Merge pull request 'Move items 89, 90 and 93 into 0.1.0' (#261) from router-rest-into-0.1.0 into main

This commit was merged in pull request #261.
This commit is contained in:
2026-10-14 22:22:06 +02:00
+17 -17
View File
@@ -26,17 +26,17 @@ Release 0.2.0 takes items 125, 114 and 124 first, whatever their scores: Shape i
| ID | Release | Exempt | Item | R | S | C | A | G | Goals | Score |
|---|---|---|---|---|---|---|---|---|---|---|
| 93 | 0.1.0 | | **Let a chunk's worker start below the project's model, once the archive shows chunks on a lower rung took no more review rounds than chunks on the project's model.** | 3 | 2 | 6 | 6 | 7 | 1.1, 1.4 | 22.4 |
| 94 | 0.1.0 | | **Every 20th cycle, have loop-claude measure token spend and routing outcomes, file them here, and interrupt the human only when they call for a routing change.** | 3 | 5 | 6 | 6 | 7 | 1.1, 1.4 | 21.6 |
| 88 | 0.1.0 | | **Define one routing ladder of (model, effort) rungs ordered by a published ranking of their competence, and map a task's R + C to a rung relative to the project's model.** | 3 | 4 | 5 | 6 | 7 | 1.4 | 21.4 |
| 90 | 0.1.0 | | **Route each reviewer by the chunk's R + C, once a replay of past chunks shows the routed reviewer finds the 🔴 and 🟡 findings on record.** | 4 | 5 | 8 | 6 | 7 | 1.4 | 20.2 |
| 91 | 0.1.0 | | **Start each chunk's worker on the rung its item's R + C picks, never below the project's model, and step it one rung up after two unfinished verdicts on one chunk.** | 4 | 5 | 6 | 6 | 7 | 1.4 | 20.2 |
| 92 | 0.1.0 | | **Have the worker route each subagent by its task's R + C, and delegate a task only when the worker needs just its conclusion.** | 2 | 2 | 5 | 5 | 6 | 1.4 | 19.0 |
| 93 | 0.2.0 | | **Let a chunk's worker start below the project's model, once the archive shows chunks on a lower rung took no more review rounds than chunks on the project's model.** | 3 | 2 | 6 | 6 | 7 | 1.1, 1.4 | 22.4 |
| 90 | 0.2.0 | | **Route each reviewer by the chunk's R + C, once a replay of past chunks shows the routed reviewer finds the 🔴 and 🟡 findings on record.** | 4 | 5 | 8 | 6 | 7 | 1.4 | 20.2 |
| 89 | 0.1.0 | | **Run the judge on the rung its chunk's R + C picks, once a replay of archived transcripts shows it returns the verdicts on record.** | 3 | 4 | 7 | 5 | 6 | 1.4 | 17.5 |
| 105 | 0.2.0 | decision | **Compare review rounds per pass and reading cost for the five chunks before and after the `worker:` decision moved the principles out of the worker's context, and file an item if either rose.** | 1 | 2 | 3 | 5 | 6 | 1.4 | 20.0 |
| 112 | 0.2.0 | | **Count an item as shipped only where its chunk built it, so an item a chunk dropped from the plan gets no outcome.** | 2 | 3 | 5 | 5 | 6 | 1.1 | 19.1 |
| 110 | 0.2.0 | | **Count, for each finished chunk, the later changes that fixed code it wrote, and show them beside its guessed R.** | 2 | 4 | 6 | 5 | 6 | 1.1, 1.4 | 18.9 |
| 120 | 0.2.0 | | **Rewrite larv-review §3's "Round cap" and "Re-entry" as one table that says, for each kind of fix, which passes rerun and what a round reviews.** | 2 | 4 | 5 | 5 | 6 | 1.1, 1.2 | 18.9 |
| 89 | 0.2.0 | | **Run the judge on the rung its chunk's R + C picks, once a replay of archived transcripts shows it returns the verdicts on record.** | 3 | 4 | 7 | 5 | 6 | 1.4 | 17.5 |
| 115 | 0.2.0 | question | **Confirm or set how many finished chunks that each shipped one item `score-outcomes` waits for before it shows a score's medians, today 10.** | 1 | 1 | 2 | 4 | 5 | 1.2 | 16.6 |
| 116 | 0.2.0 | | **Report R by the share of chunks with a 🔴, once 10 chunks that each shipped one item record finding counts and most record no 🔴.** | 1 | 2 | 2 | 4 | 5 | 1.1 | 16.4 |
| 121 | 0.2.0 | | **Check that `bin/round_titles.py`'s `PASSES` lists exactly the passes of larv-review §3's table.** | 1 | 3 | 2 | 4 | 5 | 1.1, 1.3 | 16.1 |
@@ -87,6 +87,12 @@ Release 0.2.0 takes items 125, 114 and 124 first, whatever their scores: Shape i
## Details
### 93. Let a chunk's worker start below the project's model, once the archive shows chunks on a lower rung took no more review rounds than chunks on the project's model.
Goal 1.1 asks for proven quality, so stepping down waits on the archive's evidence.
Needs 91.
### 94. Every 20th cycle, have loop-claude measure token spend and routing outcomes, file them here, and interrupt the human only when they call for a routing change.
Count the cycles across all projects in loop-claude's archive, which holds every project's transcripts on the machine. Measure in code, as `bin/reading-cost` does, so the check itself costs no model run (goal 1.4): tokens by project, subagent type, rung and tool, and per routed chunk its review rounds and unfinished verdicts. A loop on another project cannot commit here, so it writes the figures to the archive, and this repo's next chunk files or refreshes one plan item from them. The ladder's thresholds and the delegation rule are decisions, so a change to them is the human's. Interrupt when chunks on a rung took more review rounds or unfinished verdicts than chunks on the rung above, or when subagent reports fill more of the worker's peak context than the searches they replaced would have; never twice for the same figures (goal 1.2).
@@ -97,6 +103,12 @@ Needs 88, 91, 92.
One ladder, the same in every project (goal 1.3), read by loop-claude, larv-review and the worker. A rung is a model and an effort; the rungs run from Haiku at low effort to the strongest model at max effort. Order the rungs by a published source, cited in the decision recording the ladder; where no source orders two rungs, such as Sonnet at xhigh against Opus at medium, ask the human. R + C moves a task up or down from the rung holding the project's model at its default effort: a low sum steps down, a high sum steps up, the thresholds set in the decision. S never moves a rung: a huge task with low R and C runs cheap. Leave out the comprehension panel and the security pass, which stay on the project model. Record each route in loop-claude's archive, so the outcomes can tune the thresholds.
### 90. Route each reviewer by the chunk's R + C, once a replay of past chunks shows the routed reviewer finds the 🔴 and 🟡 findings on record.
larv-review passes the rung through the Agent tool's `model` and `effort`. A reviewer never steps down when the worker has stepped down on the same chunk, so a cheap worker always meets a reviewer on the project model or above.
Needs 88.
### 91. Start each chunk's worker on the rung its item's R + C picks, never below the project's model, and step it one rung up after two unfinished verdicts on one chunk.
loop-claude reads the top item of the next release with `rank_todo.py` before starting a chunk's first cycle. Change the rung only when a chunk starts or steps up: every cycle resumes the chunk's session, and a model switch discards its prompt cache. When the worker bundles an item scoring a higher R + C than the one routed on, it says so in its transcript, and the judge's verdict steps the next cycle up.
@@ -109,15 +121,9 @@ State in `CLAUDE.md` when the worker delegates: a search or read that spans many
Needs 88.
### 93. Let a chunk's worker start below the project's model, once the archive shows chunks on a lower rung took no more review rounds than chunks on the project's model.
### 89. Run the judge on the rung its chunk's R + C picks, once a replay of archived transcripts shows it returns the verdicts on record.
Goal 1.1 asks for proven quality, so stepping down waits on the archive's evidence.
Needs 91.
### 90. Route each reviewer by the chunk's R + C, once a replay of past chunks shows the routed reviewer finds the 🔴 and 🟡 findings on record.
larv-review passes the rung through the Agent tool's `model` and `effort`. A reviewer never steps down when the worker has stepped down on the same chunk, so a cheap worker always meets a reviewer on the project model or above.
loop-claude starts the judge itself (`bin/loop_claude/cycle/judge.py`), so it routes without any skill change. Replay the judge on archived cycles at the candidate rung, and switch only when its interrupts and verdicts match those on record.
Needs 88.
@@ -137,12 +143,6 @@ The `worker:` decision is valid while review rounds per pass do not rise. Count
Junior A, Mid A and the maintainability senior of PR #257's scoring run named `skills/larv-review/SKILL.md:63` "Round cap" hardest, two of them with `:74` "Re-entry"; the inherited architect named `:71` "Cap". A reader holds the trigger table's "a fix, only where …" clauses, the cap, later rounds' scope and the prose, 🟢 and ❓ exceptions at once. Junior A could not say what a fix to a stability 🟡 that also touches a comment reruns. Make those sections easier to read.
### 89. Run the judge on the rung its chunk's R + C picks, once a replay of archived transcripts shows it returns the verdicts on record.
loop-claude starts the judge itself (`bin/loop_claude/cycle/judge.py`), so it routes without any skill change. Replay the judge on archived cycles at the candidate rung, and switch only when its interrupts and verdicts match those on record.
Needs 88.
### 115. Confirm or set how many finished chunks that each shipped one item `score-outcomes` waits for before it shows a score's medians, today 10.
`ENOUGH` in `bin/score_outcomes/outcomes.py` matches the 10 figures the reading-cost trend waits for. No published standard sets either, and `CLAUDE.md` → "Web research" asks the human for a threshold a consumer relies on.