lessons-proposed.md

back to table · edit · history · where entries came from · files · download

1760 bytes, as of the version from 2026-11-19 23:48 (current). Recorded here, not run.

## Preserve storage shape when reading back a submitted table

What happened: T767 was submitted with flat Numbers records, while the API
readback normalized its one-parameter entries into a nested Numbers mapping.
A checker assuming the submitted shape stopped before comparing values.

What the skill says now: numberdb.table returns a whole plain document and
metadata is not a replacement document; the generator and stored selections
must agree.

What it should say: the server may normalize a flat upload into nested Numbers.
Use the documented flattening helper or handle both shapes before comparing
stored identities. Do not conclude that entries disappeared because a list
assertion failed.

Evidence: stored-document.json, readback_check.py, readback-results.json: all
260 entries matched after reading the actual normalized mapping.

## Match modular forms by coefficients and initialize L-function precision

What happened: PARI's newspace at weight 3, level 24 and character chi_-24
contains both g24^(1) and g24^(2), as well as an algebraic coefficient orbit.
The sign of the q^2 coefficient distinguishes the forms in Samart's identity.
The wrong form was explicitly rejected by the arithmetic checker.

What the skill says now: independent checks must fix normalization and test
the checker with deliberately wrong controls.

What it should say: a level and character need not identify a normalized
newform. Match source Fourier coefficients before evaluating its L-function,
and initialize the modular-form/L-function objects at the working precision,
not only the final printed value.

Evidence: independent_check.py and independent-results.json; Samart Theorem
1.4; PARI 2.17.1. Checks at 140 and 180 digits agree with both stored CM rows.