Skip to content

t27f: reel_factory.t27 -- content features to a unique scene plan per item (Closes #10561) - #10566

Merged
gHashTag merged 2 commits into
masterfrom
claude/t27f-content-factory
Oct 18, 2026
Merged

gHashTag merged 2 commits into
masterfrom
claude/t27f-content-factory

Conversation

@gHashTag

Copy link
Copy Markdown
Owner

Closes #10561. Part of epic #10560 (R1 + R2).

What

specs/t27f/reel_factory.t27 is the content factory. An item's own content goes in, and a scene plan comes out. Every reel of a course gets a different scene, and no reel is edited by hand.

  • Features. Each lesson's code in reel_topics.t27 is read keyword by keyword, at word boundaries, into eleven counts: test, fn, data, trit, loop, branch, use, guard, assert, name and pub. Each mention in the lesson's English title adds TITLE_WEIGHT (10). Without the title, lesson 8 "Expressions", with 8 asserts, would win the tests scene over lesson 9 "A test block", which has 1.
  • Score. An item's count over one plus the course total (term frequency over course frequency).
  • Choose. The whole course is planned at once. The best (item, construct scene) pair is assigned first, never past ARCH_CAP (4) and never within WINDOW (2) items of the same scene. Every item left over then takes the least-used generic scene: its code, title, task or place on the course map.
  • Dress. The accent is the course module, the camera motion is t mod 4 and the layout is t mod 3.
  • Formats. Tall and wide reels, a square post, a 4:5 post, a cover, a carousel (3 to 10 slides), an X card and an og image.
  • Anomaly gate. The SSIM ceiling (850 permille) that R3 will apply to the rendered stills.

The plan for course 0 (lab: t27c gen + zig run of a scratch main; the spec itself keeps no main, see below)

There are 12 archetypes and every one is used. Each lesson that teaches a construct gets that construct's scene:

lessons scenes
7 Trits trit
10 Many tests tests
13 Choices branch
14 Loops loop
16 Structs data
18 Using other modules links
19 Invariants guard
20-26 title, task, map, tests, code, graph, tests

Before this, every lesson had the same token_floor backdrop.

Self-critique, recorded in the spec's NOT CLAIMED block

  • The dressing alone already separates the signatures of a 27-lesson course, because motion and layout fix t mod 12 and the accent splits t from t+12. The uniqueness test is weak by itself; the window test is what keeps the scenes apart.
  • TITLE_WEIGHT and the SSIM bound are chosen values, not measured ones.
  • The renderers (R3) and the change to lesson_reel (R4) are not in this PR.

Checks (Railway lab)

  • t27c test-report specs/t27f/reel_factory.t27: 8/8 pass, 0 vacuous. The per-test runtime asserts are 107, 14, 7, 456, 13, 81, 54 and 3.
  • t27c seal --save and then --verify: all hashes MATCH. The closure is reel_topics.t27.
  • t27b ledger row pass. On the x86 lab, t27b corpus specs/t27f --list takes the spec through the front end and stops only at "the JIT runs only on arm64". The JIT verdict is the job of CI's t27b-native-ratchet.
  • A first version had a printing main, and t27b rejected it on std.debug.print. The ledger's not-pass cap only moves down, so main was dropped. Printing the plan becomes a tri command (R6 of the epic).

The pointer row in specs/t27f/frontend.t27 waits for #10551, where that file is introduced.

🤖 Generated with Claude Code

gHashTag and others added 2 commits October 19, 2026 00:32
… item (Closes #10561)

The content factory of epic #10560 (R1 + R2). Each lesson's code and
title are read into eleven feature counts; twelve scene archetypes are
scored by term frequency over course frequency; the course is planned
at once (best pair first, cap 4, no repeat within 2 items, then the
least-used generic scene); accent, motion and layout dress the plan;
eight output formats and the SSIM anomaly bound for R3.

Lab: test-report 8/8 pass, 0 vacuous; seal --verify MATCH; t27b front
end accepts it (JIT verdict is the arm64 ratchet's).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-18 18:29:16 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 50
PRs with All Checks Green 0
READY 0
FAILING 50
PENDING 0
NO CHECKS YET 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=fad12492937a != 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).

@gHashTag
gHashTag enabled auto-merge (squash) October 18, 2026 18:30
@gHashTag
gHashTag merged commit 480e8dc into master Oct 18, 2026
33 of 35 checks passed
gHashTag added a commit that referenced this pull request Oct 18, 2026
Ledger: keep reel_factory_plan row. Skill: keep the branch's fuller content-factory row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Oct 18, 2026
Ledger: keep the reel_beats row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.

t27f: reel_factory.t27 -- content features to a unique scene plan per item

1 participant