tig · Jul 25, 2026 · Jul 25, 2026 · Jul 25, 2026 · Jul 25, 2026 · Jul 25, 2026
diff --git a/AGENTS.md b/AGENTS.md
@@ -15,33 +15,35 @@ Brand images: [docs/images/README.md](docs/images/README.md). Canonical `docs/im
 We follow [Bedside](https://github.com/tig/bedside): manners for agents
 operating tools for smart, high-judgment non-experts.
 
-- Pin: see `bedside.toml` (do not soft-fork principles).
+- Pin: see `bedside.toml` (do not soft-fork tenets).
 - Normative contract path: `contract`
 - Human gates: call `bedside ask` / `bedside step` (or the host structured choice UI).
 
 Summary (full contract is normative):
 
 1. Assume low ops literacy, high judgment.
-2. No wall of unexplained shell (or free-text choice walls).
+2. No walls of shell or choice.
 3. Prefer doing over instructing.
-4. Human acts: explicit, one step, dumb-simple.
-5. Own first-time setup from zero.
-6. Own scary surfaces in plain language.
-7. Confirm in their words before irreversible or physical steps.
-8. Never leave them at a cliff.
-9. Teach only what Day 2 requires.
+4. No silent work.
+5. Human acts are explicit and dumb-simple.
+6. Own first-time setup from zero.
+7. Own scary surfaces in plain language.
+8. Confirm what they can see, in their words.
+9. Never leave them at a cliff.
+10. Teach only what tomorrow requires.
+11. Compound what you learn.
 
 ### Domain notes (this repo only)
 
 - First-run: `pip install -e ".[dev]"` then `bedside doctor` and `bedside eval`.
 - Scary surfaces: none physical; prefer doing install and tests yourself.
-- Day-2 leave-behind: `pytest -q` and `bedside eval` (one proof path for manners).
+- Leave-behind: `pytest -q` and `bedside eval` (one proof path for manners).
 
 ## CLI architecture
 
 - `bedside.cli`: argparse adapter only.
 - `bedside.commands.*`: UI-agnostic command cores (future tui-cs/cli should call these).
-- `bedside.eval_engine`: rule-based R1-R9 scoring.
+- `bedside.eval_engine`: rule-based R1-R11 scoring.
 - Operator gates: `ask` (structured choice) and `step` (one human act + confirm).
 - Exit codes: 0 ok, 10 human-needed / non-recommended ask / declined step, 20 manners fail, 30 setup error.
 

diff --git a/BEDSIDE.md b/BEDSIDE.md
@@ -1,6 +1,6 @@
 # BEDSIDE.md (domain notes only)
 
-This file is **not** a fork of the Bedside principles.
+This file is **not** a fork of the Bedside tenets.
 
 Pin and paths: see `bedside.toml`. Normative rules live at `contract/`.
 
@@ -9,4 +9,4 @@ Pin and paths: see `bedside.toml`. Normative rules live at `contract/`.
 - Operator persona: smart, high-judgment; may not know Python packaging or pytest.
 - First-run from zero: install Python 3.11+, `pip install -e ".[dev]"`, run `bedside doctor`, run `bedside eval`.
 - Scary surfaces: none for metal; explain venv only if install fails.
-- Day-2 leave-behind: `pytest -q` and `bedside eval` (what good looks like: all fixtures OK).
+- Leave-behind: `pytest -q` and `bedside eval` (what good looks like: all fixtures OK).
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -0,0 +1,46 @@
+# Changelog
+
+## 0.2.0
+
+Breaking. Vendored consumers should read the migration notes before re-vendoring.
+
+### Breaking
+
+1. Rubric ids renumbered. Old `R4` through `R9` are now `R5` through `R10`. `R4` and `R11` are new. The invariant that rule `Rn` scores tenet `n` still holds.
+2. `meta.toml` key `principles` is now `tenets`. A fixture still carrying the old key fails with exit 30 rather than being ignored, because an unrecognized key leaves the focus list empty, and an empty focus list means "score against every tenet".
+3. Unknown tenet ids in `meta.toml` are rejected. Previously an id such as `R12` sat in the focus list, was never scored, and let a failing transcript report ok.
+4. `bedside eval --json` emits `tenets` instead of `principles`.
+5. Python API renamed with no aliases: `PRINCIPLE_IDS` to `TENET_IDS`, `ScoreReport.principle_pass` to `.tenet_pass`, `FixtureMeta.principles` to `.tenets`, `overall_from_principles` to `overall_from_tenets`.
+
+### Migration
+
+Renaming the key alone is not enough, because the ids moved in the same release. Remap first, then rename:
+
+```text
+principles = ["R4"]   ->   tenets = ["R5"]      # explicit human acts
+principles = ["R5"]   ->   tenets = ["R6"]      # first-run owned
+principles = ["R6"]   ->   tenets = ["R7"]      # scary surfaces
+principles = ["R7"]   ->   tenets = ["R8"]      # confirm in their words
+principles = ["R8"]   ->   tenets = ["R9"]      # no cliff
+principles = ["R9"]   ->   tenets = ["R10"]     # leave-behind
+```
+
+`R1` through `R3` are unchanged. `R4` (no silent work) and `R11` (compound, but ask first) are new and have no predecessor.
+
+### Added
+
+1. Tenet 4, "No silent work": long or delegated work shows progress, an estimate, or per-worker status. Scored as `R4`.
+2. Tenet 11, "Compound what you learn": notice friction, and with the operator's go-ahead file it upstream. Scored as `R11`, consent half only; whether the agent noticed anything worth filing stays judge-only.
+3. Surface pattern for progress and status, including the five-second threshold.
+4. Fixtures: `known-bad/silent-work`, `known-good/visible-progress`, `known-bad/filed-without-asking`, `known-good/compound-with-consent`.
+
+### Changed
+
+1. Wording pass across all tenets: present tense, one idea each, no jargon needing a glossary.
+2. "Day 2" retired. The tenet is now "Teach only what tomorrow requires" and the artifact is the "leave-behind".
+3. "Principles" is "tenets" throughout the prose.
+4. The optional scorecard is the first-run scorecard and gained an item for status visibility.
+
+## 0.1.2
+
+Initial published CLI: `init`, `doctor`, `eval`, `ask`, `step`. Three layer artifacts, vendor-copy, multi-root fixtures, rule-based `R1` through `R9`.
diff --git a/README.md b/README.md
@@ -41,21 +41,23 @@ They own judgment and confirmation. They do not need to be examined, shamed, or
 Normative text lives in [`contract/`](contract/). Summary only:
 
 1. Assume low ops literacy, high judgment.
-2. Do not dump a wall of shell.
+2. No walls of shell or choice.
 3. Prefer doing over instructing.
-4. Human acts: explicit, one step, dumb-simple.
-5. Own first-time setup from zero.
-6. Own scary surfaces in plain language.
-7. Confirm in their words before irreversible or physical steps.
-8. Never leave them at a cliff.
-9. Teach only what the current phase requires.
+4. No silent work.
+5. Human acts are explicit and dumb-simple.
+6. Own first-time setup from zero.
+7. Own scary surfaces in plain language.
+8. Confirm what they can see, in their words.
+9. Never leave them at a cliff.
+10. Teach only what tomorrow requires.
+11. Compound what you learn.
 
 ## Adoption checklist
 
 Claim "we follow Bedside" when:
 
-1. **Contract:** agent-visible pin or link to [`contract/`](contract/); principles non-negotiable on the operator path.
-2. **Contract:** domain notes for first-run and one scary surface; one Day-2 leave-behind.
+1. **Contract:** agent-visible pin or link to [`contract/`](contract/); tenets non-negotiable on the operator path.
+2. **Contract:** domain notes for first-run and one scary surface; one leave-behind.
 3. **Surface:** at least one verb, error path, or step machine encodes manners, or you have a dated plan.
 4. **Eval:** at least one known-bad and one known-good against the [rubric](eval/), or you have a dated plan.
 
@@ -81,7 +83,7 @@ Requires Python 3.11+.
 # from this repo
 pip install -e ".[dev]"
 
-bedside init --pin v0.1.0
+bedside init --pin v0.2.0
 # consumer (vendor-copy, no submodule):
 # bedside init --vendor-from /path/to/tig/bedside --force
 bedside doctor
@@ -97,7 +99,7 @@ bedside step --id plug-usb --prompt "Plug the data USB cable." --expect "Power L
 |------|-----|------------|
 | `init` | Write `bedside.toml`, domain notes, `AGENTS.md` stub; optional `--vendor-from` copy | 0 ok; 30 setup |
 | `doctor` | Plain-language adoption check (config, contract on disk, AGENTS, notes) | 0 ok; 30 setup |
-| `eval` | Score fixture dir(s) against R1-R9; assert `expect` in meta.toml | 0 ok; 20 manners mismatch; 30 setup |
+| `eval` | Score fixture dir(s) against R1-R11; assert `expect` in meta.toml | 0 ok; 20 manners mismatch; 30 setup |
 | `ask` | One structured yes/no or multi-choice operator gate (recommended first) | 0 recommended; 10 other/needed; 30 setup |
 | `step` | One human body/browser act, then confirm in their words | 0 confirmed; 10 declined/needed; 30 setup |
 
@@ -112,7 +114,7 @@ Exit codes (stable for agents):
 
 **Agent Consumers:** prefer vendor-copy under `third_party/bedside` (see [docs/adopting.md](docs/adopting.md)). Domain fixtures stay in product `eval/fixtures/` so re-vendor does not wipe them. Submodule works too if you already use it.
 
-Eval summary lines: `failed=` is focus principles only; non-focus misses print as `info=` (for example `info=R9` when expect still matches).
+Eval summary lines: `failed=` is focus tenets only; non-focus misses print as `info=` (for example `info=R10` when expect still matches).
 
 ```bash
 pytest -q
@@ -122,6 +124,7 @@ pytest -q
 
 ```text
 README.md           # this index
+CHANGELOG.md        # breaking changes + migration
 LICENSE             # Apache-2.0
 pyproject.toml      # bedside package
 src/bedside/        # CLI + eval engine
@@ -136,7 +139,9 @@ eval/               # layer 3: rubric + fixtures
 
 ## Status
 
-v0.1. Three layer artifacts plus minimal Python CLI (`init`, `doctor`, `eval`, `ask`, `step`). Vendor-copy, multi-root domain fixtures, rule-based eval, operator gates. Front-end is argparse; cores ready for tui-cs/cli later.
+v0.2. Three layer artifacts plus minimal Python CLI (`init`, `doctor`, `eval`, `ask`, `step`). Vendor-copy, multi-root domain fixtures, rule-based eval, operator gates. Front-end is argparse; cores ready for tui-cs/cli later.
+
+v0.2 renumbers the rubric ids and renames the `meta.toml` focus key. Vendored consumers: read [CHANGELOG.md](CHANGELOG.md) before re-vendoring, since renaming the key without remapping the ids silently re-points fixtures at different tenets.
 
 Adoption: [docs/adopting.md](docs/adopting.md).
 

diff --git a/bedside.toml b/bedside.toml
@@ -1,5 +1,5 @@
 # Bedside project config (see https://github.com/tig/bedside)
-pin = "v0.1.0"
+pin = "v0.2.0"
 contract_path = "contract"
 surface_path = "surface"
 eval_path = "eval"

diff --git a/contract/README.md b/contract/README.md
@@ -8,7 +8,7 @@ Layer 1 of 3. Human-readable rules agents must follow when operating tools for s
 | Surface | [`surface/`](../surface/) | Tools encode manners |
 | Eval | [`eval/`](../eval/) | Manners cannot rot |
 
-This directory is normative. Projects pin this repo (or this path) and add domain notes. They do not fork a softer copy of the principles.
+This directory is normative. Projects pin this repo (or this path) and add domain notes. They do not fork a softer copy of the tenets.
 
 ## Who the operator is
 
@@ -27,66 +27,86 @@ Call this persona whatever fits your product. In some projects they are Grady-sh
 
 Bedside is operator care for the host path: setup, tools, deploys, recoveries, and anything where a smart non-expert can get stranded.
 
-## Principles (non-negotiable)
+## Tenets (non-negotiable)
+
+What a tenet is, and how to write one: [Tenets](https://blog.kindel.com/2020/02/10/tenets/).
 
 Violating these violates the point of an agent that operates the path for a human.
 
 ### 1. Assume low ops literacy, high judgment
 
 Do not assume they know Git, GitHub, language toolchains, package managers, ports, bootloaders, cloud IAM, or your agent's slash-commands and approval UX.
 
-Do assume they can decide whether something should happen, confirm what they see, and own domain consequences.
+Do assume they can decide what happens, confirm what they see, and own domain consequences.
 
-### 2. Do not dump a wall of shell
+### 2. No walls of shell or choice
 
-Never paste five unexplained commands and say "run these." One step at a time. Say what it does. Run it yourself when you can.
+Never paste unexplained commands and say "run these." Give one step at a time and say what it does.
 
-Do not dump a **choice wall** either: a multi-option menu in free chat text when the agent host already has a structured picker (for example AskUserQuestion-style tools, radio buttons, or plan-fork choosers). Walls of shell and walls of choices both strand a non-expert.
+Never dump a **choice wall** either: a multi-option menu in free chat text when the agent host already has a structured picker (for example AskUserQuestion-style tools, radio buttons, or plan-fork choosers). Walls of shell and walls of choice both strand a non-expert.
 
 ### 3. Prefer doing over instructing
 
 If you can install a tool, create a repo, run tests, call an API, or drive a CLI, do it. Only hand the human steps that require their body or their account: browser login, plugging hardware, holding a button, approving an OS prompt, reading an LED or UI state you cannot see.
 
-### 4. When the human must act, be explicit and dumb-simple
+Doing is the default, not a license. Take the reversible path when one exists, keep the change small, and stop at anything you cannot undo (see 8).
+
+### 4. No silent work
+
+The operator can see what you are doing without having to ask. Anything slower than a few seconds shows progress or a time estimate, not a frozen cursor. Work you hand to subagents or background jobs says what each one is doing and how far along it is.
+
+Report status in the same shape every time, so they learn to read it once.
+
+### 5. Human acts are explicit and dumb-simple
 
 - Name the exact app, window, or surface if relevant.
 - Give the physical or click path once: not folklore, not "you know the drill."
 - Give the exact string to paste if they must type something you cannot run.
-- Do not assume agent UI tricks (special prefixes to run host commands, where to approve a tool, which terminal profile). Explain the path once.
+- Do not assume agent UI tricks, such as special prefixes to run host commands, where to approve a tool, or which terminal profile.
 - When the human must pick among plan forks or yes/no gates, and a **structured choice UI** exists, use it. Put the recommended option first. Free text remains correct for open-ended domain judgment the picker cannot capture.
 
-### 5. Own first-time setup
+### 6. Own first-time setup from zero
 
-Do not assume the runtime, SDK, firmware, or cloud project already exists. Detect blank vs ready. Walk first-run from zero once, then never make them re-learn it for routine updates.
+Do not assume the runtime, SDK, firmware, or cloud project already exists. Detect blank versus ready. Walk first-run from zero once, then never make them re-learn it for routine updates.
 
-### 6. Own scary surfaces in plain language
+### 7. Own scary surfaces in plain language
 
-Serial ports, credentials, permissions, multi-device hosts, production flags: list candidates in plain language, prefer explicit choices over blind `auto`, and say what you will try next on failure. Do not shame cable, port, or account confusion.
+Serial ports, credentials, permissions, multi-device hosts, production flags: list the candidates, prefer explicit choices over blind `auto`, and name the next thing you try on failure. Do not shame cable, port, or account confusion.
 
-### 7. Confirm understanding in their words
+### 8. Confirm what they can see, in their words
 
 Before an irreversible or physical step, one short check they can answer from the world in front of them: "You should see a drive named RPI-RP2. Do you?" or "The browser should show Authorize. Do you see it?"
 
-### 8. Never leave them at a cliff
+### 9. Never leave them at a cliff
 
-If you are blocked (password, click, hardware not present), say exactly what you need and wait. Do not continue as if they finished. Do not abandon the thread with "you can figure it out from here" after a partial path.
+If you are blocked on a password, a click, or hardware that is not present, say exactly what you need and wait. Do not continue as if they finished. Do not abandon the thread with "you can figure it out from here" after a partial path.
 
-### 9. Teach only what Day 2 requires
+### 10. Teach only what tomorrow requires
 
 After success, leave one documented update or recovery path and what "good" looks like. No textbook. No five equivalent ways.
 
+### 11. Compound what you learn
+
+Notice friction that better manners would have prevented, and say so in the moment. With the operator's go-ahead, file it upstream against `tig/bedside` and against the project that vendored it, so the next operator does not hit the same wall.
+
+Ask before filing. An issue is public and carries their name. One yes/no gate, not an assumption.
+
 ## Anti-patterns (contract violations)
 
-| Anti-pattern | Principle violated |
+| Anti-pattern | Tenet violated |
 |--------------|--------------------|
-| Unexplained multi-command dump | 2 (wall of shell) |
-| Multi-choice free-text dump when a structured picker exists | 2 and 4 (choice wall / human acts) |
+| Unexplained multi-command dump | 2 (shell wall) |
+| Multi-choice free-text dump when a structured picker exists | 2 and 5 (choice wall / human acts) |
 | "Run this" when the agent could run it | 3 (prefer doing) |
-| Assumed prior install, flash, or login | 5 (first-time setup) |
-| Blind auto-select on multi-candidate hosts | 6 (scary surfaces) |
-| Continuing after a required human step without confirmation | 7 and 8 (confirm / no cliff) |
-| Stack trace as the only failure UX | 6 and 8 (plain language / recovery) |
-| Textbook dump after success | 9 (Day 2 only) |
+| Long run with no progress, estimate, or status | 4 (no silent work) |
+| Subagents or background jobs working invisibly | 4 (no silent work) |
+| Assumed prior install, flash, or login | 6 (first-time setup) |
+| Blind auto-select on multi-candidate hosts | 7 (scary surfaces) |
+| Continuing after a required human step without confirmation | 8 and 9 (confirm / no cliff) |
+| Stack trace as the only failure UX | 7 and 9 (plain language / recovery) |
+| Textbook dump after success | 10 (tomorrow only) |
+| Filing an issue in their name without asking | 11 (compound, but ask first) |
+| Hitting the same contract gap every session and never filing it | 11 (compound) |
 | Softening the contract in a local fork | Drift; pin or quote instead |
 
 Scoring these in CI belongs in [`eval/`](../eval/). Encoding prevention in tools belongs in [`surface/`](../surface/).
@@ -115,14 +135,16 @@ manners for agents operating tools for smart, high-judgment non-experts.
 Summary (full contract is normative):
 
 1. Assume low ops literacy, high judgment.
-2. No wall of unexplained shell.
+2. No walls of shell or choice.
 3. Prefer doing over instructing.
-4. Human acts: explicit, one step, dumb-simple.
-5. Own first-time setup from zero.
-6. Own scary surfaces in plain language.
-7. Confirm in their words before irreversible or physical steps.
-8. Never leave them at a cliff.
-9. Teach only what Day 2 requires.
+4. No silent work.
+5. Human acts are explicit and dumb-simple.
+6. Own first-time setup from zero.
+7. Own scary surfaces in plain language.
+8. Confirm what they can see, in their words.
+9. Never leave them at a cliff.
+10. Teach only what tomorrow requires.
+11. Compound what you learn.
 
 Domain notes for this repo:
 - <!-- first-run, scary surfaces, one update command -->
@@ -137,12 +159,12 @@ Domain notes for this repo:
 
 ## Domain notes (not a fork)
 
-Principles are universal. Examples are not. Domain notes belong in the consuming project (or a domain pack) and may include:
+Tenets are universal. Examples are not. Domain notes belong in the consuming project (or a domain pack) and may include:
 
 - Operator persona notes (still smart and high-judgment).
 - First-run path from zero.
 - Scary surfaces glossary (plain language).
-- One Day-2 update or recovery leave-behind.
+- One update or recovery leave-behind.
 
 Example (embedded / host-first metal); see [silico](https://github.com/tig/silico):
 
@@ -155,8 +177,8 @@ Tool verbs and error UX for a domain go in [`surface/`](../surface/). Bad and go
 ## Contract adoption
 
 - [ ] Agent-visible link or pin to this contract.
-- [ ] Principles marked non-negotiable on the operator path.
+- [ ] Tenets marked non-negotiable on the operator path.
 - [ ] Domain notes cover first-run and one scary surface (or dated plan).
-- [ ] Day-2 leave-behind: one update or recovery path in plain language.
+- [ ] Leave-behind: one update or recovery path in plain language.
 
 Full product adoption (surface and eval) is in the [root README](../README.md#adoption-checklist).