Skip to content

HermitCrab: 86 of 1,645 Mbugwe words hit the 2M step cap and carry about two-thirds of its work; causes not examined #606

Open

Description

@johnml1135

Claude-drafted.

86 of 1,645 Mbugwe words reach the 2M step cap, and they carry about two-thirds of that grammar's total work. They have not been examined one by one, so it is not known whether they share a cause.

Status: open research lead, not started.

What we know (HC-Rust port, --step-cap 2000000, work steps):

  • The count and the work share above are the only measurements. Individual causes are unknown.
  • Mbugwe's template battery is 61% of its parser work before change C; the 86 capped words are about two-thirds of the total.
  • PanGloss#16 found six other Mbugwe words that hit a 1M work cap after an alpha-variable environment fix. That is a different cap and word set, and its overlap with the 86 is not known.

Not known: whether the 86 words share a cause (template count, rule count, root count, or something else), and whether the same words are capped on hc.dll.

Proposed change: list the 86 words, record steps per word, and check for a shared pattern (template count, rule count, root count). Report in work steps. Do not raise the cap until the cause is known.

Related: #512 (cull at emission, open), #600 (open). PanGloss: sillsdev/PanGloss#108 (pointer, open); sillsdev/PanGloss#16 (closed; six words at the 1M cap).

Next: produce the per-word step table for the 86 words on the current HC-Rust build.


Evidence -- what the record states
  • Open lead 3 gives 86 of 1,645 words at the 2M step cap and about two-thirds of Mbugwe's work (L336-L337).
  • The battery-share table says the same for the 86 words under Mbugwe's 61% (L195).
  • The C2 Mbugwe run used the same step cap and compared 1,364 words (a partial run, L215 and L220); the record does not say how the 1,645 total relates to it.
  • PanGloss#16 (closed) records two words completing at a 10M cap with 2,290,810 work steps each, using an earlier eager-candidate build; that is not a result for the final iterator at 1M.
Sources -- pinned to 2eba2f5 (closed PR #490 record)

Activity

  1. changed the title [-]HermitCrab: examine the Mbugwe words that hit the 2M step cap[/-] [+]HermitCrab: 86 of 1,645 Mbugwe words hit the 2M step cap and carry about two-thirds of its work; causes not examined[/+] on Oct 31, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    • Status
      🆕 New

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions