Skip to content

v4 TASKS.md — schedule T-32: minimum never-hand-edited bootstrap seed - #3233

Merged
briansrls merged 1 commit into
mainfrom
session/still-hawk-102-t32-seed
May 17, 2026
Merged

v4 TASKS.md — schedule T-32: minimum never-hand-edited bootstrap seed#3233
briansrls merged 1 commit into
mainfrom
session/still-hawk-102-t32-seed

Conversation

@briansrls

@briansrls briansrls commented May 17, 2026

Copy link
Copy Markdown
Contributor

Summary

Schedules T-32, an operator-directed task (briansrls, 2026-05-17): reduce the bootstrap seed to the absolute minimum, in a form that never needs a hand-edit, ever.

  • Success metric — two bars, both must hold: seed size (smallest possible) and zero ongoing hand-edits (once frozen, never hand-edited for any v4 language change). A bloated never-edited seed fails; a tiny seed that needs editing every language change fails.
  • Explicitly NOT "retire v2". "Off v2" is a gameable metric — v3 proved it: v3 cleared "off v2" and was no better because its own Rust seed stayed bloated. v2-retirement is incidental, never the goal.
  • Phase 1 is design-first — an operator-reviewed definition of "minimum never-hand-edited seed" before any reduction work, resolving the open src/v3/SELF_HOSTING.md question (the frozen pinned bootstrap snapshot; the frozen language subset; how small the seed compiler for that subset can be).
  • Parallel fill, adjacent to T-15 / T-20; not on the pipeline critical path.

Planning-only TASKS.md edit on top of merged #3224/#3226/#3230.

Test plan

  • Planning-only — no code, no substrate.
  • T-32 framed by seed-size + no-hand-edit, not by "v2 removed".
  • Phase-1-design-first sequencing is explicit (no reduction work dispatched pre-operator-review).

🤖 Generated with Claude Code

@briansrls @claude
v4 TASKS.md: schedule T-32 — minimum never-hand-edited bootstrap seed
09e5f74 
Operator directive (briansrls, 2026-05-17). The goal is the bootstrap
seed reduced to absolute minimum AND never needing a hand-edit, ever —
seed-size + zero-ongoing-hand-edits, both bars. Explicitly NOT "retire
v2": "off v2" is a gameable metric (v3 cleared it and was no better,
its own seed stayed bloated). Phase 1 is a design-first, operator-
reviewed definition of "minimum never-hand-edited seed" — resolving the
open SELF_HOSTING.md question of the frozen language subset and how
small the seed compiler for it can be. Parallel fill, adjacent to
T-15/T-20, not pipeline-critical-path.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
briansrls
briansrls commented May 17, 2026
View reviewed changes

@briansrls briansrls left a comment

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review metadata

  • Provider / model: codex / unknown
  • Commit: 09e5f744 · Trigger: schedule
  • Thinking: 87s wall

Non-blocking — Strengths

  • src/v4/TASKS.md Design/docs change; T-32 preserves the Pure Bootstrap 0-floor intent by making v2 retirement a non-goal and gating seed reduction on an operator-ratified minimum/no-hand-edit definition.

✅ No blocking concerns; the added task strengthens the bootstrap success metric without diluting the 0 hand-maintained Rust target.

@briansrls
briansrls merged commit 8bd1f7b into main May 17, 2026
7 checks passed
briansrls added a commit that referenced this pull request May 17, 2026
@briansrls @claude
TASKS.md T-32: wire the plan to the bootstrap-fact-model design doc
98ae14f 
#3236 added docs/design-bootstrap-fact-model.md as "T-32's Phase-1
design content" — but T-32's TASKS.md entry (merged via #3233 before
the doc existed) did not reference it back, leaving the doc a one-way
link / effectively stray (nothing in the plan pointed to it). Add the
back-reference: T-32 Phase 1 names docs/design-bootstrap-fact-model.md
as its design content, states the Phase-1 deliverable is the operator-
ratified layer model in that doc, and that a Phase-1 worker reads it as
the brief. The doc is now consumed by the plan, not stray.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@briansrls briansrls mentioned this pull request May 17, 2026
Merged
3 tasks
briansrls added a commit that referenced this pull request May 18, 2026
@briansrls @claude
docs: bootstrap-as-complete-fact-model design (T-32 Phase 1) (#3236)
c2973ce 
* docs: design — the bootstrap as a complete fact model (T-32 Phase 1)

Operator-directed (briansrls, 2026-05-17): work out the bootstrap as a
fully modeled pipeline — every layer (compiler, seed, target, runtime,
machine) a declared fact model, every artifact a projection, zero
ambient hand-maintained code. Captures the design worked out in the
T-32 design conversation: the layer stack, the seed as
emit(snapshot, target, runtime), the comprehension boundary modeled as
a frozen sub-model + footprint fold + subset-Witness, the bootstrap
circularity modeled as a fixed point with a path-independent
self-validating witness, and the two honest floors (the permanent
physical axiom; the origin axiom that dissolves under path-
independence). This is the Phase-1 design content for TASKS.md T-32.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* docs: fix path anchors + de-overclaim bootstrap.dag state (#3236 review)

cursor APPROVE_WITH_COMMENTS on #3236 — two P1 (modeling-faithfulness)
notes: (1) the layer-table paths were repo-root-unresolvable (bare
`extdeps/...`, wrong `compiler/workflow/` segment) — all .dag paths now
src/v4/-prefixed and correct (bootstrap.dag is src/v4/workflow/); (2) §5
said bootstrap.dag "already carries this as data" — on main it is a
T-20 scaffold naming the chain in its header; the staged-stage records
+ FixptStage1Stage2 + bit_identical_check as actual workflow data are
the in-flight T-20 expansion (#3213). Reworded to distinguish "this
Phase-1 model specifies the shape" from "carried as .dag data on main".

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* TASKS.md T-32: wire the plan to the bootstrap-fact-model design doc

#3236 added docs/design-bootstrap-fact-model.md as "T-32's Phase-1
design content" — but T-32's TASKS.md entry (merged via #3233 before
the doc existed) did not reference it back, leaving the doc a one-way
link / effectively stray (nothing in the plan pointed to it). Add the
back-reference: T-32 Phase 1 names docs/design-bootstrap-fact-model.md
as its design content, states the Phase-1 deliverable is the operator-
ratified layer model in that doc, and that a Phase-1 worker reads it as
the brief. The doc is now consumed by the plan, not stray.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* design-bootstrap-fact-model: fix the trusting-trust + LanguageModel-authority claims (#3236 review)

codex BLOCKING ×2, both valid:

1. The fixed-point witness was doing double duty — reproduction AND
   trust-erasure. Corrected §5/§6: the bit-identical witness proves
   the fixed point is REACHED (reproduction), NOT that the seed was
   honest — a compromised seed (Thompson trusting-trust) reaches a
   compromised, self-consistent fixed point that passes the witness.
   Path-independence covers only benign entry paths. The "origin axiom
   dissolves" claim was wrong: it is a real second floor — the seed-
   honesty axiom — discharged by diverse double-compilation (Wheeler),
   not dissolved. End state = TWO floors, not one.

2. The doc treated the `LanguageModel` carrier as an existing authority;
   it is not declared (Theme-A #9). §2/§4 now mark it as open substrate
   T-32 Phase 1 must land (declare the carrier, or fix "a model IS a
   Node"), not an authority to point at.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* v4 TASKS: align the T-32 summary with the design doc's §5/§6 (codex/cursor #3236)

The T-32 blurb pointing to docs/design-bootstrap-fact-model.md still carried
the pre-trusting-trust-fix framing — "a path-independent witness" and "the one
permanent physical axiom" — which is a stronger claim than the corrected design
doc makes. §5 qualifies path-independence to benign entry paths and scopes the
bit-identical witness to reproduction (not honesty); §6 names two floors. A
reader stopping at TASKS.md got the skewed summary. Tightened to match.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* WIP: Sat May 16

* design-bootstrap-fact-model: honest compiler-pipeline status + footprint over full projection closure

Two BLOCKING review findings (#3236):

- §2 layer table overstated the v4 compiler pipeline as flatly
  "modeled". Only the front-end (01_tokenize / 02_parse) is modeled
  (CP-1a #3214); 00_compile / 03_normalize / 03_resolve / 04_infer /
  05_emit / 05_eval are T-8/T-9/T-10/T-22 scaffolds. Status now states
  this honestly under P1.

- §4 defined footprint as a fold over the snapshot alone, but §3
  defines the seed as emit(snapshot, target_model, runtime_model). The
  gate's seed_capability == footprint then undercounts the
  lowering/runtime constructs the seed must consume. footprint now folds
  over the full projection-input closure {snapshot, target_model,
  runtime_model}, matching seed_capability's "lowering the seed
  performs" component.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* design-bootstrap-fact-model: honest layer-model status — only rust.dag is modeled

BLOCKING review finding (#3236, docs/design-bootstrap-fact-model.md:31):
"the layer models needed already exist individually" overstated live
authority under INVARIANTS P1 (Documentation Describes Live State).
Verified on origin/main: rust.dag is modeled (2097 lines, T-4 rust-slice);
machine_code.dag (T-4.13), process.dag and file_system.dag (T-4.5) are
module-declaration-only scaffolds.

Fix: §2 status cells for target-language and runtime now distinguish the
modeled rust.dag from the three scaffolds; the "already exist" sentence
now states only rust.dag is modeled and names the per-file modeling as a
T-32 precondition.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* design-bootstrap-fact-model: cross-reference Practice 10 (now on main)

PR #3241 merged Practice 10 ("Don't hand-roll a derived operation") into
modeling-discipline.md. §1's principle — the seed is a projection, never
hand-code — is exactly Practice 10 at bootstrap scale: the seed is
produced by translation + emission (derived-operations registry rows 3
and 6), so a hand-written seed is the bootstrap-scale dissolution
finding. Added a one-paragraph cross-reference so the design doc leans on
the landed keystone Practice rather than restating the principle in
isolation.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* WIP: Sat May 16

* design-bootstrap-fact-model: §7 Phase-1 deliverable names both honest floors

BLOCKING review finding (#3236:238): §6 establishes seed-honesty as a
real second floor discharged by diverse double-compilation, but §7's
Phase-1 deliverable enumeration named only "the named physical axiom" —
so the trust-discharge witness could fall out of T-32 (INVARIANTS P1/P2).

Fix: §7 now lists "both honest floors of §6 — the physical axiom and the
seed-honesty axiom with its diverse-double-compilation discharge
witness" as load-bearing T-32 Phase-1 deliverables, consistent with §6.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@briansrls
briansrls deleted the session/still-hawk-102-t32-seed branch June 1, 2026 18:43
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.

1 participant

@briansrls