| 1 | #!/usr/bin/env node |
| 2 | // Language-invariant coverage floor for the compaction survival contract. |
| 3 | // Later TypeScript strategies should keep this check; Rust remains the B1 |
| 4 | // enforcement path. Run: node validate_survival_contract.mjs |
| 5 | |
| 6 | import { readFileSync } from "node:fs"; |
| 7 | import { dirname, join } from "node:path"; |
| 8 | import { fileURLToPath } from "node:url"; |
| 9 | |
| 10 | const MARKERS = [ |
| 11 | "Another language model started to solve this problem", |
| 12 | "Conversation Summary (Auto-Generated)", |
| 13 | ]; |
| 14 | |
| 15 | const root = dirname(fileURLToPath(import.meta.url)); |
| 16 | const matrix = JSON.parse( |
| 17 | readFileSync(join(root, "fixtures/matrix.json"), "utf8"), |
| 18 | ); |
| 19 | |
| 20 | function userTextOf(message) { |
| 21 | if (message.role !== "user") return null; |
| 22 | const text = (message.content ?? []) |
| 23 | .filter((block) => block.type === "text") |
| 24 | .map((block) => block.text) |
| 25 | .join("\n") |
| 26 | .trim(); |
| 27 | return text || null; |
| 28 | } |
| 29 | |
| 30 | function isCheckpoint(message) { |
| 31 | const text = userTextOf(message); |
| 32 | return Boolean(text && MARKERS.some((marker) => text.includes(marker))); |
| 33 | } |
| 34 | |
| 35 | function isPlainUserText(message) { |
| 36 | return !isCheckpoint(message) && Boolean(userTextOf(message)); |
| 37 | } |
| 38 | |
| 39 | function lastPlainUserIndex(messages, end) { |
| 40 | for (let idx = end - 1; idx >= 0; idx -= 1) { |
| 41 | if (isPlainUserText(messages[idx])) return idx; |
| 42 | } |
| 43 | return null; |
| 44 | } |
| 45 | |
| 46 | function sliceHasToolResult(messages, start) { |
| 47 | return messages.slice(start).some((message) => |
| 48 | (message.content ?? []).some((block) => block.type === "tool_result"), |
| 49 | ); |
| 50 | } |
| 51 | |
| 52 | export function lastRoundStart(messages) { |
| 53 | const lastUser = lastPlainUserIndex(messages, messages.length); |
| 54 | if (lastUser === null) return 0; |
| 55 | if (sliceHasToolResult(messages, lastUser)) return lastUser; |
| 56 | let candidate = lastUser; |
| 57 | for (;;) { |
| 58 | const prev = lastPlainUserIndex(messages, candidate); |
| 59 | if (prev === null) return lastUser; |
| 60 | if (sliceHasToolResult(messages, prev)) return prev; |
| 61 | candidate = prev; |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | function toolResultIds(message) { |
| 66 | return (message.content ?? []) |
| 67 | .filter((block) => block.type === "tool_result") |
| 68 | .map((block) => block.tool_use_id); |
| 69 | } |
| 70 | |
| 71 | function toolUseIds(message) { |
| 72 | return (message.content ?? []) |
| 73 | .filter((block) => block.type === "tool_use") |
| 74 | .map((block) => block.id); |
| 75 | } |
| 76 | |
| 77 | function isAssistantLike(message) { |
| 78 | return message.role === "assistant" || message.role === "assistant_interrupted"; |
| 79 | } |
| 80 | |
| 81 | function assistantTextOf(message) { |
| 82 | if (!isAssistantLike(message)) return null; |
| 83 | const text = (message.content ?? []) |
| 84 | .filter((block) => block.type === "text") |
| 85 | .map((block) => block.text) |
| 86 | .join("\n") |
| 87 | .trim(); |
| 88 | return text || null; |
| 89 | } |
| 90 | |
| 91 | // A retained copy may be truncated, so a prefix either way counts as survival |
| 92 | // -- but nothing weaker does. |
| 93 | function survives(text, replacement, of) { |
| 94 | return replacement.some((message) => { |
| 95 | const kept = of(message); |
| 96 | return ( |
| 97 | kept && |
| 98 | (kept === text || text.startsWith(kept) || kept.startsWith(text)) |
| 99 | ); |
| 100 | }); |
| 101 | } |
| 102 | |
| 103 | export function validateSurvivalContract(original, replacement, anchors) { |
| 104 | const start = lastRoundStart(original); |
| 105 | const round = original.slice(start); |
| 106 | const pending = new Set(); |
| 107 | const boundaries = []; |
| 108 | for (const [idx, message] of round.entries()) { |
| 109 | const calls = toolUseIds(message); |
| 110 | if (calls.length && pending.size === 0) boundaries.push(idx); |
| 111 | for (const id of calls) pending.add(id); |
| 112 | for (const id of toolResultIds(message)) pending.delete(id); |
| 113 | } |
| 114 | const tail = boundaries.length > 2 ? boundaries.at(-2) : 0; |
| 115 | const lastRound = round.filter((message, idx) => !isCheckpoint(message) && (idx >= tail || isPlainUserText(message))); |
| 116 | // Every user turn in the round, not the first one `find` reaches: the round |
| 117 | // spans a tool-bearing turn plus the toolless tail after it, so checking one |
| 118 | // let a rewrite drop the latest turn. |
| 119 | for (const text of lastRound.map(userTextOf).filter(Boolean)) { |
| 120 | if (!survives(text, replacement, userTextOf)) { |
| 121 | return "a last-round user message was dropped"; |
| 122 | } |
| 123 | } |
| 124 | for (const id of lastRound.flatMap(toolResultIds)) { |
| 125 | const kept = replacement.some((message) => |
| 126 | (message.content ?? []).some( |
| 127 | (block) => block.type === "tool_result" && block.tool_use_id === id, |
| 128 | ), |
| 129 | ); |
| 130 | if (!kept) { |
| 131 | return `last-round tool result ${id} was dropped`; |
| 132 | } |
| 133 | } |
| 134 | // The call, not just its result: a tool_result whose tool_use was summarized |
| 135 | // away is an orphan providers reject. |
| 136 | for (const id of lastRound.flatMap(toolUseIds)) { |
| 137 | const kept = replacement.some((message) => |
| 138 | (message.content ?? []).some( |
| 139 | (block) => block.type === "tool_use" && block.id === id, |
| 140 | ), |
| 141 | ); |
| 142 | if (!kept) { |
| 143 | return `last-round tool call ${id} was dropped`; |
| 144 | } |
| 145 | } |
| 146 | // Match the text: "some assistant message survived" is satisfied by the |
| 147 | // summary the rewrite itself just wrote. |
| 148 | for (const text of lastRound.map(assistantTextOf).filter(Boolean)) { |
| 149 | if (!survives(text, replacement, assistantTextOf)) { |
| 150 | return "last-round assistant output was dropped"; |
| 151 | } |
| 152 | } |
| 153 | if ( |
| 154 | lastRound.some(isAssistantLike) && |
| 155 | !replacement.some(isAssistantLike) |
| 156 | ) { |
| 157 | return "last-round assistant output was dropped"; |
| 158 | } |
| 159 | const checkpoints = replacement.filter(isCheckpoint).length; |
| 160 | if (checkpoints === 0) return "checkpoint receipt was dropped"; |
| 161 | if (checkpoints > 1) return "prior summaries were duplicated"; |
| 162 | if (anchors && !replacement.some((message) => |
| 163 | (message.content ?? []).some((block) => { |
| 164 | if (block.type === "text") return (block.text ?? "").includes(anchors); |
| 165 | if (block.type === "tool_result") { |
| 166 | return (block.content ?? "").includes(anchors); |
| 167 | } |
| 168 | return false; |
| 169 | }), |
| 170 | )) { |
| 171 | return "pinned /anchor text was dropped"; |
| 172 | } |
| 173 | return null; |
| 174 | } |
| 175 | |
| 176 | function main() { |
| 177 | if (matrix.schema_version !== 2) { |
| 178 | throw new Error(`unexpected schema_version ${matrix.schema_version}`); |
| 179 | } |
| 180 | let failed = 0; |
| 181 | for (const fixture of matrix.cases) { |
| 182 | if (typeof fixture.last_round_start === "number") { |
| 183 | const start = lastRoundStart(fixture.original); |
| 184 | if (start !== fixture.last_round_start) { |
| 185 | failed += 1; |
| 186 | console.error( |
| 187 | `${fixture.id}: last_round_start ${start} != ${fixture.last_round_start}`, |
| 188 | ); |
| 189 | } |
| 190 | } |
| 191 | const error = validateSurvivalContract( |
| 192 | fixture.original, |
| 193 | fixture.replacement, |
| 194 | fixture.anchors, |
| 195 | ); |
| 196 | const passed = error === null; |
| 197 | if (fixture.expect === "pass" && !passed) { |
| 198 | failed += 1; |
| 199 | console.error(`${fixture.id}: expected pass, got ${error}`); |
| 200 | } else if (fixture.expect === "fail" && passed) { |
| 201 | failed += 1; |
| 202 | console.error(`${fixture.id}: expected fail closed`); |
| 203 | } |
| 204 | } |
| 205 | if (failed > 0) { |
| 206 | console.error(`${failed} fixture(s) failed`); |
| 207 | process.exit(1); |
| 208 | } |
| 209 | console.log(`ok ${matrix.cases.length} survival-contract fixtures`); |
| 210 | } |
| 211 | |
| 212 | const entry = process.argv[1] && fileURLToPath(import.meta.url) === process.argv[1]; |
| 213 | if (entry || process.argv[1]?.endsWith("validate_survival_contract.mjs")) { |
| 214 | main(); |
| 215 | } |
| 216 |