evidgate 0.9.2.post1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (29) hide show
  1. evidgate-0.9.2.post1/LICENSE +21 -0
  2. evidgate-0.9.2.post1/PKG-INFO +221 -0
  3. evidgate-0.9.2.post1/README.md +203 -0
  4. evidgate-0.9.2.post1/evidgate/__init__.py +6 -0
  5. evidgate-0.9.2.post1/evidgate/cli.py +599 -0
  6. evidgate-0.9.2.post1/evidgate/core/__init__.py +10 -0
  7. evidgate-0.9.2.post1/evidgate/core/builtin.py +105 -0
  8. evidgate-0.9.2.post1/evidgate/core/check.py +175 -0
  9. evidgate-0.9.2.post1/evidgate/core/evidence.py +243 -0
  10. evidgate-0.9.2.post1/evidgate/core/gate.py +154 -0
  11. evidgate-0.9.2.post1/evidgate/core/packs.py +90 -0
  12. evidgate-0.9.2.post1/evidgate/core/suggest.py +185 -0
  13. evidgate-0.9.2.post1/evidgate/ledger/__init__.py +4 -0
  14. evidgate-0.9.2.post1/evidgate/ledger/table.py +135 -0
  15. evidgate-0.9.2.post1/evidgate/packs/build.yaml +25 -0
  16. evidgate-0.9.2.post1/evidgate/packs/figures.yaml +29 -0
  17. evidgate-0.9.2.post1/evidgate/packs/repo.yaml +17 -0
  18. evidgate-0.9.2.post1/evidgate/plugins/__init__.py +1 -0
  19. evidgate-0.9.2.post1/evidgate/plugins/artifact.py +303 -0
  20. evidgate-0.9.2.post1/evidgate/plugins/prose.py +181 -0
  21. evidgate-0.9.2.post1/evidgate.egg-info/PKG-INFO +221 -0
  22. evidgate-0.9.2.post1/evidgate.egg-info/SOURCES.txt +27 -0
  23. evidgate-0.9.2.post1/evidgate.egg-info/dependency_links.txt +1 -0
  24. evidgate-0.9.2.post1/evidgate.egg-info/entry_points.txt +2 -0
  25. evidgate-0.9.2.post1/evidgate.egg-info/top_level.txt +1 -0
  26. evidgate-0.9.2.post1/pyproject.toml +50 -0
  27. evidgate-0.9.2.post1/setup.cfg +4 -0
  28. evidgate-0.9.2.post1/tests/test_evidgate.py +944 -0
  29. evidgate-0.9.2.post1/tests/test_skill_commands.py +92 -0
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2026 BringCo
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
@@ -0,0 +1,221 @@
1
+ Metadata-Version: 2.4
2
+ Name: evidgate
3
+ Version: 0.9.2.post1
4
+ Summary: Evidence gates for deliverables: every claim gets a command, an exit code, and a ledger line.
5
+ Author: BringCo
6
+ License-Expression: MIT
7
+ Project-URL: Homepage, https://github.com/BringCo/evidgate
8
+ Keywords: evidence,gate,audit,ledger,reproducibility,documentation
9
+ Classifier: Programming Language :: Python :: 3
10
+ Classifier: Operating System :: OS Independent
11
+ Classifier: Topic :: Documentation
12
+ Classifier: Topic :: Software Development :: Testing
13
+ Classifier: Topic :: Software Development :: Quality Assurance
14
+ Requires-Python: >=3.9
15
+ Description-Content-Type: text/markdown
16
+ License-File: LICENSE
17
+ Dynamic: license-file
18
+
19
+ # evidgate
20
+
21
+ [![PyPI](https://img.shields.io/pypi/v/evidgate.svg)](https://pypi.org/project/evidgate/)
22
+ [![Python](https://img.shields.io/pypi/pyversions/evidgate.svg)](https://pypi.org/project/evidgate/)
23
+ [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
24
+ [![CI](https://github.com/BringCo/evidgate/actions/workflows/ci.yml/badge.svg)](https://github.com/BringCo/evidgate/actions/workflows/ci.yml)
25
+
26
+ **English** · [简体中文](README.zh-CN.md)
27
+
28
+ **A command-line tool that turns a deliverable's claims into checks — and refuses to
29
+ load a check that cannot go red.**
30
+
31
+ It does not judge whether the content is right. It judges whether the evidence you claim
32
+ for it is real, still there, and reproducible, then writes that judgement into one ledger
33
+ you can hand to somebody else.
34
+
35
+ One ledger line = one check = the command that ran = the exit code it produced = the
36
+ point where you resume = the decision to release or block.
37
+
38
+ ## The failure this exists to catch
39
+
40
+ A check that reports green for the wrong reason is worse than no check: you stop
41
+ looking. Three shapes of that, all encountered while building this:
42
+
43
+ **1. The command never ran.** On Windows there is no `grep`. Running a missing command
44
+ exits `1` — and `exit:1` is exactly how you assert "grep found nothing". So a check that
45
+ never executed reported PASS, indefinitely. evidgate now tells "the command did not run"
46
+ apart from "the command ran and returned this code"; only the second can pass.
47
+
48
+ **2. The check had nothing to check.** A 172-file manifest where every entry was a
49
+ source made the `figures` pack report **3 PASS with 100% coverage**: with no derivative
50
+ there is nothing to compare, so every rule held vacuously. There is now a `GF0` check
51
+ whose only job is to fail when a manifest contains no derivative at all.
52
+
53
+ **3. Nobody ever asked the check to fail.** This one is silent by construction: a rule
54
+ written straight into CI and never run against a broken version is an untested rule.
55
+ So `counter` — *how does this check go red* — is a required field, and a check without
56
+ it is refused at load time, not at run time.
57
+
58
+ ## How it differs
59
+
60
+ | | typical linter | checklist (human) | evidgate |
61
+ |---|---|---|---|
62
+ | Who decides pass | the rule | a person ticking a box | the command + its exit code |
63
+ | What proves it | nothing | nothing | command, exit code, output tail, artifact hash, timestamp |
64
+ | If the rule silently stops working | silent | silent | `evidgate mutate` reports "detection 1/1" or "NOT caught" |
65
+ | Where "what is left" lives | nowhere | another sheet | the same ledger line |
66
+ | Can a rule exist with no way to fail | yes | yes | **refused at load time** |
67
+
68
+ `counter` is a required field, so **a gate that cannot fail is not a gate, it is a wish.**
69
+
70
+ ## 30 seconds
71
+
72
+ ```bash
73
+ pip install evidgate
74
+ cd <your project>
75
+ evidgate init --dir . # writes evidgate.yaml; picks a file that actually exists
76
+ evidgate doctor --dir . # yaml parseable? every command resolvable? — before blaming the project
77
+ evidgate run --dir . # ledger + verdicts; exit 1 = blocked
78
+
79
+ # straight from a clone, no install needed (stdlib only):
80
+ ./bin/evidgate --version
81
+ ```
82
+
83
+ If `doctor` prints `BAD`, fix that first — the most common failure is a `method` whose
84
+ command is not on `PATH`, or a `method` that needs `shell:` and does not declare it.
85
+
86
+ Running the bundled three-domain example:
87
+
88
+ ```console
89
+ $ evidgate run --dir examples/mixed
90
+ evidgate 0.9.2.post1 · mixed-delivery (fictional) · 7 checks
91
+ D1 FAIL(红) 草稿里的绝对化词带范围或检索日期 ─ -exit 2 (want 0) | got: L3 [首次实现] 本方案首次实现全自动估价流程。 | [absolutes] 1 finding(s) (exit 2)
92
+ D2 DIFF(黄) 草稿里的统计数字带口径 ─ -exit 3 (want 0) | got: L3 抽样调参后误差降低 18.4%。 | [numeral_scope] 1 finding(s) (exit 3)
93
+ A1 PASS 产物存在且非空 ─ +exit 0 (want 0) ; +regex OK ─ \d -> 3
94
+ A2 FAIL(红) 正本与其派生副本逐字节一致 ─ -exit 1 (want 0)
95
+ C1 PASS 代码单测全绿 ─ +exit 0 (want 0)
96
+ C2 PASS 源码里没有明文密钥 ─ +exit 0 (want 0) ; +not_contains OK:api_key ; +not_contains OK:secret_key
97
+ C3 PASS 源码里没有残留 TODO ─ +exit 1 (want 1)
98
+
99
+ PASS 4 · DIFF(黄) 1 · FAIL(红) 2 · 待补 0 · 未覆盖 0 | coverage 7/7 (100.0%)
100
+ RELEASE BLOCKED blocked by D1, A2
101
+ note: PASS 表示「证据存在且可复现」,不表示数值本身正确。
102
+ ```
103
+
104
+ The example deliberately mixes Chinese check names with English output: the shipped
105
+ `prose` rules target Chinese drafts, and one ledger holds either language.
106
+
107
+ Red blocks the release. Yellow goes to a confirmation list. **待补 is never "done".**
108
+ Exit code 1 means blocked — a non-zero run cannot be reported as a pass.
109
+
110
+ ## What it finds on real things
111
+
112
+ | target | size | result |
113
+ |---|---|---|
114
+ | Mature open-source repos | 8 | **0 findings** |
115
+ | Personal deliverable repos | 4 | **3 real defects** |
116
+ | A real paper draft | 581 sentences | 222 findings before calibration → **32** |
117
+
118
+ The zero is the boundary, not a weakness: mature projects do not commit credentials or
119
+ leave `.bak` files behind. Personal repos produced a crash on a broken symlink, the
120
+ vacuous-green manifest above, and a `suggest` that could not see `.fig`/`.dia`/`.dat` as
121
+ figure sources. This tool is for **the last look before you hand something over**.
122
+
123
+ ## Who this is for
124
+
125
+ - You are about to hand a draft, a report, an application package or a build directory to
126
+ someone who will ask "where did this number come from?"
127
+ - An agent produced the deliverable, and "done" arrived as one sentence with no command
128
+ behind it. The same holds across your own windows: a claim from another context is
129
+ something to check, not to trust.
130
+ - You already run tests and linters, and none of them leave a record of *what was
131
+ verified, with which command, when*.
132
+
133
+ ## Who it is not for
134
+
135
+ - **Content correctness** — whether the number is right or the argument holds.
136
+ - **Typography, terminology, AI-tone linting** — [Vale](https://vale.sh/) and
137
+ `deliverable-qa` have far more rules for that.
138
+ - **Auditing mature repositories** — expect zero findings.
139
+ - **Approval workflows** — no roles, no multi-user sign-off; this is a single-user CLI.
140
+
141
+ ## Install as an agent skill
142
+
143
+ `skill/evidgate/SKILL.md` is a written instruction set for an agent: when to reach for
144
+ the tool, the first-run commands, what a check looks like, and the three shapes of false
145
+ green. Copy or link the directory into your agent's skills folder:
146
+
147
+ ```bash
148
+ cp -r skill/evidgate ~/.hermes/skills/ # Hermes
149
+ ln -s "$PWD/skill/evidgate" ~/.hermes/skills/evidgate # or link instead of copying
150
+ ```
151
+
152
+ After that, an instruction as short as "run evidgate before you tell me it is done" has
153
+ something concrete behind it. `evidgate run` exits 1 when blocked, and an agent that
154
+ reports success off a non-zero run is doing precisely what this tool exists to catch.
155
+
156
+ ## Commands
157
+
158
+ ```bash
159
+ evidgate init --dir <proj> # starter evidgate.yaml
160
+ evidgate suggest --dir <proj> [--write] # read the disk, name the pack that applies
161
+ evidgate doctor --dir <proj> # yaml + command resolvability, before you blame the project
162
+ evidgate run --dir <proj> [--only C1,P2] # run + rebuild the ledger; exit 1 = blocked
163
+ evidgate resume --dir <proj> # rerun only unresolved checks
164
+ evidgate report --dir <proj> # gate_report.md: table, yellows, unverified, verdict
165
+ evidgate mutate --dir <proj> --baseline A --mutant B [--only C1] # does the check actually go red?
166
+ evidgate prose-init --dir <proj> --file <draft> # append the Chinese prose checks (P1..P5)
167
+ evidgate artifact-init --dir <proj> --out artifacts.json # scan a directory into a manifest
168
+ evidgate artifact-update --dir <proj> # recompute an artifact's own md5 (never source_md5)
169
+ evidgate artifact-stamp --dir <proj> --path P # after regenerating: record the source hash
170
+ evidgate pack [list|show <name>] # what a pack expands to
171
+ ```
172
+
173
+ ## Documentation
174
+
175
+ | doc | covers |
176
+ |---|---|
177
+ | [Writing a check](docs/writing-a-check.md) | `method` forms, `optional`, every `expect` clause, the evidence self-check, adding your own domain |
178
+ | [Artifact plugin](docs/artifact-plugin.md) | the five manifest rules, why `stale` hashes instead of comparing mtimes, `orphan` scope and cost, the one writer of `source_md5` |
179
+ | [Prose plugin](docs/prose-plugin.md) | the five Chinese rules, the calibration story, sentence boundaries, `EVIDGATE_EVIDENCE` |
180
+ | [Shell and Windows](docs/shell-and-windows.md) | why no shell by default, Git Bash discovery, the Windows false green |
181
+ | [suggest and packs](docs/suggest-and-packs.md) | the first command, the three packs and their parameters |
182
+
183
+ ## Examples
184
+
185
+ All examples are **fictional**; every defect is planted on purpose.
186
+
187
+ - `examples/mixed/` — prose + artifact + code in one ledger (the output above)
188
+ - `examples/` — a Chinese draft, six checks, 2 PASS / 2 DIFF / 2 FAIL
189
+ - `examples/figures/` — one SVG source, three derivatives, all three `stale` verdicts
190
+ side by side
191
+ - `examples/artifact/` — one markdown source with two derivatives, four asset checks
192
+
193
+ Mutation demo:
194
+
195
+ ```bash
196
+ ./bin/evidgate mutate --dir examples --baseline examples/base --mutant examples/mut --only C1
197
+ # detection power: 1/1
198
+ ```
199
+
200
+ ## Scope
201
+
202
+ **Does not** verify that a reference exists or says what you claim — use
203
+ [refchecker](https://github.com/markrussinovich/refchecker) for that.
204
+ **Does not** detect AI-generated text, judge typography, or decide whether a number is
205
+ correct.
206
+ **Does** check that the evidence you claim for your own statements, numbers and
207
+ artifacts is real, reproducible and recorded — in whatever domain your command can
208
+ observe.
209
+
210
+ `prose` is a writing prompt, not a proofreader. Measured honestly: of 22 problems found
211
+ by reading a draft by hand, and 26 reported by the rules, the overlap was small.
212
+
213
+ ## Attribution
214
+
215
+ Evidence triples and three-state verdicts follow obra/superpowers'
216
+ `verification-before-completion`; the evidence-gate framing follows TestifySec Pushgate.
217
+ What is new here is listed in [ATTRIBUTION.md](ATTRIBUTION.md).
218
+
219
+ ## License
220
+
221
+ MIT
@@ -0,0 +1,203 @@
1
+ # evidgate
2
+
3
+ [![PyPI](https://img.shields.io/pypi/v/evidgate.svg)](https://pypi.org/project/evidgate/)
4
+ [![Python](https://img.shields.io/pypi/pyversions/evidgate.svg)](https://pypi.org/project/evidgate/)
5
+ [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
6
+ [![CI](https://github.com/BringCo/evidgate/actions/workflows/ci.yml/badge.svg)](https://github.com/BringCo/evidgate/actions/workflows/ci.yml)
7
+
8
+ **English** · [简体中文](README.zh-CN.md)
9
+
10
+ **A command-line tool that turns a deliverable's claims into checks — and refuses to
11
+ load a check that cannot go red.**
12
+
13
+ It does not judge whether the content is right. It judges whether the evidence you claim
14
+ for it is real, still there, and reproducible, then writes that judgement into one ledger
15
+ you can hand to somebody else.
16
+
17
+ One ledger line = one check = the command that ran = the exit code it produced = the
18
+ point where you resume = the decision to release or block.
19
+
20
+ ## The failure this exists to catch
21
+
22
+ A check that reports green for the wrong reason is worse than no check: you stop
23
+ looking. Three shapes of that, all encountered while building this:
24
+
25
+ **1. The command never ran.** On Windows there is no `grep`. Running a missing command
26
+ exits `1` — and `exit:1` is exactly how you assert "grep found nothing". So a check that
27
+ never executed reported PASS, indefinitely. evidgate now tells "the command did not run"
28
+ apart from "the command ran and returned this code"; only the second can pass.
29
+
30
+ **2. The check had nothing to check.** A 172-file manifest where every entry was a
31
+ source made the `figures` pack report **3 PASS with 100% coverage**: with no derivative
32
+ there is nothing to compare, so every rule held vacuously. There is now a `GF0` check
33
+ whose only job is to fail when a manifest contains no derivative at all.
34
+
35
+ **3. Nobody ever asked the check to fail.** This one is silent by construction: a rule
36
+ written straight into CI and never run against a broken version is an untested rule.
37
+ So `counter` — *how does this check go red* — is a required field, and a check without
38
+ it is refused at load time, not at run time.
39
+
40
+ ## How it differs
41
+
42
+ | | typical linter | checklist (human) | evidgate |
43
+ |---|---|---|---|
44
+ | Who decides pass | the rule | a person ticking a box | the command + its exit code |
45
+ | What proves it | nothing | nothing | command, exit code, output tail, artifact hash, timestamp |
46
+ | If the rule silently stops working | silent | silent | `evidgate mutate` reports "detection 1/1" or "NOT caught" |
47
+ | Where "what is left" lives | nowhere | another sheet | the same ledger line |
48
+ | Can a rule exist with no way to fail | yes | yes | **refused at load time** |
49
+
50
+ `counter` is a required field, so **a gate that cannot fail is not a gate, it is a wish.**
51
+
52
+ ## 30 seconds
53
+
54
+ ```bash
55
+ pip install evidgate
56
+ cd <your project>
57
+ evidgate init --dir . # writes evidgate.yaml; picks a file that actually exists
58
+ evidgate doctor --dir . # yaml parseable? every command resolvable? — before blaming the project
59
+ evidgate run --dir . # ledger + verdicts; exit 1 = blocked
60
+
61
+ # straight from a clone, no install needed (stdlib only):
62
+ ./bin/evidgate --version
63
+ ```
64
+
65
+ If `doctor` prints `BAD`, fix that first — the most common failure is a `method` whose
66
+ command is not on `PATH`, or a `method` that needs `shell:` and does not declare it.
67
+
68
+ Running the bundled three-domain example:
69
+
70
+ ```console
71
+ $ evidgate run --dir examples/mixed
72
+ evidgate 0.9.2.post1 · mixed-delivery (fictional) · 7 checks
73
+ D1 FAIL(红) 草稿里的绝对化词带范围或检索日期 ─ -exit 2 (want 0) | got: L3 [首次实现] 本方案首次实现全自动估价流程。 | [absolutes] 1 finding(s) (exit 2)
74
+ D2 DIFF(黄) 草稿里的统计数字带口径 ─ -exit 3 (want 0) | got: L3 抽样调参后误差降低 18.4%。 | [numeral_scope] 1 finding(s) (exit 3)
75
+ A1 PASS 产物存在且非空 ─ +exit 0 (want 0) ; +regex OK ─ \d -> 3
76
+ A2 FAIL(红) 正本与其派生副本逐字节一致 ─ -exit 1 (want 0)
77
+ C1 PASS 代码单测全绿 ─ +exit 0 (want 0)
78
+ C2 PASS 源码里没有明文密钥 ─ +exit 0 (want 0) ; +not_contains OK:api_key ; +not_contains OK:secret_key
79
+ C3 PASS 源码里没有残留 TODO ─ +exit 1 (want 1)
80
+
81
+ PASS 4 · DIFF(黄) 1 · FAIL(红) 2 · 待补 0 · 未覆盖 0 | coverage 7/7 (100.0%)
82
+ RELEASE BLOCKED blocked by D1, A2
83
+ note: PASS 表示「证据存在且可复现」,不表示数值本身正确。
84
+ ```
85
+
86
+ The example deliberately mixes Chinese check names with English output: the shipped
87
+ `prose` rules target Chinese drafts, and one ledger holds either language.
88
+
89
+ Red blocks the release. Yellow goes to a confirmation list. **待补 is never "done".**
90
+ Exit code 1 means blocked — a non-zero run cannot be reported as a pass.
91
+
92
+ ## What it finds on real things
93
+
94
+ | target | size | result |
95
+ |---|---|---|
96
+ | Mature open-source repos | 8 | **0 findings** |
97
+ | Personal deliverable repos | 4 | **3 real defects** |
98
+ | A real paper draft | 581 sentences | 222 findings before calibration → **32** |
99
+
100
+ The zero is the boundary, not a weakness: mature projects do not commit credentials or
101
+ leave `.bak` files behind. Personal repos produced a crash on a broken symlink, the
102
+ vacuous-green manifest above, and a `suggest` that could not see `.fig`/`.dia`/`.dat` as
103
+ figure sources. This tool is for **the last look before you hand something over**.
104
+
105
+ ## Who this is for
106
+
107
+ - You are about to hand a draft, a report, an application package or a build directory to
108
+ someone who will ask "where did this number come from?"
109
+ - An agent produced the deliverable, and "done" arrived as one sentence with no command
110
+ behind it. The same holds across your own windows: a claim from another context is
111
+ something to check, not to trust.
112
+ - You already run tests and linters, and none of them leave a record of *what was
113
+ verified, with which command, when*.
114
+
115
+ ## Who it is not for
116
+
117
+ - **Content correctness** — whether the number is right or the argument holds.
118
+ - **Typography, terminology, AI-tone linting** — [Vale](https://vale.sh/) and
119
+ `deliverable-qa` have far more rules for that.
120
+ - **Auditing mature repositories** — expect zero findings.
121
+ - **Approval workflows** — no roles, no multi-user sign-off; this is a single-user CLI.
122
+
123
+ ## Install as an agent skill
124
+
125
+ `skill/evidgate/SKILL.md` is a written instruction set for an agent: when to reach for
126
+ the tool, the first-run commands, what a check looks like, and the three shapes of false
127
+ green. Copy or link the directory into your agent's skills folder:
128
+
129
+ ```bash
130
+ cp -r skill/evidgate ~/.hermes/skills/ # Hermes
131
+ ln -s "$PWD/skill/evidgate" ~/.hermes/skills/evidgate # or link instead of copying
132
+ ```
133
+
134
+ After that, an instruction as short as "run evidgate before you tell me it is done" has
135
+ something concrete behind it. `evidgate run` exits 1 when blocked, and an agent that
136
+ reports success off a non-zero run is doing precisely what this tool exists to catch.
137
+
138
+ ## Commands
139
+
140
+ ```bash
141
+ evidgate init --dir <proj> # starter evidgate.yaml
142
+ evidgate suggest --dir <proj> [--write] # read the disk, name the pack that applies
143
+ evidgate doctor --dir <proj> # yaml + command resolvability, before you blame the project
144
+ evidgate run --dir <proj> [--only C1,P2] # run + rebuild the ledger; exit 1 = blocked
145
+ evidgate resume --dir <proj> # rerun only unresolved checks
146
+ evidgate report --dir <proj> # gate_report.md: table, yellows, unverified, verdict
147
+ evidgate mutate --dir <proj> --baseline A --mutant B [--only C1] # does the check actually go red?
148
+ evidgate prose-init --dir <proj> --file <draft> # append the Chinese prose checks (P1..P5)
149
+ evidgate artifact-init --dir <proj> --out artifacts.json # scan a directory into a manifest
150
+ evidgate artifact-update --dir <proj> # recompute an artifact's own md5 (never source_md5)
151
+ evidgate artifact-stamp --dir <proj> --path P # after regenerating: record the source hash
152
+ evidgate pack [list|show <name>] # what a pack expands to
153
+ ```
154
+
155
+ ## Documentation
156
+
157
+ | doc | covers |
158
+ |---|---|
159
+ | [Writing a check](docs/writing-a-check.md) | `method` forms, `optional`, every `expect` clause, the evidence self-check, adding your own domain |
160
+ | [Artifact plugin](docs/artifact-plugin.md) | the five manifest rules, why `stale` hashes instead of comparing mtimes, `orphan` scope and cost, the one writer of `source_md5` |
161
+ | [Prose plugin](docs/prose-plugin.md) | the five Chinese rules, the calibration story, sentence boundaries, `EVIDGATE_EVIDENCE` |
162
+ | [Shell and Windows](docs/shell-and-windows.md) | why no shell by default, Git Bash discovery, the Windows false green |
163
+ | [suggest and packs](docs/suggest-and-packs.md) | the first command, the three packs and their parameters |
164
+
165
+ ## Examples
166
+
167
+ All examples are **fictional**; every defect is planted on purpose.
168
+
169
+ - `examples/mixed/` — prose + artifact + code in one ledger (the output above)
170
+ - `examples/` — a Chinese draft, six checks, 2 PASS / 2 DIFF / 2 FAIL
171
+ - `examples/figures/` — one SVG source, three derivatives, all three `stale` verdicts
172
+ side by side
173
+ - `examples/artifact/` — one markdown source with two derivatives, four asset checks
174
+
175
+ Mutation demo:
176
+
177
+ ```bash
178
+ ./bin/evidgate mutate --dir examples --baseline examples/base --mutant examples/mut --only C1
179
+ # detection power: 1/1
180
+ ```
181
+
182
+ ## Scope
183
+
184
+ **Does not** verify that a reference exists or says what you claim — use
185
+ [refchecker](https://github.com/markrussinovich/refchecker) for that.
186
+ **Does not** detect AI-generated text, judge typography, or decide whether a number is
187
+ correct.
188
+ **Does** check that the evidence you claim for your own statements, numbers and
189
+ artifacts is real, reproducible and recorded — in whatever domain your command can
190
+ observe.
191
+
192
+ `prose` is a writing prompt, not a proofreader. Measured honestly: of 22 problems found
193
+ by reading a draft by hand, and 26 reported by the rules, the overlap was small.
194
+
195
+ ## Attribution
196
+
197
+ Evidence triples and three-state verdicts follow obra/superpowers'
198
+ `verification-before-completion`; the evidence-gate framing follows TestifySec Pushgate.
199
+ What is new here is listed in [ATTRIBUTION.md](ATTRIBUTION.md).
200
+
201
+ ## License
202
+
203
+ MIT
@@ -0,0 +1,6 @@
1
+ """evidgate — evidence gates for documents.
2
+
3
+ Core idea: one ledger line == one check == one piece of evidence == one resume point
4
+ == one release decision. Nothing lives in two places.
5
+ """
6
+ __version__ = "0.9.2.post1"