@andresmassello/uscha 1.82.0 → 1.83.0
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.
- package/README.md +1 -1
- package/package.json +1 -1
- package/uscha-kit/.claude/skills/uscha-devloop/qa_ledger.py +21 -5
- package/uscha-kit/.claude-plugin/plugin.json +1 -1
- package/uscha-kit/.codex-plugin/plugin.json +1 -1
- package/uscha-kit/README.md +1 -1
- package/uscha-kit/VERSION +1 -1
- package/uscha-kit/reports/junit/.sched-cases.json +1 -0
- package/uscha-kit/skills/uscha-devloop/qa_ledger.py +21 -5
- package/uscha-kit/uscha.config.json +1 -1
package/README.md
CHANGED
|
@@ -40,7 +40,7 @@ Requires **Python 3.8+** on the machine (the engine is Python stdlib — no pip
|
|
|
40
40
|
runtime dependencies). The npm package is a thin router; the canonical installer is
|
|
41
41
|
`uscha-kit/install-uscha.py`.
|
|
42
42
|
|
|
43
|
-
**Kit v1.
|
|
43
|
+
**Kit v1.83.0** <!-- uscha:version --> · [uscha.dev](https://uscha.dev) ·
|
|
44
44
|
[changelog](https://github.com/andresmassello/uscha/blob/main/uscha-kit/CHANGELOG.md)
|
|
45
45
|
(the per-release changelogs live in the repo, not in the npm tarball)
|
|
46
46
|
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@andresmassello/uscha",
|
|
3
|
-
"version": "1.
|
|
3
|
+
"version": "1.83.0",
|
|
4
4
|
"description": "Spec-driven development for LLM coding agents: 9 skills + a stdlib evidence engine. Facts block, guesses advise; the human approves.",
|
|
5
5
|
"author": {
|
|
6
6
|
"name": "Andres Massello",
|
|
@@ -5976,7 +5976,7 @@ def cmd_lang_compare(args):
|
|
|
5976
5976
|
"""Compare a FREE-prose arm and a CONTROLLED (EARS+STE) arm of the same canonical package
|
|
5977
5977
|
(ADR-019). The two arms MUST share one withheld oracle -- a differing oracle is a mechanical
|
|
5978
5978
|
refusal, because the whole comparison rests on the arms targeting the same behaviour. Emits
|
|
5979
|
-
the per-arm metrics, the delta, and a COMPUTED verdict (REDUCED / NO EFFECT / WORSE); a null
|
|
5979
|
+
the per-arm metrics, the delta, and a COMPUTED verdict (REDUCED / IMPROVED / MIXED / NO EFFECT / WORSE); a null
|
|
5980
5980
|
is a first-class result. Consults no model."""
|
|
5981
5981
|
hf, hc = _oracle_hash(args.free), _oracle_hash(args.controlled)
|
|
5982
5982
|
if hf is None or hc is None:
|
|
@@ -5999,15 +5999,22 @@ def cmd_lang_compare(args):
|
|
|
5999
5999
|
d_green = ctrl["greens"] - free["greens"]
|
|
6000
6000
|
pf, pc = free["mean_passrate"], ctrl["mean_passrate"]
|
|
6001
6001
|
d_pass = (pc - pf) if (pf is not None and pc is not None) else None
|
|
6002
|
-
# Verdict is BEHAVIOUR-FIRST
|
|
6003
|
-
#
|
|
6002
|
+
# Verdict is BEHAVIOUR-FIRST in BOTH directions. The M4 lesson: lower variance toward a
|
|
6003
|
+
# WORSE answer is not a win (MIXED). Its mirror, the ADR-025 scheduler lesson: higher
|
|
6004
|
+
# variance toward a BETTER answer is not a loss (IMPROVED) -- two compilers converging on
|
|
6005
|
+
# the SAME bug reads as low variance, and a rewrite that separates them raises variance
|
|
6006
|
+
# while fixing behaviour. A regression is a lost all-green OR a mean pass-rate drop beyond
|
|
6007
|
+
# the pass-rate margin; an improvement is the exact mirror. ADR-019 + ADR-026.
|
|
6004
6008
|
variance_reduced = d_var is not None and d_var <= -_LANG_MARGIN
|
|
6005
6009
|
variance_worse = d_var is not None and d_var >= _LANG_MARGIN
|
|
6006
6010
|
regressed = (d_green < 0) or (d_pass is not None and d_pass <= -_LANG_PR_MARGIN)
|
|
6011
|
+
improved = (d_green > 0) or (d_pass is not None and d_pass >= _LANG_PR_MARGIN)
|
|
6007
6012
|
if variance_reduced and regressed:
|
|
6008
6013
|
verdict = "MIXED" # variance down but behaviour regressed
|
|
6009
6014
|
elif variance_reduced:
|
|
6010
|
-
verdict = "REDUCED" # variance down, behaviour held
|
|
6015
|
+
verdict = "REDUCED" # variance down, behaviour held or better
|
|
6016
|
+
elif improved and not regressed:
|
|
6017
|
+
verdict = "IMPROVED" # behaviour up, variance not down
|
|
6011
6018
|
elif variance_worse or regressed:
|
|
6012
6019
|
verdict = "WORSE"
|
|
6013
6020
|
else:
|
|
@@ -6016,6 +6023,7 @@ def cmd_lang_compare(args):
|
|
|
6016
6023
|
"unresolved_intent_count": round(d_ui, 3), "oracle_green": d_green,
|
|
6017
6024
|
"mean_passrate": round(d_pass, 4) if d_pass is not None else None}
|
|
6018
6025
|
report = {"free": free, "controlled": ctrl, "delta": delta, "verdict": verdict,
|
|
6026
|
+
"regressed": regressed, "improved": improved,
|
|
6019
6027
|
"margin": _LANG_MARGIN, "passrate_margin": _LANG_PR_MARGIN, "oracle_shared": True}
|
|
6020
6028
|
if args.out:
|
|
6021
6029
|
with open(args.out, "w", encoding="utf-8", newline="\n") as fh:
|
|
@@ -6078,7 +6086,15 @@ def _render_lang_md(r):
|
|
|
6078
6086
|
"the delta here. A null result, reported as a null — not a failure.",
|
|
6079
6087
|
"WORSE": "Controlled authoring increased variance, or regressed behaviour (lost an "
|
|
6080
6088
|
"all-green or dropped mean pass-rate) beyond the margins — reported "
|
|
6081
|
-
"honestly."
|
|
6089
|
+
"honestly.",
|
|
6090
|
+
"IMPROVED": "Controlled authoring IMPROVED behaviour (gained an all-green or raised "
|
|
6091
|
+
"mean pass-rate beyond the %.2f margin) while inter-compiler variance "
|
|
6092
|
+
"did not fall. Behaviour won; variance did not — check whether the free "
|
|
6093
|
+
"arm's low variance was convergence on a shared reading (right or "
|
|
6094
|
+
"wrong): two compilers resolving the same ambiguity the same way read "
|
|
6095
|
+
"as agreement, and a rewrite that separates them raises variance "
|
|
6096
|
+
"while moving behaviour."
|
|
6097
|
+
% r.get("passrate_margin", 0.02)}[r["verdict"]],
|
|
6082
6098
|
"", "*The oracle is byte-identical across both arms, so behaviour is held fixed; "
|
|
6083
6099
|
"the only variable is the authoring discipline. The judgement of \"same semantic "
|
|
6084
6100
|
"content\" between the two canonical packages is human — a stated limitation.*", ""]
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
|
|
3
3
|
"name": "uscha",
|
|
4
|
-
"version": "1.
|
|
4
|
+
"version": "1.83.0",
|
|
5
5
|
"displayName": "Uscha",
|
|
6
6
|
"description": "Spec-driven development for LLM coding agents: 9 skills (discovery, adr-refine, reverse-discovery, characterize, devloop, sysdoc, rubric, mirador, status) + a stdlib measurement engine (qa_ledger.py, 49 subcommands + universal installer + npm/npx router). Facts block, guesses advise; the human approves.",
|
|
7
7
|
"author": {
|
package/uscha-kit/README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# uscha-kit
|
|
2
2
|
|
|
3
|
-
**Kit version:** v1.
|
|
3
|
+
**Kit version:** v1.83.0 <!-- uscha:version --> · **[uscha.dev](https://uscha.dev)**
|
|
4
4
|
|
|
5
5
|
Spec-driven orchestrator + multi-repo QA for Claude Code, with a deterministic ledger.
|
|
6
6
|
**Nine skills** (`uscha-discovery`, `uscha-adr-refine`, `uscha-devloop`, `uscha-sysdoc`, `uscha-reverse-discovery`,
|
package/uscha-kit/VERSION
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
uscha-kit 1.
|
|
1
|
+
uscha-kit 1.83.0
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"AC-SH-01": true, "AC-SH-02": true, "AC-SH-03": true, "AC-LI-01": true, "AC-LI-02": true, "AC-LI-03": true}
|
|
@@ -5976,7 +5976,7 @@ def cmd_lang_compare(args):
|
|
|
5976
5976
|
"""Compare a FREE-prose arm and a CONTROLLED (EARS+STE) arm of the same canonical package
|
|
5977
5977
|
(ADR-019). The two arms MUST share one withheld oracle -- a differing oracle is a mechanical
|
|
5978
5978
|
refusal, because the whole comparison rests on the arms targeting the same behaviour. Emits
|
|
5979
|
-
the per-arm metrics, the delta, and a COMPUTED verdict (REDUCED / NO EFFECT / WORSE); a null
|
|
5979
|
+
the per-arm metrics, the delta, and a COMPUTED verdict (REDUCED / IMPROVED / MIXED / NO EFFECT / WORSE); a null
|
|
5980
5980
|
is a first-class result. Consults no model."""
|
|
5981
5981
|
hf, hc = _oracle_hash(args.free), _oracle_hash(args.controlled)
|
|
5982
5982
|
if hf is None or hc is None:
|
|
@@ -5999,15 +5999,22 @@ def cmd_lang_compare(args):
|
|
|
5999
5999
|
d_green = ctrl["greens"] - free["greens"]
|
|
6000
6000
|
pf, pc = free["mean_passrate"], ctrl["mean_passrate"]
|
|
6001
6001
|
d_pass = (pc - pf) if (pf is not None and pc is not None) else None
|
|
6002
|
-
# Verdict is BEHAVIOUR-FIRST
|
|
6003
|
-
#
|
|
6002
|
+
# Verdict is BEHAVIOUR-FIRST in BOTH directions. The M4 lesson: lower variance toward a
|
|
6003
|
+
# WORSE answer is not a win (MIXED). Its mirror, the ADR-025 scheduler lesson: higher
|
|
6004
|
+
# variance toward a BETTER answer is not a loss (IMPROVED) -- two compilers converging on
|
|
6005
|
+
# the SAME bug reads as low variance, and a rewrite that separates them raises variance
|
|
6006
|
+
# while fixing behaviour. A regression is a lost all-green OR a mean pass-rate drop beyond
|
|
6007
|
+
# the pass-rate margin; an improvement is the exact mirror. ADR-019 + ADR-026.
|
|
6004
6008
|
variance_reduced = d_var is not None and d_var <= -_LANG_MARGIN
|
|
6005
6009
|
variance_worse = d_var is not None and d_var >= _LANG_MARGIN
|
|
6006
6010
|
regressed = (d_green < 0) or (d_pass is not None and d_pass <= -_LANG_PR_MARGIN)
|
|
6011
|
+
improved = (d_green > 0) or (d_pass is not None and d_pass >= _LANG_PR_MARGIN)
|
|
6007
6012
|
if variance_reduced and regressed:
|
|
6008
6013
|
verdict = "MIXED" # variance down but behaviour regressed
|
|
6009
6014
|
elif variance_reduced:
|
|
6010
|
-
verdict = "REDUCED" # variance down, behaviour held
|
|
6015
|
+
verdict = "REDUCED" # variance down, behaviour held or better
|
|
6016
|
+
elif improved and not regressed:
|
|
6017
|
+
verdict = "IMPROVED" # behaviour up, variance not down
|
|
6011
6018
|
elif variance_worse or regressed:
|
|
6012
6019
|
verdict = "WORSE"
|
|
6013
6020
|
else:
|
|
@@ -6016,6 +6023,7 @@ def cmd_lang_compare(args):
|
|
|
6016
6023
|
"unresolved_intent_count": round(d_ui, 3), "oracle_green": d_green,
|
|
6017
6024
|
"mean_passrate": round(d_pass, 4) if d_pass is not None else None}
|
|
6018
6025
|
report = {"free": free, "controlled": ctrl, "delta": delta, "verdict": verdict,
|
|
6026
|
+
"regressed": regressed, "improved": improved,
|
|
6019
6027
|
"margin": _LANG_MARGIN, "passrate_margin": _LANG_PR_MARGIN, "oracle_shared": True}
|
|
6020
6028
|
if args.out:
|
|
6021
6029
|
with open(args.out, "w", encoding="utf-8", newline="\n") as fh:
|
|
@@ -6078,7 +6086,15 @@ def _render_lang_md(r):
|
|
|
6078
6086
|
"the delta here. A null result, reported as a null — not a failure.",
|
|
6079
6087
|
"WORSE": "Controlled authoring increased variance, or regressed behaviour (lost an "
|
|
6080
6088
|
"all-green or dropped mean pass-rate) beyond the margins — reported "
|
|
6081
|
-
"honestly."
|
|
6089
|
+
"honestly.",
|
|
6090
|
+
"IMPROVED": "Controlled authoring IMPROVED behaviour (gained an all-green or raised "
|
|
6091
|
+
"mean pass-rate beyond the %.2f margin) while inter-compiler variance "
|
|
6092
|
+
"did not fall. Behaviour won; variance did not — check whether the free "
|
|
6093
|
+
"arm's low variance was convergence on a shared reading (right or "
|
|
6094
|
+
"wrong): two compilers resolving the same ambiguity the same way read "
|
|
6095
|
+
"as agreement, and a rewrite that separates them raises variance "
|
|
6096
|
+
"while moving behaviour."
|
|
6097
|
+
% r.get("passrate_margin", 0.02)}[r["verdict"]],
|
|
6082
6098
|
"", "*The oracle is byte-identical across both arms, so behaviour is held fixed; "
|
|
6083
6099
|
"the only variable is the authoring discipline. The judgement of \"same semantic "
|
|
6084
6100
|
"content\" between the two canonical packages is human — a stated limitation.*", ""]
|