jared/cc-os

Watch

Fork 0

decision-system S4c: lever fields in PredictionParser and PredictionLog, and the os-me-route CLI #1263

New issue

Closed

opened 2026-10-23 17:10:39 +00:00 by jared · 2 comments

jared commented

2026-10-23 17:10:39 +00:00

Owner

Problem

The prediction log does not record the lever labels of a prediction. PredictionParser reads only the ruling text and the Confidence: line. os-sdlc flows (S5) must get the labels of a row without a gem dependency between plugins, and no command prints them. The design is in docs/decision-system/operations.md (PredictionParser extension) and docs/decision-system/build-plan.md S4. This ticket is the third of three S4 tickets. It uses Row, AnswerSource, WrongCost::Assessor, and RequiredBand from S4a.

Approach

PredictionParser#parse reads four more lines from the prediction text. It adds their values to each ruling hash. It removes these lines from the ruling text, as it removes the Confidence: line today. A field whose line is absent is not in the hash.

Line	Pattern (as in `operations.md`)	Hash keys and types
`Source: precedent`	`/^Source:\s*(\w+)/`	`source:` String
`Cost: medium`	`/^Cost:\s*(\w+)/`	`cost:` String
`Band: 80`	`/^Band:\s*(\d+)/`	`band:` Integer
`Evidence: 55% -> 82%`	`/^Evidence:\s(\d+)%\s->\s*(\d+)%/`	`evidence_before:` Integer, `evidence_after:` Integer

PredictionLog::FIELDS gains source cost band evidence_before evidence_after, after beat_note. Old records read these fields as nil.

The command os-me-route ROW_FILE reads the whole file as the row text. It asks the model through DecisionModel::Client with the HeadlessClaude adapter and no model flag. It asks four questions in this order: the AnswerSource question, then the WrongCost::Assessor questions hard_to_undo, compounds, stays_hidden. It prints one line of JSON (JSON.generate) and exits 0. The keys are in this order:

{"source":"facts","max_rung":2,"novel":false,"wrong_cost":{"hard_to_undo":false,"compounds":true,"stays_hidden":false},"level":"medium","required_band":70}

required_band is null for the level low. O-9 is settled: os-sdlc calls this CLI; the levers stay in os-me. The ADR challenge rejected an ADR.

Required work:

Read the Source:, Cost:, Band:, and Evidence: lines in PredictionParser — check: present Evidence: in plugins/os-me/lib/os_me/prediction_parser.rb
Add source cost band evidence_before evidence_after to PredictionLog::FIELDS — check: present evidence_before in plugins/os-me/lib/os_me/prediction_log.rb
Add RowLabels, which builds the labels of one row and gives the hash for the JSON line — check: present class RowLabels in plugins/os-me/lib/os_me/row_labels.rb
Add the route ROW_FILE command to OsMe::CLI — check: present def route(row_file) in plugins/os-me/lib/os_me/cli.rb
Add the executable bin/os-me-route, which starts OsMe::CLI with route — check: present ['route', *ARGV] in plugins/os-me/bin/os-me-route

Acceptance criteria

Given the text a. Keep StepOutcome in Gate.\nConfidence: 80%\nSource: precedent\nCost: medium\nBand: 80\nEvidence: 55% -> 82%, when PredictionParser#parse runs, then it returns [{ id: "session:a", text: "Keep StepOutcome in Gate.", confidence: 80, triggers: [], source: "precedent", cost: "medium", band: 80, evidence_before: 55, evidence_after: 82 }].
Given the text Jared would accept the plan.\nConfidence: 60%, when PredictionParser#parse runs, then it returns [{ id: "session:1", text: "Jared would accept the plan.", confidence: 60, triggers: [] }].
Given the text a. Split it.\nb. Defer it.\nConfidence: 65%\nBand: 65\nEvidence: 50%->65%, when PredictionParser#parse runs, then both rulings have band: 65, evidence_before: 50, and evidence_after: 65, and their texts are Split it. and Defer it.
Given a log line {"id":"session:a","session_id":"s1","confidence":80}, when PredictionLog#records runs, then the record has "source", "cost", "band", "evidence_before", and "evidence_after", each nil.
Given a transcript whose last assistant message is the text of criterion 1, when os-me-eval append TRANSCRIPT runs, then the log record has "source" => "precedent", "cost" => "medium", "band" => 80, "evidence_before" => 55, and "evidence_after" => 82.
Given a row file with the text Move Gate::StepOutcome to OsSdlc::StepOutcome. and a fake claude that answers facts, no, yes, no in turn, when os-me-route ROW_FILE runs, then stdout is exactly the JSON line in Approach plus a newline, and the exit status is 0.
Given a row file and a fake claude that answers taste, no, no, no, when os-me-route ROW_FILE runs, then stdout is {"source":"taste","max_rung":0,"novel":true,"wrong_cost":{"hard_to_undo":false,"compounds":false,"stays_hidden":false},"level":"low","required_band":null} plus a newline.
Given a row file and a fake claude that answers facts, yes, yes, yes, when os-me-route ROW_FILE runs, then the JSON has "level":"high" and "required_band":85.
Given the row file of criterion 6, when os-me-route ROW_FILE runs, then the fake claude gets four calls, and each prompt includes Move Gate::StepOutcome to OsSdlc::StepOutcome.
Given a fake claude whose first answer is not one of the five sources, when os-me-route ROW_FILE runs, then the exit status is not 0 and stdout is empty.

Owner sketch

class OsMe::PredictionParser
  LEVER_FIELDS = {
    source: /^Source:\s*(\w+)/, cost: /^Cost:\s*(\w+)/, band: /^Band:\s*(\d+)/,
    evidence: /^Evidence:\s*(\d+)%\s*->\s*(\d+)%/
  }.freeze
  def parse(text); end
end

class OsMe::RowLabels
  def self.of(row, client:); end
  def to_h; end
end

class OsMe::CLI < Thor
  desc 'route ROW_FILE', 'print the lever labels of one row as JSON'
  def route(row_file); end
end

Non-goals

The PredictLoop text in plugins/os-me/skills/predict/SKILL.md. The orchestrator writes it through the fast path.
The Store SUGGESTIONS.md line for brief.md. The orchestrator writes it.
The stage gate (five real predictions carry all four fields). The orchestrator checks it after the skill text ships.
Bold markdown forms such as **Band:** 80. The parser reads the plain forms in the table only.
Routing in os-sdlc flows and RowRouter (S5).

Blocked by: #1246 (Row, AnswerSource, WrongCost::Assessor, RequiredBand).

Origin

Trigger: User asked for the decision-system build tickets this session. Rewritten 2026-10-23: split #1246 with exact criteria after the #1245 lesson.
Improvised this session: none
Chain: off-design S3 build ← acceptance criteria too loose ← first ticket draft. Stop at DESIGN (docs/decision-system/operations.md).
Root candidate: none found; searched WrongCost, levers, os-me, predict
Where: docs/decision-system/build-plan.md S4; docs/decision-system/operations.md PredictionParser; plugins/os-me/lib/os_me/prediction_parser.rb; plugins/os-me/lib/os_me/prediction_log.rb; plugins/os-me/lib/os_me/cli.rb; plugins/os-me/bin/os-me-route
Session: 4f3f95ea-3c6f-4d23-a0ea-e349d6ab02a5
Transcript: /home/jared/.claude/projects/-home-jared-dev-cc-os/4f3f95ea-3c6f-4d23-a0ea-e349d6ab02a5.jsonl

VERDICT: CREATE (user-approved stage split of #1246; backlog search only)

## Problem The prediction log does not record the lever labels of a prediction. `PredictionParser` reads only the ruling text and the `Confidence:` line. os-sdlc flows (S5) must get the labels of a row without a gem dependency between plugins, and no command prints them. The design is in `docs/decision-system/operations.md` (`PredictionParser` extension) and `docs/decision-system/build-plan.md` S4. This ticket is the third of three S4 tickets. It uses `Row`, `AnswerSource`, `WrongCost::Assessor`, and `RequiredBand` from S4a. ## Approach `PredictionParser#parse` reads four more lines from the prediction text. It adds their values to each ruling hash. It removes these lines from the ruling text, as it removes the `Confidence:` line today. A field whose line is absent is not in the hash. | Line | Pattern (as in `operations.md`) | Hash keys and types | |---|---|---| | `Source: precedent` | `/^Source:\s*(\w+)/` | `source:` String | | `Cost: medium` | `/^Cost:\s*(\w+)/` | `cost:` String | | `Band: 80` | `/^Band:\s*(\d+)/` | `band:` Integer | | `Evidence: 55% -> 82%` | `/^Evidence:\s*(\d+)%\s*->\s*(\d+)%/` | `evidence_before:` Integer, `evidence_after:` Integer | `PredictionLog::FIELDS` gains `source cost band evidence_before evidence_after`, after `beat_note`. Old records read these fields as `nil`. The command `os-me-route ROW_FILE` reads the whole file as the row text. It asks the model through `DecisionModel::Client` with the `HeadlessClaude` adapter and no model flag. It asks four questions in this order: the `AnswerSource` question, then the `WrongCost::Assessor` questions `hard_to_undo`, `compounds`, `stays_hidden`. It prints one line of JSON (`JSON.generate`) and exits 0. The keys are in this order: ```json {"source":"facts","max_rung":2,"novel":false,"wrong_cost":{"hard_to_undo":false,"compounds":true,"stays_hidden":false},"level":"medium","required_band":70} ``` `required_band` is `null` for the level `low`. O-9 is settled: os-sdlc calls this CLI; the levers stay in os-me. The ADR challenge rejected an ADR. Required work: 1. Read the `Source:`, `Cost:`, `Band:`, and `Evidence:` lines in `PredictionParser` — check: present `Evidence:` in `plugins/os-me/lib/os_me/prediction_parser.rb` 2. Add `source cost band evidence_before evidence_after` to `PredictionLog::FIELDS` — check: present `evidence_before` in `plugins/os-me/lib/os_me/prediction_log.rb` 3. Add `RowLabels`, which builds the labels of one row and gives the hash for the JSON line — check: present `class RowLabels` in `plugins/os-me/lib/os_me/row_labels.rb` 4. Add the `route ROW_FILE` command to `OsMe::CLI` — check: present `def route(row_file)` in `plugins/os-me/lib/os_me/cli.rb` 5. Add the executable `bin/os-me-route`, which starts `OsMe::CLI` with `route` — check: present `['route', *ARGV]` in `plugins/os-me/bin/os-me-route` ## Acceptance criteria 1. Given the text `a. Keep StepOutcome in Gate.\nConfidence: 80%\nSource: precedent\nCost: medium\nBand: 80\nEvidence: 55% -> 82%`, when `PredictionParser#parse` runs, then it returns `[{ id: "session:a", text: "Keep StepOutcome in Gate.", confidence: 80, triggers: [], source: "precedent", cost: "medium", band: 80, evidence_before: 55, evidence_after: 82 }]`. 2. Given the text `Jared would accept the plan.\nConfidence: 60%`, when `PredictionParser#parse` runs, then it returns `[{ id: "session:1", text: "Jared would accept the plan.", confidence: 60, triggers: [] }]`. 3. Given the text `a. Split it.\nb. Defer it.\nConfidence: 65%\nBand: 65\nEvidence: 50%->65%`, when `PredictionParser#parse` runs, then both rulings have `band: 65`, `evidence_before: 50`, and `evidence_after: 65`, and their texts are `Split it.` and `Defer it.` 4. Given a log line `{"id":"session:a","session_id":"s1","confidence":80}`, when `PredictionLog#records` runs, then the record has `"source"`, `"cost"`, `"band"`, `"evidence_before"`, and `"evidence_after"`, each `nil`. 5. Given a transcript whose last assistant message is the text of criterion 1, when `os-me-eval append TRANSCRIPT` runs, then the log record has `"source" => "precedent"`, `"cost" => "medium"`, `"band" => 80`, `"evidence_before" => 55`, and `"evidence_after" => 82`. 6. Given a row file with the text `Move Gate::StepOutcome to OsSdlc::StepOutcome.` and a fake `claude` that answers `facts`, `no`, `yes`, `no` in turn, when `os-me-route ROW_FILE` runs, then stdout is exactly the JSON line in Approach plus a newline, and the exit status is 0. 7. Given a row file and a fake `claude` that answers `taste`, `no`, `no`, `no`, when `os-me-route ROW_FILE` runs, then stdout is `{"source":"taste","max_rung":0,"novel":true,"wrong_cost":{"hard_to_undo":false,"compounds":false,"stays_hidden":false},"level":"low","required_band":null}` plus a newline. 8. Given a row file and a fake `claude` that answers `facts`, `yes`, `yes`, `yes`, when `os-me-route ROW_FILE` runs, then the JSON has `"level":"high"` and `"required_band":85`. 9. Given the row file of criterion 6, when `os-me-route ROW_FILE` runs, then the fake `claude` gets four calls, and each prompt includes `Move Gate::StepOutcome to OsSdlc::StepOutcome.` 10. Given a fake `claude` whose first answer is not one of the five sources, when `os-me-route ROW_FILE` runs, then the exit status is not 0 and stdout is empty. ## Owner sketch ```ruby class OsMe::PredictionParser LEVER_FIELDS = { source: /^Source:\s*(\w+)/, cost: /^Cost:\s*(\w+)/, band: /^Band:\s*(\d+)/, evidence: /^Evidence:\s*(\d+)%\s*->\s*(\d+)%/ }.freeze def parse(text); end end class OsMe::RowLabels def self.of(row, client:); end def to_h; end end class OsMe::CLI < Thor desc 'route ROW_FILE', 'print the lever labels of one row as JSON' def route(row_file); end end ``` ## Non-goals - The `PredictLoop` text in `plugins/os-me/skills/predict/SKILL.md`. The orchestrator writes it through the fast path. - The Store `SUGGESTIONS.md` line for `brief.md`. The orchestrator writes it. - The stage gate (five real predictions carry all four fields). The orchestrator checks it after the skill text ships. - Bold markdown forms such as `**Band:** 80`. The parser reads the plain forms in the table only. - Routing in os-sdlc flows and `RowRouter` (S5). Blocked by: #1246 (`Row`, `AnswerSource`, `WrongCost::Assessor`, `RequiredBand`). ## Origin - Trigger: User asked for the decision-system build tickets this session. Rewritten 2026-10-23: split #1246 with exact criteria after the #1245 lesson. - Improvised this session: none - Chain: off-design S3 build ← acceptance criteria too loose ← first ticket draft. Stop at `DESIGN` (`docs/decision-system/operations.md`). - Root candidate: none found; searched WrongCost, levers, os-me, predict - Where: docs/decision-system/build-plan.md S4; docs/decision-system/operations.md PredictionParser; plugins/os-me/lib/os_me/prediction_parser.rb; plugins/os-me/lib/os_me/prediction_log.rb; plugins/os-me/lib/os_me/cli.rb; plugins/os-me/bin/os-me-route - Session: 4f3f95ea-3c6f-4d23-a0ea-e349d6ab02a5 - Transcript: /home/jared/.claude/projects/-home-jared-dev-cc-os/4f3f95ea-3c6f-4d23-a0ea-e349d6ab02a5.jsonl VERDICT: CREATE (user-approved stage split of #1246; backlog search only)

jared added the

project/cc-os

filed-by/user

category/code-change

labels

2026-10-23 17:10:39 +00:00

jared referenced this issue

2026-10-23 17:17:51 +00:00

decision-system S4: levers and os-me-route in predict #1246

jared commented

2026-10-23 17:31:15 +00:00

Author

Owner

Work started on branch ticket-1263 (/os-sdlc:implement, S4c).

jared commented

2026-10-23 17:39:45 +00:00

Author

Owner

Resolution

Done: S4c built: Source/Cost/Band/Evidence fields in PredictionParser and PredictionLog::FIELDS, RowLabels, and the os-me-route ROW_FILE CLI that prints the labels as JSON; implementation 130 complete.

Evidence: Branch ticket-1263, Runner commit; one test-review round; suite, lint and required-work gates green; contract audit passed first time; lever_fields_test.rb and route_cli_test.rb.

Follow-ups: The predict SKILL.md PredictLoop text and the Store SUGGESTIONS.md brief line are orchestrator work for S4 (fast path), not tickets. S5 #1247 wires os-sdlc flows to os-me-route.

## Resolution **Done:** S4c built: Source/Cost/Band/Evidence fields in PredictionParser and PredictionLog::FIELDS, RowLabels, and the os-me-route ROW_FILE CLI that prints the labels as JSON; implementation 130 complete. **Evidence:** Branch ticket-1263, Runner commit; one test-review round; suite, lint and required-work gates green; contract audit passed first time; lever_fields_test.rb and route_cli_test.rb. **Follow-ups:** The predict SKILL.md PredictLoop text and the Store SUGGESTIONS.md brief line are orchestrator work for S4 (fast path), not tickets. S5 #1247 wires os-sdlc flows to os-me-route.

jared closed this issue

2026-10-23 17:39:45 +00:00

jared referenced this issue from a commit

2026-10-23 17:40:48 +00:00

feat(os-me): Audited POODR boundaries for S4c lever fields and os-me-route; placement, dependency direction, and contracts are sound; R1-R5 present. (#1263)

jared referenced this issue from a commit

2026-10-23 17:43:40 +00:00

docs(decision-system): S4 build records (#1246, #1262, #1263)