|
117 | 117 | { |
118 | 118 | "case_id": "07_summary_create_update", |
119 | 119 | "baseline": "in_memory", |
120 | | - "status": "match", |
| 120 | + "status": "different", |
121 | 121 | "comparisons": [ |
122 | 122 | { |
123 | 123 | "backend": "sql", |
124 | 124 | "session_id": "replay-session", |
125 | | - "status": "match", |
| 125 | + "status": "different", |
126 | 126 | "diffs": [ |
127 | 127 | { |
128 | 128 | "path": "events[0].summary_metadata.summary_replaces_event_ids[4]", |
129 | 129 | "expected": "event-10", |
130 | 130 | "actual": null, |
131 | | - "allowed": true, |
| 131 | + "allowed": false, |
132 | 132 | "session_id": "replay-session", |
133 | 133 | "event_index": 0, |
134 | | - "target": "session", |
135 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 134 | + "target": "session" |
136 | 135 | }, |
137 | 136 | { |
138 | 137 | "path": "events[0].summary_metadata.summary_replaces_event_ids[5]", |
139 | 138 | "expected": "event-10", |
140 | 139 | "actual": null, |
141 | | - "allowed": true, |
| 140 | + "allowed": false, |
142 | 141 | "session_id": "replay-session", |
143 | 142 | "event_index": 0, |
144 | | - "target": "session", |
145 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 143 | + "target": "session" |
146 | 144 | }, |
147 | 145 | { |
148 | 146 | "path": "events[1].author", |
149 | 147 | "expected": "user", |
150 | 148 | "actual": "agent", |
151 | | - "allowed": true, |
| 149 | + "allowed": false, |
152 | 150 | "session_id": "replay-session", |
153 | 151 | "event_index": 1, |
154 | | - "target": "session", |
155 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 152 | + "target": "session" |
156 | 153 | }, |
157 | 154 | { |
158 | 155 | "path": "events[1].text", |
159 | 156 | "expected": "Give me a final short itinerary", |
160 | 157 | "actual": "Shenzhen will be sunny and warm", |
161 | | - "allowed": true, |
| 158 | + "allowed": false, |
162 | 159 | "session_id": "replay-session", |
163 | 160 | "event_index": 1, |
164 | | - "target": "session", |
165 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 161 | + "target": "session" |
166 | 162 | }, |
167 | 163 | { |
168 | 164 | "path": "events[2].author", |
169 | 165 | "expected": "agent", |
170 | 166 | "actual": "user", |
171 | | - "allowed": true, |
| 167 | + "allowed": false, |
172 | 168 | "session_id": "replay-session", |
173 | 169 | "event_index": 2, |
174 | | - "target": "session", |
175 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 170 | + "target": "session" |
176 | 171 | }, |
177 | 172 | { |
178 | 173 | "path": "events[2].text", |
179 | 174 | "expected": "Take the train, stay near the station, and pack light clothes", |
180 | 175 | "actual": "Give me a final short itinerary", |
181 | | - "allowed": true, |
| 176 | + "allowed": false, |
182 | 177 | "session_id": "replay-session", |
183 | 178 | "event_index": 2, |
184 | | - "target": "session", |
185 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 179 | + "target": "session" |
186 | 180 | } |
187 | 181 | ] |
188 | 182 | }, |
189 | 183 | { |
190 | 184 | "backend": "redis_mock", |
191 | 185 | "session_id": "replay-session", |
192 | | - "status": "match", |
| 186 | + "status": "different", |
193 | 187 | "diffs": [ |
194 | 188 | { |
195 | 189 | "path": "events[0].summary_metadata.summary_replaces_event_ids[4]", |
196 | 190 | "expected": "event-10", |
197 | 191 | "actual": null, |
198 | | - "allowed": true, |
| 192 | + "allowed": false, |
199 | 193 | "session_id": "replay-session", |
200 | 194 | "event_index": 0, |
201 | | - "target": "session", |
202 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 195 | + "target": "session" |
203 | 196 | }, |
204 | 197 | { |
205 | 198 | "path": "events[0].summary_metadata.summary_replaces_event_ids[5]", |
206 | 199 | "expected": "event-10", |
207 | 200 | "actual": null, |
208 | | - "allowed": true, |
| 201 | + "allowed": false, |
209 | 202 | "session_id": "replay-session", |
210 | 203 | "event_index": 0, |
211 | | - "target": "session", |
212 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 204 | + "target": "session" |
213 | 205 | }, |
214 | 206 | { |
215 | 207 | "path": "events[1].author", |
216 | 208 | "expected": "user", |
217 | 209 | "actual": "agent", |
218 | | - "allowed": true, |
| 210 | + "allowed": false, |
219 | 211 | "session_id": "replay-session", |
220 | 212 | "event_index": 1, |
221 | | - "target": "session", |
222 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 213 | + "target": "session" |
223 | 214 | }, |
224 | 215 | { |
225 | 216 | "path": "events[1].text", |
226 | 217 | "expected": "Give me a final short itinerary", |
227 | 218 | "actual": "Shenzhen will be sunny and warm", |
228 | | - "allowed": true, |
| 219 | + "allowed": false, |
229 | 220 | "session_id": "replay-session", |
230 | 221 | "event_index": 1, |
231 | | - "target": "session", |
232 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 222 | + "target": "session" |
233 | 223 | }, |
234 | 224 | { |
235 | 225 | "path": "events[2].author", |
236 | 226 | "expected": "agent", |
237 | 227 | "actual": "user", |
238 | | - "allowed": true, |
| 228 | + "allowed": false, |
239 | 229 | "session_id": "replay-session", |
240 | 230 | "event_index": 2, |
241 | | - "target": "session", |
242 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 231 | + "target": "session" |
243 | 232 | }, |
244 | 233 | { |
245 | 234 | "path": "events[2].text", |
246 | 235 | "expected": "Take the train, stay near the station, and pack light clothes", |
247 | 236 | "actual": "Give me a final short itinerary", |
248 | | - "allowed": true, |
| 237 | + "allowed": false, |
249 | 238 | "session_id": "replay-session", |
250 | 239 | "event_index": 2, |
251 | | - "target": "session", |
252 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 240 | + "target": "session" |
253 | 241 | } |
254 | 242 | ] |
255 | 243 | } |
|
258 | 246 | { |
259 | 247 | "case_id": "08_summary_truncation", |
260 | 248 | "baseline": "in_memory", |
261 | | - "status": "match", |
| 249 | + "status": "different", |
262 | 250 | "comparisons": [ |
263 | 251 | { |
264 | 252 | "backend": "sql", |
265 | 253 | "session_id": "replay-session", |
266 | | - "status": "match", |
| 254 | + "status": "different", |
267 | 255 | "diffs": [ |
268 | 256 | { |
269 | 257 | "path": "events[5].author", |
270 | 258 | "expected": "user", |
271 | 259 | "actual": "agent", |
272 | | - "allowed": true, |
| 260 | + "allowed": false, |
273 | 261 | "session_id": "replay-session", |
274 | 262 | "event_index": 5, |
275 | | - "target": "session", |
276 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 263 | + "target": "session" |
277 | 264 | }, |
278 | 265 | { |
279 | 266 | "path": "events[5].text", |
280 | 267 | "expected": "Please add one visualization exercise", |
281 | 268 | "actual": "Create a matplotlib chart from a CSV file", |
282 | | - "allowed": true, |
| 269 | + "allowed": false, |
283 | 270 | "session_id": "replay-session", |
284 | 271 | "event_index": 5, |
285 | | - "target": "session", |
286 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 272 | + "target": "session" |
287 | 273 | }, |
288 | 274 | { |
289 | 275 | "path": "events[6]", |
|
297 | 283 | "summary_metadata": {} |
298 | 284 | }, |
299 | 285 | "actual": null, |
300 | | - "allowed": true, |
| 286 | + "allowed": false, |
301 | 287 | "session_id": "replay-session", |
302 | 288 | "event_index": 6, |
303 | | - "target": "session", |
304 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 289 | + "target": "session" |
305 | 290 | }, |
306 | 291 | { |
307 | 292 | "path": "events[7]", |
|
315 | 300 | "summary_metadata": {} |
316 | 301 | }, |
317 | 302 | "actual": null, |
318 | | - "allowed": true, |
| 303 | + "allowed": false, |
319 | 304 | "session_id": "replay-session", |
320 | 305 | "event_index": 7, |
321 | | - "target": "session", |
322 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 306 | + "target": "session" |
323 | 307 | } |
324 | 308 | ] |
325 | 309 | }, |
326 | 310 | { |
327 | 311 | "backend": "redis_mock", |
328 | 312 | "session_id": "replay-session", |
329 | | - "status": "match", |
| 313 | + "status": "different", |
330 | 314 | "diffs": [ |
331 | 315 | { |
332 | 316 | "path": "events[5].author", |
333 | 317 | "expected": "user", |
334 | 318 | "actual": "agent", |
335 | | - "allowed": true, |
| 319 | + "allowed": false, |
336 | 320 | "session_id": "replay-session", |
337 | 321 | "event_index": 5, |
338 | | - "target": "session", |
339 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 322 | + "target": "session" |
340 | 323 | }, |
341 | 324 | { |
342 | 325 | "path": "events[5].text", |
343 | 326 | "expected": "Please add one visualization exercise", |
344 | 327 | "actual": "Create a matplotlib chart from a CSV file", |
345 | | - "allowed": true, |
| 328 | + "allowed": false, |
346 | 329 | "session_id": "replay-session", |
347 | 330 | "event_index": 5, |
348 | | - "target": "session", |
349 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 331 | + "target": "session" |
350 | 332 | }, |
351 | 333 | { |
352 | 334 | "path": "events[6]", |
|
360 | 342 | "summary_metadata": {} |
361 | 343 | }, |
362 | 344 | "actual": null, |
363 | | - "allowed": true, |
| 345 | + "allowed": false, |
364 | 346 | "session_id": "replay-session", |
365 | 347 | "event_index": 6, |
366 | | - "target": "session", |
367 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 348 | + "target": "session" |
368 | 349 | }, |
369 | 350 | { |
370 | 351 | "path": "events[7]", |
|
378 | 359 | "summary_metadata": {} |
379 | 360 | }, |
380 | 361 | "actual": null, |
381 | | - "allowed": true, |
| 362 | + "allowed": false, |
382 | 363 | "session_id": "replay-session", |
383 | 364 | "event_index": 7, |
384 | | - "target": "session", |
385 | | - "allowed_reason": "Known InMemory event aliasing after update_session() can duplicate post-summary events until the backend bug is fixed." |
| 365 | + "target": "session" |
386 | 366 | } |
387 | 367 | ] |
388 | 368 | } |
|
0 commit comments