction retry, so outputs like
'The equation is: x ◇ y = y ◇ x' now land as parse-failure; candidate small
PR to oracle/strip_noise.
@karthikreddy-02
karthikreddy-02 requested a review from corpaci July 28, 2026 17:27
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