pi-lens 4.4.7 → 4.4.8

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 (167) hide show
  1. package/CHANGELOG.md +149 -0
  2. package/README.md +12 -2
  3. package/config/markdownlint/run-no-fix.mjs +36 -0
  4. package/dist/{chunk-WME2OQST.js → chunk-2453RRJ2.js} +11 -6
  5. package/dist/{chunk-FZKFVUEC.js → chunk-2BKUFQKR.js} +276 -208
  6. package/dist/chunk-2TS4GLOJ.js +31 -0
  7. package/dist/{chunk-QHSNL7B2.js → chunk-2ZE3AUJQ.js} +12 -14
  8. package/dist/{chunk-MPKRJBPK.js → chunk-35WKJWO5.js} +28 -30
  9. package/dist/chunk-3JAHQBSY.js +27 -0
  10. package/dist/{chunk-4LM3VAZH.js → chunk-3O5EII3W.js} +2 -2
  11. package/dist/chunk-447YZL4S.js +25 -0
  12. package/dist/{chunk-H7OR7DFZ.js → chunk-4B2YFN4U.js} +25 -21
  13. package/dist/{chunk-OSTKYB55.js → chunk-4BKU7PV3.js} +1 -1
  14. package/dist/{chunk-RRV3FNNM.js → chunk-4LL2VRNY.js} +27 -28
  15. package/dist/{chunk-BZARLRJW.js → chunk-4NBWFYWW.js} +9 -9
  16. package/dist/{chunk-KKCK636N.js → chunk-4QG55RRC.js} +12 -12
  17. package/dist/{chunk-RQS4QQSD.js → chunk-4W52ZKZX.js} +2 -2
  18. package/dist/{chunk-Z2XAOKHK.js → chunk-565VRUP4.js} +2 -2
  19. package/dist/{chunk-HWDAVBFU.js → chunk-57X2CDRX.js} +23 -23
  20. package/dist/{chunk-MB6PXBUA.js → chunk-5F3XF3SX.js} +7 -15
  21. package/dist/{chunk-O4BJFESU.js → chunk-5RNB4HVK.js} +1 -1
  22. package/dist/{chunk-ZMAUPOMG.js → chunk-5T5A6YAJ.js} +35 -34
  23. package/dist/{chunk-4ZYJHQMB.js → chunk-6FIRFP6A.js} +12 -12
  24. package/dist/{chunk-4YHYPW2G.js → chunk-6GMGEKQN.js} +9 -5
  25. package/dist/{chunk-4LOE5R5P.js → chunk-6LJOJ6EO.js} +20 -15
  26. package/dist/{chunk-3YY7IXIL.js → chunk-6YUNI7B5.js} +3 -3
  27. package/dist/{chunk-PKES3ISE.js → chunk-7ANZ7XDD.js} +2 -2
  28. package/dist/{chunk-FB5SCCPT.js → chunk-7CGIDRIK.js} +16 -15
  29. package/dist/{chunk-L3CMBQMH.js → chunk-7MAD5ZAY.js} +16 -17
  30. package/dist/{chunk-WDBICWH3.js → chunk-7TNILKGY.js} +1 -1
  31. package/dist/{chunk-RORGS5II.js → chunk-A5GRLYPZ.js} +9 -9
  32. package/dist/{chunk-MNJIXFX4.js → chunk-AGGUWKRK.js} +1 -1
  33. package/dist/{chunk-ABPKUVA4.js → chunk-BBET6HD6.js} +1 -1
  34. package/dist/{chunk-L5JYIU47.js → chunk-BHWW6FAX.js} +14 -13
  35. package/dist/{chunk-URCQWAQ4.js → chunk-BOPDQ7AZ.js} +23 -21
  36. package/dist/{chunk-KDX35VXP.js → chunk-BQG2QFVO.js} +158 -133
  37. package/dist/{chunk-CB4WHS4H.js → chunk-BUZIQDFS.js} +3 -3
  38. package/dist/{chunk-2PHMIAKO.js → chunk-CM73PPGD.js} +4 -4
  39. package/dist/{chunk-DH5JPM3K.js → chunk-CUDD35TW.js} +10 -9
  40. package/dist/{chunk-AWDSEDER.js → chunk-CZL44WFN.js} +2 -2
  41. package/dist/{chunk-N5WSK2K7.js → chunk-DIWK3LBC.js} +12 -15
  42. package/dist/{chunk-24BILSZW.js → chunk-DYRZGUMA.js} +2 -2
  43. package/dist/{chunk-PR6PO3ES.js → chunk-FHEV6T4L.js} +22 -18
  44. package/dist/{chunk-DQZFQSR5.js → chunk-FRLAUP2B.js} +4 -4
  45. package/dist/{chunk-RZOXUSPC.js → chunk-G5GV2Q7K.js} +2 -2
  46. package/dist/{chunk-EKWX4WD3.js → chunk-G6JKM4O7.js} +7 -7
  47. package/dist/chunk-GGYOZWHW.js +27 -0
  48. package/dist/{chunk-XT2HTCJ5.js → chunk-GIRBSBGL.js} +6 -5
  49. package/dist/{chunk-PQJFG24Z.js → chunk-GMROJJRN.js} +1292 -946
  50. package/dist/{chunk-U6VN4NVL.js → chunk-HHR7WGCI.js} +3 -3
  51. package/dist/{chunk-TIQJPRDD.js → chunk-HKIL6DCL.js} +3 -2
  52. package/dist/{chunk-SNDXCFRS.js → chunk-HUAM7QG3.js} +11 -11
  53. package/dist/{chunk-QOE342SU.js → chunk-HXFUTMX2.js} +2 -2
  54. package/dist/{chunk-U7BD2QL7.js → chunk-I22THIHI.js} +12 -12
  55. package/dist/chunk-IE5D3RA5.js +58 -0
  56. package/dist/{chunk-DPGEAHZT.js → chunk-IYVYZQRL.js} +10 -16
  57. package/dist/{chunk-3F6BMY53.js → chunk-JBP7KESM.js} +1 -1
  58. package/dist/{chunk-U5HPROLA.js → chunk-JE6HKLC7.js} +2 -2
  59. package/dist/{chunk-ZE3KXHS7.js → chunk-JPRDTY5S.js} +85 -3
  60. package/dist/{chunk-HO6LWE2E.js → chunk-JYLOYTZD.js} +17 -1
  61. package/dist/{chunk-KW5JQPO5.js → chunk-JZRP4DGP.js} +2 -2
  62. package/dist/{chunk-CPSR3AVL.js → chunk-KZ4ALBCH.js} +2 -2
  63. package/dist/{chunk-6IR3ZYIB.js → chunk-L3CLUGDY.js} +3 -3
  64. package/dist/{chunk-LV4UMYVH.js → chunk-L4NEVZ4T.js} +28 -27
  65. package/dist/chunk-LPYTPNOJ.js +22 -0
  66. package/dist/{chunk-WGRJOXED.js → chunk-LSFXZLPW.js} +1016 -636
  67. package/dist/{chunk-LBLBAFG3.js → chunk-MC6VB2E5.js} +4 -4
  68. package/dist/{chunk-U2SDMG7C.js → chunk-NM2NWH55.js} +5 -5
  69. package/dist/{chunk-T7A2HK2B.js → chunk-NSPVR3FN.js} +19 -18
  70. package/dist/{chunk-ZGSJDWQU.js → chunk-O2ARHQKF.js} +2 -2
  71. package/dist/{chunk-OXOI55RU.js → chunk-O4N5BQ3T.js} +10 -10
  72. package/dist/{chunk-3FRLWYBG.js → chunk-OC7VWJBZ.js} +39 -40
  73. package/dist/chunk-OFNMHPBT.js +89 -0
  74. package/dist/{chunk-HR3OJ3TH.js → chunk-OWTG6UMO.js} +21 -21
  75. package/dist/{chunk-75GAYVNO.js → chunk-OY7WZI7Q.js} +6 -6
  76. package/dist/{chunk-ZJTT47IG.js → chunk-OZWXLJMX.js} +2 -2
  77. package/dist/{chunk-F6X3YJWD.js → chunk-PIQQM32Y.js} +2 -2
  78. package/dist/chunk-PKLB2XJL.js +8 -0
  79. package/dist/{chunk-EAOK7MIZ.js → chunk-PKSUMEPT.js} +17 -4
  80. package/dist/{chunk-YISME6ON.js → chunk-PPKECBPP.js} +1 -1
  81. package/dist/{chunk-4L36HG27.js → chunk-R2QQC33W.js} +14 -14
  82. package/dist/{chunk-LUBPZN36.js → chunk-R4QWJQ2H.js} +11 -11
  83. package/dist/{chunk-V63BC55Q.js → chunk-RGGH6M34.js} +1 -1
  84. package/dist/{chunk-Y7F7IEE3.js → chunk-RZG472YW.js} +65 -7
  85. package/dist/{chunk-PUZ3E2J4.js → chunk-SDFKOLSF.js} +6 -6
  86. package/dist/{chunk-YHC2O2WO.js → chunk-SFLHSL5R.js} +9 -9
  87. package/dist/{chunk-CCV2EU7T.js → chunk-SP4TFY5F.js} +14 -13
  88. package/dist/{chunk-J6JP25XU.js → chunk-SQC7K3SP.js} +4 -3
  89. package/dist/{chunk-3K2F6UDZ.js → chunk-T2S3FNNJ.js} +44 -42
  90. package/dist/{chunk-2TF6RXWB.js → chunk-TV6RXRYH.js} +10 -10
  91. package/dist/{chunk-GNSIWBAT.js → chunk-U3AS2FR6.js} +3 -3
  92. package/dist/{chunk-F7K5IBZG.js → chunk-UHHLR2VY.js} +40 -40
  93. package/dist/chunk-UYXV5NIS.js +20 -0
  94. package/dist/{chunk-VSFLRN5Y.js → chunk-VLQ7UKZH.js} +1 -1
  95. package/dist/{chunk-VUZWQ7UG.js → chunk-VN3L2JWZ.js} +46 -72
  96. package/dist/{chunk-NA7SEXHE.js → chunk-VOP4PJKA.js} +7 -7
  97. package/dist/{chunk-F5Z6URKM.js → chunk-VOXTCULR.js} +2 -2
  98. package/dist/{chunk-LLS45EWB.js → chunk-VPLBLWFO.js} +6 -8
  99. package/dist/chunk-W3BGKBBK.js +27 -0
  100. package/dist/{chunk-ZKXQKGWZ.js → chunk-WC6WRNQT.js} +4 -3
  101. package/dist/{chunk-YVHBNSME.js → chunk-WMDUT4KH.js} +1148 -1132
  102. package/dist/{chunk-KXF5R4UN.js → chunk-WQV6BI4Q.js} +1 -1
  103. package/dist/chunk-WXUDRAG6.js +24 -0
  104. package/dist/{chunk-KPPJTH44.js → chunk-X35S7HSB.js} +28 -10
  105. package/dist/{chunk-JMEKXFOL.js → chunk-X66YJ2SS.js} +40 -34
  106. package/dist/{chunk-CE6VMBCB.js → chunk-XOMGO63G.js} +22 -21
  107. package/dist/{chunk-LS3QJRUB.js → chunk-XWIME22V.js} +13 -13
  108. package/dist/{chunk-R5MOQM2H.js → chunk-XYLC35E4.js} +119 -87
  109. package/dist/{chunk-OE5ZBP7F.js → chunk-YHGYAEFH.js} +18 -18
  110. package/dist/{chunk-7E2563GA.js → chunk-ZFORSIED.js} +13 -13
  111. package/dist/{chunk-E7HJ56Y7.js → chunk-ZKLTNLAW.js} +64 -73
  112. package/dist/chunk-ZONZ6YY3.js +74 -0
  113. package/dist/chunk-ZSM35JF4.js +30 -0
  114. package/dist/{chunk-RC2Z6K47.js → chunk-ZZDIOAAJ.js} +15 -6
  115. package/dist/index.js +6993 -5396
  116. package/dist/mcp/analyze-cli.js +10 -10
  117. package/dist/mcp/cli.js +44 -42
  118. package/dist/mcp/server.js +1627 -504
  119. package/dist/mcp/worker.js +62 -61
  120. package/dist/probes/installer.js +20 -19
  121. package/docs/adr/0003-git-guard-latch-writer.md +11 -6
  122. package/docs/agent-tools.md +18 -8
  123. package/docs/alignment-families.md +29 -0
  124. package/docs/ast-grep_rules_catalog.md +6 -4
  125. package/docs/ci-bot-pr-checks.md +126 -0
  126. package/docs/configuration.md +99 -3
  127. package/docs/defect-shapes.md +55 -12
  128. package/docs/engineering-principles.md +27 -2
  129. package/docs/features.md +2 -2
  130. package/docs/invariants/git-guard-host.md +7 -3
  131. package/docs/invariants/language-config.md +15 -4
  132. package/docs/invariants/lsp-trust-exec.md +16 -2
  133. package/docs/invariants/paths-os.md +4 -2
  134. package/docs/lsp-capability-matrix.md +119 -39
  135. package/docs/merge-queue-rollout.md +1 -1
  136. package/docs/pi-lens-fixer.md +24 -17
  137. package/docs/pi-lens-merge-policy.md +20 -11
  138. package/docs/pi-lens-monitor.md +9 -0
  139. package/docs/pi-lens-orchestration.md +4 -1
  140. package/docs/pi-lens-reviewer.md +5 -5
  141. package/docs/pi-lens-subagent.md +6 -3
  142. package/docs/public-api-stability.md +18 -2
  143. package/docs/schema/pi-lens-config-v1.json +10 -0
  144. package/docs/settings.md +10 -6
  145. package/docs/word-index.md +17 -5
  146. package/package.json +5 -3
  147. package/rules/ast-grep-rules/rule-tests/no-accumulating-spread-js-test.yml +37 -0
  148. package/rules/ast-grep-rules/rule-tests/ts-redundant-filter-map-test.yml +12 -1
  149. package/rules/ast-grep-rules/rule-tests-disabled/filter-then-map-js-test.yml +12 -0
  150. package/rules/ast-grep-rules/rule-tests-disabled/filter-then-map-test.yml +25 -0
  151. package/rules/ast-grep-rules/rules/no-accumulating-spread-js.yml +49 -1
  152. package/rules/ast-grep-rules/rules/ts-redundant-filter-map.yml +9 -1
  153. package/rules/ast-grep-rules/rules-disabled/filter-then-map-js.yml +17 -0
  154. package/rules/ast-grep-rules/rules-disabled/filter-then-map.yml +36 -0
  155. package/rules/ast-grep-rules/self-scan-baseline.json +1 -1
  156. package/dist/chunk-46TNM7XZ.js +0 -88
  157. package/dist/chunk-72OKUXSD.js +0 -31
  158. package/dist/chunk-7GLOLOSH.js +0 -57
  159. package/dist/chunk-AKUUMGD5.js +0 -26
  160. package/dist/chunk-ASUXV767.js +0 -22
  161. package/dist/chunk-CTADOKYM.js +0 -25
  162. package/dist/chunk-JJ2RRESP.js +0 -58
  163. package/dist/chunk-LSC3VADM.js +0 -23
  164. package/dist/chunk-SHLDT3OI.js +0 -27
  165. package/dist/chunk-UF7RTZSO.js +0 -27
  166. package/dist/chunk-W4WP2ERB.js +0 -30
  167. package/dist/chunk-YES5CACN.js +0 -73
@@ -179,9 +179,12 @@ file-scoped worklist and receives no project-wide scanner output.
179
179
  decides only delivery, and a run with no verdict (failed, unanswered, in
180
180
  flight) observes nothing, so neither retires a recorded blocker. A skip
181
181
  has a verdict only with a stated not-applicable reason
182
- (`RUNNER_SKIP_REASONS`). A recorded blocker is still observed only at its
183
- own line, aligned to the current bytes: another occurrence of the same
184
- rule is not it. A fallback for missing evidence narrows what retires:
182
+ (`RUNNER_SKIP_REASONS`). `not-enabled` (an opt-in runner left off, #5042)
183
+ is the one stated reason that stays a no-verdict: admitting it would retire
184
+ recorded helm-render blockers and stop their owed ranges. A recorded blocker
185
+ is still observed only at its own line, aligned to the current bytes:
186
+ another occurrence of the same rule is not it. A fallback for missing
187
+ evidence narrows what retires:
185
188
  with no alignment, its identity anywhere keeps it. It stays a per-write
186
189
  hold and never widens what counts as the same finding: the match still
187
190
  needs the entry's own line text, and owed lines are matched by content,
@@ -195,6 +198,18 @@ file-scoped worklist and receives no project-wide scanner output.
195
198
  kept entry with no place in a write's bytes is recorded as such
196
199
  (`positionStale`), never inferred later from one line's hash, and a line
197
200
  with no word character owes nothing by its text.
201
+ A non-authoritative answer never stands in for a primary that never
202
+ answered: a `fallbackFor` alternate (python-jedi beside a pyright that was
203
+ trust-refused, failed to start, sat in a breaker cool-down or was not
204
+ installed) is not its owner, so the file reads unavailable with the owner
205
+ named, never "confirmed clean" (#5014). The rule admits the known-good
206
+ forms (a non-alternate primary acquired, an alternate standing in for an
207
+ owner the operator disabled, an alternate whose owner resolved no root) in
208
+ `unansweredPrimaries` (`clients/lsp/index.ts`); every other cause, listed or
209
+ not, lands unconfirmed. Screen: can a file read clean when its primary
210
+ never answered? Population enforcer:
211
+ `tests/clients/lsp/unanswered-primaries-fallback-5014.test.ts` (every
212
+ registered `fallbackFor` alternate).
198
213
  Enforced by `tests/clients/dispatch/runners/run-outcome-ratchet.test.ts`,
199
214
  `tests/clients/dispatch/runners/parsed-nothing-sweep.test.ts` and
200
215
  `tests/clients/delivery-surface-ratchet.test.ts` (partial: no detector
@@ -605,11 +620,13 @@ file-scoped worklist and receives no project-wide scanner output.
605
620
  the lane's timeout scale and is refused; a same-named local constant is
606
621
  admitted and an import alias of the shared one is refused (no live member
607
622
  either way). Screen: does this case call a walker or read a binding one
608
- produced, and is its third argument one of those two forms? Census: 151 in
609
- `tests/config`, all budgeted; 79 outside it still keep the default (a probe,
610
- not a command). Enforcer: `tests/config/tree-census-timeout.test.ts` (it
611
- derives the declaration callee from the file's vitest imports, resolves
612
- walker aliases, and counts a test-shaped call it cannot trace as a census).
623
+ produced, and is its third argument one of those two forms? Census: 244
624
+ under `tests/` (154 in `tests/config`, all budgeted; 90 outside it, of which
625
+ 87 are refused and admitted by name in
626
+ `tests/config/tree-census-timeout.baseline.json`, and 3 budgeted). Enforcer:
627
+ `tests/config/tree-census-timeout.test.ts` (it derives the declaration
628
+ callee from the file's vitest imports, resolves walker aliases, and counts a
629
+ test-shaped call it cannot trace as a census).
613
630
  It does not read a walker reached through a helper in another module, a
614
631
  variable-bound arrow-function walker, a local spawn wrapper, or a walk named
615
632
  only by an enclosing `describe.each` or loop; `Unit tests Windows` is the
@@ -622,8 +639,15 @@ file-scoped worklist and receives no project-wide scanner output.
622
639
  (`tests/clients/interleaving-kit.ts`), which reads a real-clock deadline
623
640
  scaled by `PI_LENS_TEST_TIMEOUT_SCALE`, and give the test a timeout above
624
641
  that deadline. Screen: does the wait end on the event it waits for or on a
625
- real-clock deadline, never on a count? Enforcer: none yet; the detector is
626
- tracked in #4865. Instances and dates: `HISTORY.md`.
642
+ real-clock deadline, never on a count? Enforcer: the governance ast-grep
643
+ rules `turn-count-wait` and `turn-count-wait-test-hook`
644
+ (`scripts/governance-rules/`, run by `npm run astgrep:self-scan`), with a
645
+ reasoned baseline. It does not see a count loop with no exit besides the
646
+ count (a pure `for (N) await yield` flush), a wait spelled as recursion, a
647
+ count written as iteration over a generated range
648
+ (`for (const _ of Array.from({ length: N }))`), or a loop that compares a
649
+ clock value for some other purpose.
650
+ Instances and dates: `HISTORY.md`.
627
651
 
628
652
  </important>
629
653
 
@@ -718,8 +742,27 @@ file-scoped worklist and receives no project-wide scanner output.
718
742
  <reason>`, and a `session` owner states `sessionSeamException`, why a scope
719
743
  cell cannot hold it yet. A moved store has no process container and no
720
744
  entry; `scanScopeCellKeys` pins every cell by file and key
721
- (`SESSION_CELL_KEYS`). Partial: it reads the registered stores only, so an
722
- unregistered module-level field is still not scanned.
745
+ (`SESSION_CELL_KEYS`). The same `auditOwnerKeys` reads all three lists:
746
+ `SESSION_STATE_REGISTRY`, the file-level `EXEMPT_SESSION_STATE_FILES` rows
747
+ and the declared `SESSION_STORE_REGISTRY` rows (#5027); a file with several
748
+ owners states the narrowest. The scan behind them
749
+ (`scanSessionStateCandidates`) flags a file for a module-level container
750
+ with or without a reset, except one admitted constant: an UPPER_SNAKE
751
+ built-in `Map`/`Set`/`WeakMap`/`WeakSet` declared `const` whose every use
752
+ in `clients/`, `tools/`, `mcp/` and `index.ts` is a read; and for a class
753
+ field or `this.x` assignment bound to `new <container or clients/ class>`
754
+ or `[]`, or a depth-1 `key: new <container>` member of a module-level
755
+ object literal (`globalState = { reads: new Map() }`). An admitted
756
+ constant removes the symbol, not the file: a file that also exports a
757
+ reset stays a candidate with a 0 pin. One symbol-count pin counts the
758
+ containers, fields and members. Partial: it walks `clients/`;
759
+ `tools/` holds no candidate today, `index.ts` is censused by
760
+ `tests/config/session-scope-sweep.test.ts` (#3611), and `mcp/server.ts`
761
+ (`lspReadyCwds`) is outside the scan root. A field built by a factory call,
762
+ through another binding or an aliased constructor is not read, and an
763
+ admitted constant's `forEach` callback or stored value can be mutated
764
+ unseen (`SWEEP_HEURISTIC_LIMITS`; live population 0), and the owner value
765
+ itself is a judgment the audit only forces to be stated.
723
766
 
724
767
  87. **Factory-time read of a host value set later:** pi applies an extension flag's command-line value after every factory has run, so a flag read while the factory runs answers the registered default. Screen: does any flag read run while the factory runs, and does the host test apply the value after the factory, as pi does? Read it in a handler, a command or a tool, or once in `session_start`. Enforcer: `tests/config/factory-flag-read-sweep.test.ts` (partial: a reader handed to a callee that invokes it synchronously is not seen). Instances and dates: `HISTORY.md`.
725
768
 
@@ -72,8 +72,10 @@ verdict (a trust check, a lint, a census, a body check), it admits a short list
72
72
  of known-good forms and refuses everything else. It never enumerates known-bad
73
73
  spellings. Each review of an enumerated gate finds the next spelling it missed;
74
74
  one gate took sixteen rounds that way. Once a second round finds a new
75
- spelling, stop patching and redesign. Screen: can an input nobody planned for
76
- pass the gate?
75
+ spelling, stop patching and redesign. A value the gate cannot resolve to a
76
+ known-good literal (computed, concatenated, aliased, templated, reached through
77
+ an indirection) is unrecognized: it counts as a hit, never a pass. Screen: can
78
+ an input nobody planned for pass the gate?
77
79
 
78
80
  **Blast radius as a call-tree diff.** For any change to a shared seam, list the
79
81
  changed symbol, its callers above, its callees below, with plus and minus on the
@@ -115,6 +117,13 @@ it, and paste the red. A checked box with no transcript is treated as false. A
115
117
  mutation that fails to compile proves nothing about the test; redo it as a
116
118
  compile-valid change. A guard that cannot be made to red does not need to exist.
117
119
 
120
+ **A precision figure names its judgement and an outside sample.** A claim that
121
+ a rule, check or heuristic is N% precise states what counts as a true positive,
122
+ and someone other than its author judges a random sample. "The pattern matched"
123
+ is not "the finding is worth acting on": one rule reported 611 of 611 hits real
124
+ on a type check, and an independent sample of 58 found none worth the rewrite.
125
+ Below the stated bar, the rule ships disabled with the count recorded.
126
+
118
127
  **Tautological tests are harmful.** A test observes behaviour through the real
119
128
  seam and asserts an independent effect. It never restates the implementation,
120
129
  never asserts a value its own setup supplied, and never swaps a real in-process
@@ -138,6 +147,13 @@ thing is required, a registry entry, a baseline row and lane membership. A
138
147
  mocked seam is not a real spawn and gets an exemption with its reason, never an
139
148
  admission.
140
149
 
150
+ **A gate judges the actor's delta, not inherited state.** A check run on an
151
+ artifact someone edits (a tracker body, a shared file, a long-lived record)
152
+ judges only what that edit adds or changes. Inherited violations belong in a
153
+ baseline that shrinks on purpose. Refusing an honest edit for debt it did not
154
+ create blocks the work or invites routing around the gate. If the gate cannot
155
+ compute the delta, it checks the whole artifact, failing closed.
156
+
141
157
  **A new ratchet sweeps the open changes before it lands.** A governance test
142
158
  that rejects a new pattern fails every open change that already adds one, and
143
159
  none of them conflict textually, so nothing warns. Before merging a ratchet,
@@ -219,6 +235,15 @@ cleanup, triage and status upkeep. The script surfaces unfinished work and
219
235
  never hides it: stalled work is restarted, not unassigned, archived or deleted.
220
236
  Screen: is this the second time this step was done by hand?
221
237
 
238
+ **Process carries the net-count rule too.** A new guard, hook, script, lint
239
+ rule, ratchet or scheduled routine retires one of the same kind in the same
240
+ change, or it cites fire data showing what the existing ones miss. Each rule
241
+ must also earn its keep: count how often it fired, and what each fire led to
242
+ (a real fix, a wording edit, or a route-around). One that never fired in its
243
+ window and names no recurrence is demoted or deleted. Process cost is paid on
244
+ every change, while the bug a rule prevents may never arrive. Screen: what
245
+ does this rule retire, and when did the last one like it fire?
246
+
222
247
  ## 4. Honesty and records
223
248
 
224
249
  **Closes versus refs follows delivery, not optimism.** Close only when every
package/docs/features.md CHANGED
@@ -164,7 +164,7 @@ instead of a flat blob of source. See
164
164
  [`module-report-read-symbol.md`](module-report-read-symbol.md)
165
165
  for the full design and token-efficiency numbers.
166
166
 
167
- **`module_report(filePath, maxRefsPerSymbol?, focus?, view?, blastRadius?, blastRadiusDepth?)`** —
167
+ **`module_report(path, maxRefsPerSymbol?, focus?, view?, blastRadius?, blastRadiusDepth?)`** —
168
168
  returns a structured outline: every symbol's name/kind/startLine/endLine/signature
169
169
  (plus a first-line `doc` summary when a doc comment is attached — #512),
170
170
  exported vs internal split (with class/interface members nested under their
@@ -184,7 +184,7 @@ instead of JSON — same data, opt-in. Tree-sitter extract + cached review-graph
184
184
  lookup; never builds the graph, never calls LSP on this path.
185
185
  `semantic.source` reports what backed the data (`review-graph` | `none`).
186
186
 
187
- **`read_symbol(filePath, symbol)`** — returns the verbatim body of one
187
+ **`read_symbol(path, symbol)`** — returns the verbatim body of one
188
188
  named symbol plus a one-line header (`<kind> <name> <basename>:<startLine>-<endLine>`).
189
189
  Records the read against the read-guard so a follow-up edit anywhere in
190
190
  that symbol's range passes the read-before-edit check (an outline from
@@ -44,9 +44,13 @@
44
44
  The commit gate reads every record, whoever owns it, because the tree is
45
45
  shared. A turn end reads only its own session's records: it delivers,
46
46
  sweeps, re-judges and consumes those, and files its guard section under its
47
- own id. Only a write by the coordinator's session latches
48
- `session_mismatch`; a concurrent session's write leaves another
49
- session's record and the latch alone.
47
+ own id. The gate reads every record but retires only its own session's: its
48
+ missing-file sweep (#5020) persists into the cache of the acting session's
49
+ own records, whose cache is the coordinator root's cache or the session
50
+ cwd's, whichever wrote them, so another session's dead-cite row waits for
51
+ that session's own gate or turn end. Only a write by the coordinator's
52
+ session latches `session_mismatch`; a concurrent session's write leaves
53
+ another session's record and the latch alone.
50
54
  - The commit gate is keyed by the repository being committed (#4941), not by
51
55
  session. A commit/push is judged by the unresolved blockers whose files lie
52
56
  in the git checkout it targets (`resolveGitCheckout`, the seam the review
@@ -19,10 +19,21 @@
19
19
  legacy migrations, and namespaces live in the config-location/schema modules.
20
20
  A new config key or environment flag needs a forcing function, stability tier,
21
21
  diagnostic code, tests, and docs.
22
- - `rules.<id>.ignorePaths` is an experimental, project-relative path denial
23
- resolved by config-core with array-union semantics: global entries cannot be
24
- cleared by project config. Ast-grep and tree-sitter apply it before scanning;
25
- other runner output uses the shared rule-policy filter.
22
+ - `rules.<id>.ignorePaths` is an experimental path denial resolved by
23
+ config-core with array-union semantics: global entries cannot be cleared by
24
+ project config. Each entry is a gitignore line, matched as the `ignore` field
25
+ matches (#783: last match wins, `!` re-includes, a leading `/` roots it, a
26
+ trailing `/` is directory-only) through `isIgnoredByGitignoreLines`
27
+ (`clients/file-utils.ts`), relative to the directory of the file that
28
+ declares it (the project root for a global entry). Each file is its own
29
+ scope and a path is carved out when any scope ignores it, so a `!` never
30
+ lifts another file's carve-out. A rule's own YAML `ignores:` stays on the
31
+ ast-grep glob matcher (`isRuleIgnoredForPath`), which mirrors
32
+ `ast-grep scan`. Ast-grep and tree-sitter apply `ignorePaths` before
33
+ scanning; other runner output uses the shared rule-policy filter.
34
+ - `rules` and `dispatch.disabledRunners` are read per file from its config
35
+ owner (`resolveDispatchPolicyConfigForFile`, `loadProjectRulePolicy`), so a
36
+ nested `.pi-lens.json` governs its own files on every surface.
26
37
  - `lens_diagnostics` has one model-facing diagnostic surface. `source` is
27
38
  `session` or `lsp`; `scope` is `paths` or `workspace`; explicit paths always
28
39
  win. Severity is a threshold. Retired compatibility names must not widen a
@@ -102,8 +102,9 @@
102
102
  timers on every removal path.
103
103
  - Production callers use the grouped experimental `LspCapabilities` adapter in
104
104
  `clients/lsp/capabilities.ts`; direct `LSPService` module imports are limited
105
- to that adapter and are enforced by
106
- `tests/config/lsp-capabilities-import-sweep.test.ts` (#2372/#277).
105
+ to that adapter and are enforced by oxlint's `no-restricted-imports` in
106
+ `.oxlintrc.json` (`npm run lint:js`, #2372/#277; it replaced the
107
+ `lsp-capabilities-import-sweep` test in #4865).
107
108
  - Idle-eviction policy is the registry's `idleEviction` field, declared per
108
109
  server. The nightly (`scripts/measure-lsp-idle-eviction.mjs`) measures every
109
110
  registry server's eviction cost and respawn safety into
@@ -187,6 +188,19 @@
187
188
  - Per-path LSP notifications serialize read/build/send/record work. Pull
188
189
  cancellation blocks a same-path replacement until settlement. Waits are
189
190
  deadline- and abort-bounded, and silence is never clean.
191
+ - A cached answer is evidence about the state it was published for, never about a
192
+ later change. `touchFile` skips a notify for bytes the server already holds
193
+ (see the debounce bullet below), so the diagnostics it reports can be an
194
+ entry the server published before the caller's edit; it names that case
195
+ (`TouchFileResult.reusedAnswer`, server plus publish stamp) and never claims
196
+ it as the touch's own answer. The cascade reads it (`isPreEditAnswer`,
197
+ `clients/dispatch/integration.ts`): an answer published before the run's edit
198
+ is unconfirmed, never clean, exactly as `isPostEditObservation` already rules
199
+ for the passive snapshot and both fallbacks. An answer published after the
200
+ edit (a push or pull inside the touch, the `typescript.tsserverRequest` sync
201
+ confirm, or a server that republished for the file after the edit) keeps its
202
+ verdict. A client that cannot date its cache reports no reuse and keeps the
203
+ pre-#5040 verdict.
190
204
  - Document line addressing has one owner, `clients/lsp/content-scan.ts`: the
191
205
  `\n` count behind the `lsp_document_send` record's `contentLineCount` and the
192
206
  LSP sync bound (`countLfNewlines`), the last LSP-addressable line an
@@ -15,8 +15,10 @@
15
15
  base expansion) or `resolveToolCallFilePath` (base plus pi's variant
16
16
  ladder), never a bare `path.resolve(base, raw)`; see shape 74.
17
17
  - Windows-shaped paths use `path.win32` functions. Use `toPosix`,
18
- `splitPathSegments`, `isUnderDir`, `isSameOrWithin`, and
19
- `isAtOrAboveHomeDir` instead of inline separator or containment logic.
18
+ `splitPathSegments`, `isUnderDir`, `isSameOrWithin`, `relativeInside`
19
+ (lexical containment with the relative text), `isWithinIdentity`
20
+ (containment on the identity axis), and `isAtOrAboveHomeDir` instead of
21
+ inline separator or containment logic.
20
22
  - Data and log locations follow **Data directories and logs**; never hardcode
21
23
  `.pi-lens` paths in writes or user-facing text.
22
24
  - Probes and child processes pin `PI_LENS_HOME`, `PILENS_DATA_DIR`, `HOME`, and
@@ -129,7 +129,7 @@ belonged to another live client of the same server id became `typescript`'s
129
129
  | **2 — push, publishes-versioned** | `publishDiagnostics([])` **with version** on every scan, incl. clean→clean | YES, currency-proven via version | ast-grep |
130
130
  | **2\* — push, publishes-unversioned** | re-publishes on a clean scan but **version-less** — the wait still early-returns (the client accepts a version-less publish as fresh: it can't be proven stale), but currency is only *temporally correlated*, not proven. Since #3484 the client fences a server measured to answer a `documentSymbol` fence before it publishes (`diagnosticsFence: "reply-first"`: yaml, intelephense): a pre-edit publish is dropped until the fence's reply. Every other 2\* server (docker publishes before answering; prisma, taplo, zls, dart, gleam, clojure unmarked) keeps this behaviour | YES at runtime, with a staleness-risk caveat (not a latency cost) | opengrep |
131
131
  | **3 — push, silent on clean** | server publishes nothing when nothing changed | **NO** — budget-wait floor (safe; a timeout is *not* a false clean). **This tier is #458's learned-deadline target set.** | typescript-language-server |
132
- | **Navigation-only — custom, no evidence** | custom `lsp.servers.*` entry has no pull provider and has not published in this session | **NO** — diagnostics are unsupported; skip the wait and report navigation-only | Dexter |
132
+ | **Unconfirmed — custom, no evidence** | custom `lsp.servers.*` entry has no pull provider and has not published in this session | **NO** — diagnostics are unconfirmed (reported `inconclusive`, never `unsupported`); skip the wait until the server publishes | Dexter, groovy-language-server |
133
133
 
134
134
  Detection is **cached** at `initialize` (`detectWorkspaceDiagnosticsSupport` →
135
135
  `state.workspaceDiagnosticsSupport.mode`, upgraded on `client/registerCapability`),
@@ -139,8 +139,22 @@ Custom servers without a pull provider begin with one bounded first-contact
139
139
  probe. The probe uses the smaller of the server's push-wait budget and the live
140
140
  hook deadline. If the hook cuts it off first, the result remains unconfirmed and
141
141
  the next touch probes again. Only silence through the full push budget latches
142
- that server id as navigation-only for the service session. A publish upgrades it
143
- to the ordinary push-wait policy.
142
+ that server id for the service session: the latch skips the wait and the file
143
+ reads **unconfirmed** (`inconclusive` in a batch outcome, "not the same as 0
144
+ diagnostics" in a single-file result), never `unsupported` and never clean. Any
145
+ publish, however late and whatever file it is for, clears the latch (#5128), and
146
+ the server returns to the ordinary push-wait policy. A server that publishes
147
+ only for files with errors (groovy-language-server sends nothing for a file that
148
+ compiles) therefore reads unconfirmed on clean files, and its findings arrive as
149
+ soon as it publishes them. A publish that lands while the first-contact wait is
150
+ still in flight prevents the latch.
151
+
152
+ The first-contact budget on the per-edit path is the strategy's
153
+ `aggregateWaitMs` (1500 ms). It is a cost, not a verdict: the latch is
154
+ recoverable, so a server slower than the budget pays 1.5 s once and reads
155
+ unconfirmed until its first publish, instead of being lost for the session.
156
+ Raising it would charge up to the larger budget on the first edit of every
157
+ custom server that never publishes for a clean file.
144
158
 
145
159
  ## Matrix (dev box + CI nightly; mode last refreshed 2026-06-17 from run 27713958681, clean-behavior probed in run 35914033696 — #460)
146
160
 
@@ -328,55 +342,121 @@ and were characterized `push-only` in the run above).
328
342
 
329
343
  Population: every server that declares `textDocumentSync.save` in
330
344
  `docs/servercapabilities.md` (15 `save`, 5 `save+text`), plus zizmor, which
331
- registers `textDocument/didSave` dynamically. Verdicts are moved from the #4768
332
- PR body (merged as 11cf30f8d) and re-read on origin/master d21f7d267: each
333
- server's registry entry is in `clients/lsp/server.ts`, and each save declaration
334
- matches `docs/servercapabilities.md`. `publishes after save only` is `no` when
335
- the server publishes without a save, `yes` when only a save publishes, and
336
- `unknown` when nothing was measured. A `source` row cites an external
337
- repository at the pinned commit; those commits were not re-fetched here.
345
+ registers `textDocument/didSave` dynamically and therefore never receives one
346
+ from pi-lens (its `didSave` registration is dynamic and pi-lens maps only the
347
+ static `initialize` reply). The `server` column is the registry id in
348
+ `clients/lsp/server.ts`, so a row can be matched against the capability
349
+ inventory mechanically.
350
+
351
+ `publishes after save only` states whether only a save makes the server answer:
352
+
353
+ | value | meaning |
354
+ |---|---|
355
+ | `no` | the seeded finding arrived without a save |
356
+ | `no (pull)` | pull mode: the `textDocument/diagnostic` request answered without a save, so no publish is involved |
357
+ | `yes` | the same bytes produced the finding only after a save |
358
+ | `yes (saved write)` | only a diagnostic-neutral edit plus a save produced it |
359
+ | `inconclusive` | no arm produced the finding, or the answer landed in the untriggered window between two arms and so belongs to no arm |
360
+ | `cannot measure` | no runner can install the server; the reason follows |
361
+ | `unknown` | no run has measured this row yet |
362
+
363
+ Regenerate it with
364
+
365
+ ```text
366
+ node scripts/probe-save-verdict.mjs [lang ...] [--install] [--wait-ms <ms>] [--doc <path>]
367
+ ```
368
+
369
+ The nightly `tool-smoke.yml` step "LSP save verdict (#3405)" runs it with
370
+ `--install` on the tool-smoke fixtures, so the measurement columns self-populate
371
+ through the same docs-refresh PR as `mode` and `clean-behavior` (#390). The probe
372
+ drives the real client and reports four arms of equal length: the seeded bytes
373
+ with `saved: false` (didOpen/didChange only), an untriggered window of the same
374
+ length, the SAME bytes with `saved: true` (so a publish has the save as its only
375
+ new input), and, only when nothing else answered, a diagnostic-neutral edit
376
+ plus a save. `scripts/lib/save-verdict.mjs` owns the classification, the wording
377
+ and the merge guard; a run rewrites the three measurement columns of the rows it
378
+ reached and copies every other row verbatim, so a runner that could not install a
379
+ server never blanks a measured cell.
380
+
381
+ A night that could not tell never erases an answer (#5060). A row that holds a
382
+ settled answer (`no`, `no (pull)`, `yes`, `yes (saved write)`) keeps it, with its
383
+ evidence, when the run measures `inconclusive` or `cannot measure`; only a
384
+ settled answer replaces a settled one, and a `yes` to `no` change stays a visible
385
+ flip. This table has no column for a night a row was not answered, so the held
386
+ night goes to the run log (`save-verdict: held back <server>`), to one
387
+ `::warning::` annotation per held server under GitHub Actions, and to the
388
+ `--summary` JSON (`heldBack`). No counter or issue tracks consecutive held
389
+ nights.
390
+
391
+ A pull-mode server answers a `textDocument/diagnostic` request rather than
392
+ publishing, so no arm of its run has the save as its only new input: it reads
393
+ `no (pull)` when the request answered without a save and `inconclusive`
394
+ otherwise, never `yes`.
395
+
396
+ The cells carry what a run OBSERVED, never when or where: each arm's elapsed time
397
+ and the `src` token that says CI or a dev box go to the run log, so a
398
+ re-measurement that changes only those is not a capability flip and the refresh PR
399
+ stays armed for auto-merge (#4512).
400
+
401
+ A row whose evidence names this command is a measurement by it. A row that names
402
+ a prior probe, a workflow run or a source commit instead is provenance from
403
+ before the command existed (or from a server no local run could reach), and the
404
+ dispatched nightly replaces it on the first night that reaches the server.
338
405
 
339
406
  | server | save as declared | publishes after save only | verdict | evidence |
340
407
  |---|---|---|---|---|
341
- | expert | `save` | no | Measured: answers with no `didSave` once the seed is under `lib/`. Gated (`lspGate: true`). | #4768 arms A, B, C; nightly run 39852672584 (answered on call 6, 10955 ms); `tests/fixtures/nightly-floors/expert-readiness.json` |
342
- | cpp (clangd 22.1.0) | `save` | no | Measured: publishes on `didOpen` alone (97 ms, 3 diagnostics). Gated. Open conflict with #4527, see note 2. | #4768 local probe; #4527 wire scenario (issue comment) |
343
- | lua | `save` | no | Measured: the first publish carries 6 diagnostics at 135 ms; a save adds a late republish at 3.7 s. Gated. | #4768 local probe |
344
- | python-jedi | `save` | no | Measured: publishes on `didOpen` alone (about 1 s). Gated. | #4768 local probe; `clients/python-environment.ts:421` (id) |
345
- | opengrep (aux) | `save` | no | Measured: a save rescans and republishes (#3482); it also publishes on open once its rules load. Auxiliary, not gated. | `clients/lsp/wait-policy/strategies.ts:100`; Key findings above (#4763) |
346
- | gleam | `save` | no | Source: a save only marks the project changed and drops the in-memory cache; the publish comes from `CompilePlease`. Gated. | gleam-lang/gleam `0d9fa18ddb69`, `language-server/src/server.rs` (`SourceFileSaved`, `discard_in_memory_cache`, `compile_please`) |
347
- | go (gopls) | `save` | no | Source: `DidSave` calls `didModifyFiles`, the path a change takes. Gated. | golang/tools `02c9e4cb4c2f`, `gopls/internal/server/text_synchronization.go` |
348
- | zig (zls) | `save` | no | Source: a save runs the build-on-save watch update and autofix; ast-check diagnostics come from changes. Gated. | zigtools/zls `82ba5351ca8a`, `src/Server.zig` (`saveDocumentHandler`) |
349
- | rust (rust-analyzer) | `save` | no | Source: a save starts flycheck (`cargo check`); native diagnostics come from pull mode and do not wait for it. Exempt: runner availability. | rust-lang/rust-analyzer `0c4d74686b21`, `crates/rust-analyzer/src/handlers/notification.rs` (`handle_did_save_text_document`); matrix tier 1 |
350
- | terraform | `save` | no | Source: `terraform validate` runs on save only with the experimental `ValidateOnSave`, off by default. Gated. | hashicorp/terraform-ls `3dbe85a2b4e1`, `internal/langserver/handlers/did_save.go` |
408
+ | expert | `save` | no | Prior measurement: answers with no `didSave` once the seed is under `lib/`. Gated. | #4768 arms A, B, C; nightly run 39852672584 (answered on call 6, 10955 ms); `tests/fixtures/nightly-floors/expert-readiness.json` |
409
+ | cpp | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture cpp) |
410
+ | lua | `save` | no | Measured: answers without a save (2 publish events on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture lua) |
411
+ | python-jedi | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture jedi) |
412
+ | opengrep | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Auxiliary, not gated. | `node scripts/probe-save-verdict.mjs` (fixture opengrep) |
413
+ | gleam | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture gleam) |
414
+ | go | `save` | no | Source: `DidSave` calls `didModifyFiles`, the path a change takes. Gated. | golang/tools `02c9e4cb4c2f`, `gopls/internal/server/text_synchronization.go` |
415
+ | zig | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture zig) |
416
+ | rust | `save` | inconclusive | Measured: inconclusive — pull mode: every arm is a textDocument/diagnostic request, so no arm has the save as its only new input (arms: no-save silent (6000 ms), save alone silent (6000 ms), saved write silent (6000 ms)). Exempt from the clean gate. | `node scripts/probe-save-verdict.mjs` (fixture rust) |
417
+ | terraform | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture terraform) |
351
418
  | fsharp | `save+text` | no | Source: a save holds the source and refreshes code lenses; diagnostics come from open and change checks. Exempt: 0 findings in every run. | ionide/FsAutoComplete `8f011f1395a4`, `src/FsAutoComplete/LspServers/AdaptiveFSharpLspServer.fs` (`TextDocumentDidSave`); nightly run 39845687189 |
352
- | zizmor (dynamic) | not declared (`textDocumentSync` only) | no | Source: `didSave` is registered dynamically and pi-lens does not map that registration, so no save is sent. Auxiliary. | zizmor `crates/zizmor/src/lsp.rs` (`initialized`); `clients/lsp/wait-policy/strategies.ts:400-410` (#3548) |
353
- | clojure | `save+text` | unknown | Unverified: not measured. Gated; one save per write. | none |
354
- | fish | `save+text` | unknown | Unverified: not measured. Gated; one save per write. | none |
355
- | tinymist | `save` | unknown | Unverified: not measured. Gated (typst fixture); one save per write. | none |
356
- | cue | `save` | unknown | Unverified: not measured. Gated; one save per write. | none |
357
- | csharp | `save+text` | unknown | Unverified: not measured. Pull mode, so diagnostics are requested and a save is unlikely to be the only trigger. Gated. | matrix row (csharp-ls, tier 1) |
358
- | deno | `save` | unknown | Unverified: not measured. Pull mode. Gated. | matrix row (deno, tier 1) |
359
- | svelte | `save` | unknown | Unverified: not measured. Pull mode. Gated. | matrix row (svelte-language-server, tier 1) |
419
+ | zizmor | · (registered dynamically) | no | Source: `didSave` is registered dynamically and pi-lens does not map that registration, so no save is sent. Auxiliary, not gated. | zizmor `crates/zizmor/src/lsp.rs` (`initialized`); `clients/lsp/wait-policy/strategies.ts:400-410` (#3548) |
420
+ | clojure | `save+text` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture clojure) |
421
+ | fish | `save+text` | unknown | Unverified: not measured. Gated. | none |
422
+ | tinymist | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: 1 publish event. Gated. | `node scripts/probe-save-verdict.mjs` (fixture typst) |
423
+ | cue | `save` | no | Measured: answers without a save (1 publish event on didOpen); a later save alone: silent (6000 ms). Gated. | `node scripts/probe-save-verdict.mjs` (fixture cue) |
424
+ | csharp | `save+text` | unknown | Unverified: not measured. Pull mode, so diagnostics are requested and a save is unlikely to be the only trigger. Gated. | none |
425
+ | deno | `save` | no (pull) | Measured: pull mode — no save needed; a later save+request answered: 1 diagnostic. Gated. | `node scripts/probe-save-verdict.mjs` (fixture deno) |
426
+ | svelte | `save` | no (pull) | Measured: pull mode — no save needed; a later save+request answered: 2 diagnostics. Gated. | `node scripts/probe-save-verdict.mjs` (fixture svelte) |
360
427
  | powershell | `save+text` | unknown | Unverified: not measured. Exempt: intermittent. | none |
361
- | cmake | `save` | unknown | Unverified: the local probe published nothing in either arm (12 s each) with an `if(` seed, so inconclusive. Exempt: upstream documents no diagnostics. | #4768 local probe (no named command) |
428
+ | cmake | `save` | inconclusive | Measured: inconclusive — no arm published, so this fixture does not distinguish a save-triggered server from one that never diagnoses it (arms: no-save silent (6000 ms), save alone silent (6000 ms), saved write silent (6000 ms)). Exempt from the clean gate. | `node scripts/probe-save-verdict.mjs` (fixture cmake) |
362
429
 
363
430
  Notes.
364
431
 
365
- 1. Probe. The only command #4768 names is `node scripts/smoke-tools.mjs --lsp-gate`
366
- (the nightly `tool-smoke.yml` step "LSP diagnostics clean-gate"). It runs
367
- the gated fixture rows and does not write per-server save verdicts. The local
368
- probes behind the cpp, lua, python-jedi and cmake rows, and the Expert arms
369
- run outside CI, are not named in #4768's body, so no command regenerates this
370
- table. Each row cites its own evidence instead.
371
- 2. Open conflict, cpp. #4527 records a clangd 22.1.0 wire scenario in which the
372
- save sender changes the outcome (base 1 of 5, head 5 of 5). #4768 records clangd
373
- publishing on `didOpen` alone. The two records are not reconciled here; the row
374
- keeps #4768's verdict.
432
+ 1. Probe. `node scripts/probe-save-verdict.mjs` is the command that produces this
433
+ table (see the paragraph above). It reuses the tool-smoke fixtures and the
434
+ nightly that already installs those servers, so it is one measurement path, not
435
+ a second driver. The only other command #4768 named is
436
+ `node scripts/smoke-tools.mjs --lsp-gate`; that step drives the clean gate and
437
+ writes no save verdict.
438
+ 2. clangd reconciled. #4527 recorded "base 1 of 5, head 5 of 5" and #4768 recorded
439
+ clangd publishing on `didOpen` alone; both hold, and they counted different
440
+ things. The 5 checks are the rows of the `lsp-did-save/analyze` verify scenario,
441
+ four of which require a `textDocument/didSave` on the wire, so its one passing
442
+ check on base is `control: clangd answered for a.c`: clangd published the
443
+ finding with `0 didSave(s)` captured. The sender was missing, not the
444
+ publish. The probe reproduces both arms with one seed and one window (6000 ms
445
+ each, one client): the no-save arm published on `didOpen`, and the save arm,
446
+ with the identical bytes and the save on the wire, published nothing, because
447
+ clangd does not republish identical bytes, the same property #4527 measured
448
+ when its own re-push left the cold query inconclusive. So #4768's verdict
449
+ stands, #4527's 1-of-5 → 5-of-5 is a sender measurement, and no clangd arm
450
+ reads `yes`.
375
451
  3. Citations corrected against master. #4768 cites `clients/lsp/client.ts:4410`
376
452
  for `sendDidSave`; it is at `clients/lsp/client.ts:4909`, with the save-capability
377
453
  gate at line 4917. The opengrep rescan claim is at
378
454
  `clients/lsp/wait-policy/strategies.ts:100`; line 345 is the
379
455
  `emptyUntilRulesRefreshed` hold.
456
+ 4. Clean-gate policy in a measured row is the fixture's own declaration, not
457
+ hand prose: `Gated.` is `lspGate: true`, `Auxiliary, not gated.` an auxiliary
458
+ scanner, and `Exempt from the clean gate.` a fixture whose `lspGateExempt`
459
+ reason lives in `scripts/smoke-tools.mjs`.
380
460
 
381
461
  ## Completing the matrix
382
462
  The fixtures (`tests/fixtures/tool-smoke/<lang>/`) are durable and cover every
@@ -16,7 +16,7 @@ neither is needed. Sharded Unit tests (#3753) keep one queue run short.
16
16
  Together they produce every required check: `Lint & type-check`,
17
17
  `Unit tests`, `Install test (ubuntu-latest)`, `Install test (windows-latest)`,
18
18
  `Install test (macos-latest)`, `knip`, `oxfmt format check`, `TLA+ models`.
19
- - PR-only jobs (PR title, PR body, changelog fast-fail, targeted advisory) are
19
+ - PR-only jobs (PR title, PR body, changelog fast-fail, changed tests shuffled) are
20
20
  guarded by `github.event_name == 'pull_request'` and are skipped on
21
21
  `merge_group`; none of them is required.
22
22
  - `tests/config/merge-queue-workflows.test.ts` pins the trigger, the required
@@ -41,8 +41,9 @@ registry; other doubles only as `AGENTS.md` admits them.
41
41
  compose, and flag merge order in the PR body. Branch `fix/<N>-<slug>` from
42
42
  `origin/master` unless the brief names a branch. Preserve contributor
43
43
  authorship on a contributor's branch.
44
- - Write the failure list in the PR body before the first edit (principles §2):
45
- inputs, states, orderings, and platforms; a space with two axes is a table.
44
+ - Write the failure list under `### Tests` in the PR body before the first edit
45
+ (principles §2): inputs, states, orderings, and platforms; a space with two
46
+ axes is a table.
46
47
  The tests cover that list, not only the happy path. For a change to N of M
47
48
  members, the list also names M, the excluded default, and the planned
48
49
  generalization verdict: *widen in this PR*, *follow-up issue* with its seam
@@ -85,7 +86,7 @@ registry; other doubles only as `AGENTS.md` admits them.
85
86
  tests stay green; put any behaviour change in a separate commit (#3817).
86
87
  - Delete vacuous tests in the files you touch (a case that reds on no
87
88
  mutation, asserts a constant, or duplicates a sibling), with the sweep
88
- transcript quoted in `Test assessment`. A redundant test that still guards
89
+ transcript quoted in `### Tests`. A redundant test that still guards
89
90
  goes only with the named survivor that covers it.
90
91
  - Apply folded review follow-ups per `docs/pi-lens-subagent.md`
91
92
  "Follow-ups", and list every gap left open in one **Accepted residuals** section of
@@ -232,7 +233,8 @@ registry; other doubles only as `AGENTS.md` admits them.
232
233
  (#2654).
233
234
  - Spot-check one prior round's mutation on the new head (#2583).
234
235
  - Re-read the changelog fragment for any claim the round retracts (#3155), then
235
- add an honest review-round section to the PR body.
236
+ update the PR body in place so it describes the current head. Do not append a
237
+ round section; commits and review threads hold the history.
236
238
  - Lint the PR body (`--lint-local`) before the first push of a round, not only
237
239
  at hand-back: `PR body (advisory)` re-reads the same body on every push, so
238
240
  an unfixed finding reds each one.
@@ -247,19 +249,24 @@ rigor (#2599).
247
249
 
248
250
  ## Handoff
249
251
 
250
- - The PR body is the whole `.github/PULL_REQUEST_TEMPLATE.md`, every
251
- heading present in order: `## Why` (one sentence), `## Notes for the
252
- reviewer`, `## Change outline`, `## Summary`, `## Type of change`,
253
- `## Area`, `## Checklist`, `## Tests`, `## Blast radius`,
254
- `## Observability` (a record literal from the runtime diff, `none: <reason>`,
252
+ - The PR body follows `.github/PULL_REQUEST_TEMPLATE.md`, which a reader must
253
+ be able to scan (#5003): `## Summary` (at most three sentences, with the
254
+ closing keyword, which GitHub ignores in a title), `## Notes for the reviewer`
255
+ (at most five bullets, no tables or fences), `## Change outline`, and
256
+ `## Proof and checks` with `### Tests`, `### Class sweep` (ending in one
257
+ `Alignment:` line that names the families of `docs/alignment-families.md` the
258
+ change touches, or `none: <reason>` of 12+ characters),
259
+ `### Observability` (a record literal from the runtime diff, `none: <reason>`,
255
260
  or exactly `No new failure path; no record added.`, which a new decision
256
- branch on a session, lifecycle or delivery seam refuses, #3875),
257
- `## Class sweep`, `## Fold check` (one line per candidate, `#N folded`,
258
- `#N not folded` or `#N sequenced`, then its reason, or `none: <reason>`;
259
- the body lint owns the form and the local script owns the candidate list,
260
- #4711), and `## Test assessment`. A brief that names only some
261
- headings does not shorten this list. The closing keyword lives in the body;
262
- GitHub ignores it in a title.
261
+ branch on a session, lifecycle or delivery seam refuses, #3875), and
262
+ `### Fold check` (one line per candidate, `#N folded`, `#N not folded` or
263
+ `#N sequenced`, then its reason, or `none: <reason>`; the body lint owns the
264
+ form and the local script owns the candidate list, #4711). Add `### Blast
265
+ radius` only when the change reaches outside the touched module or hits a hot
266
+ path. The lint caps visible text at 5000 characters and each code fence at 20
267
+ lines: quote failing lines, not whole runs, and give each mutation one table
268
+ row, not a paragraph. A brief that names only some headings does not shorten
269
+ this list.
263
270
  - Run `node scripts/check-pr-body.mjs --lint-local PR_BODY.md` and
264
271
  `node scripts/check-changelog-fragments.mjs` before the hand-back; both
265
272
  must pass, and the hand-back quotes them. The body lint reads the closed
@@ -272,7 +279,7 @@ rigor (#2599).
272
279
  reference written as `owner/repo#N` or a URL is refused unless the body
273
280
  already closes a bug, which owes the section whatever else it cites. CI refuses a label
274
281
  it cannot read; the local preflight, with no working gh, warns and skips it
275
- as for a closed issue (#4968). A non-bug PR either deletes the `## Detection` section or
282
+ as for a closed issue (#4968). A non-bug PR either deletes the `### Detection` section or
276
283
  fills the three lines, and a `Shape:` or `Enforcer:` line it does write is
277
284
  validated too; free text in between reds the local preflight. A bug-closing
278
285
  PR that answers a layer, the `Gap: none:` reason, the `Shape: new:` name or