Skip to content

docs(mother): DSH best-practice adoption pipeline tested against the Phase 0 pin (Closes #12952) - #12960

Merged
gHashTag merged 1 commit into
masterfrom
claude/cool-fermi-286d59
Oct 31, 2026
Merged

docs(mother): DSH best-practice adoption pipeline tested against the Phase 0 pin (Closes #12952)#12960
gHashTag merged 1 commit into
masterfrom
claude/cool-fermi-286d59

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 31, 2026

Copy link
Copy Markdown
Collaborator

Refs #12945

Closes #12952

What

The DSH best-practice adoption pipeline for T27 M, tested against the Phase 0 pin of #12948 (dsh ce7073d512 = dsh-v0.2.2-alpha.2-351, read-only sandbox, Apple M1 Pro 16 GB, 2026-10-31).

  • specs/tri/dsh_practice_adoption_plan.t27: the adoption gate and the 15 records. Four conditions (applies to the pin, keeps the Phase 0 baseline, has evidence, needs no fork change); the recorded status of every record must equal the verdict the gate computes. 10 tests, negative controls for each condition.
  • specs/tri/dsh_best_practices.md: sources, layer map L0 to L5, the summary table, 10 run receipts (command, version, machine, date), one record per practice with the saving chain, findings, what was not measured.

Result

Evaluated Adopted Rejected Deferred
15 9 2 4

Only 2 adoptions carry a measured saving: P4 resume (8 s against 24 s for the fresh run) and P9 fast gate (0.43 s per tick against 106 s for the deep rerun). The other 7 are adopted because they keep the baseline and have evidence; their saving reads "not measured".

Checks

  • t27c check: typecheck ok, 0 errors, 0 warnings.
  • t27c test-report: 10 of 10 pass, 0 vacuous passes.
  • Counting commands (documented in the md, section 11), from the repo root: grep -c 'status: S_[A]DOPTED, reason' specs/tri/dsh_practice_adoption_plan.t27 prints 9; the same with S_[R]EJECTED prints 2 and with S_[D]EFERRED prints 4.
  • English and ASCII only, no home paths, no secret, no token. Nothing under gen/. No foreign code.

Notes for the reviewer

  • No JSON receipts file. t27c gen-json prints one IGLA board event and cannot express the receipts table, so a JSON file would be hand-built outside t27. The md says so (section 10).
  • Official docs site gave no content. https://deepseek-harness.github.io/deepseek-harness/ answered HTTP 200 but the fetch returned only the page title (script-rendered), so it is recorded as unresolved as a source of content. The 14 official documents cited were read in the clone at the pin and every cited path exists there.
  • The 200-row HTTP cold start was not rerun (it needs a polling loop, which would be a non-t27 script); --dump-config timing is the boot proxy.
  • The task prompts are reworded copies of the Phase 0 tasks, so a time against Phase 0 is the same kind of task, not the same bytes.
  • Findings that bound the adoption: MCP tool calls ran with no approval event under read-only plus approval never (27 calls, 7 of them Telegram tools); the default telemetry is FEEDBACK_ONLY to a third-party collector; a providers patch replaces the map; dsh records no version, prompt SHA or tool version; the served model differed from the requested one in every run that asked for glm-5.3-flash, so no routing saving is claimed.
  • Nothing was committed to the dsh clone or to gHashTag/deepseek-harness. The running dsh and the launchd agents were not touched.

🤖 Generated with Claude Code

@gHashTag @claude
docs(mother): DSH best-practice adoption pipeline tested against the …
5aa27a6 
…Phase 0 pin (Closes #12952)

The adoption gate as a spec with tests (four conditions, fifteen records, the
recorded status equals the gate verdict) and the prose record with sources,
layer map, receipts and the reasons for each decision: 15 evaluated, 9 adopted,
2 rejected, 4 deferred. No JSON is published: gen-json cannot express the table.

Refs #12945

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Oct 31, 2026
Closed
@github-actions

github-actions Bot commented Oct 31, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-31 05:04:46 UTC

Summary

Status Count
Total Open PRs 28
PRs with Failing Checks 27
PRs with All Checks Green 1
READY 0
FAILING 27
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 27 + 0 + 0 = 27, and there are 28 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 31, 2026
Closed
Open
@github-actions github-actions Bot mentioned this pull request Oct 31, 2026
Merged
@gHashTag
gHashTag merged commit 0067777 into master Oct 31, 2026
32 of 38 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

uplift(mother): DSH documentation and best-practice adoption pipeline, tested against the Phase 0 pin (Refs #12945)

1 participant

@gHashTag