add checkpoint token comparison · Entire
add checkpoint token comparison
e94a5d0·
peyton-alt·3w ago·2 files·+569 added/-13 removed
Sessions
cea04e74bd49View transcript
Changes
2
- cmd/entire/cli
- Mcheckpoint_tokens.go+182/-13
- Msessions_test.go+387
4 unmodified lines
5
6
7
8
9
10
11
16 unmodified lines
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
4 unmodified lines
71
72
73
45
74
75
76
77
78
79
48
80
81
82
83
84
85
86
87
88
56
57
89
90
91
92
93
1 unmodified line
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
67
124
125
126
127
128
72
129
130
131
132
133
77
134
135
136
80
81
82
83
84
85
137
138
139
140
217 unmodified lines
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
32 unmodified lines
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
4 unmodified lines
\t\t\t\t\t\t\t\t\t\t\t"encoding/json"
"fmt"
"io"
"strconv"
"strings"
"github.com/entireio/cli/cmd/entire/cli/agent"
16 unmodified lines
\t\t\t\t\t\t\t\t\t\t\tContext *sessionTokensContext `json:"context,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tContributors []sessionTokensContributor `json:"contributors,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tRecommendations []sessionTokensRecommendation `json:"recommendations,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tComparison *checkpointTokensComparison `json:"comparison,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tLimitations []string `json:"limitations,omitempty"`
\t\t\t\t\t\t\t\t\t\t\t}
\n\ttype checkpointTokensComparison struct {
\t\t\t\t\t\t\t\t\t\t\tBaselineCheckpointID string `json:"baseline_checkpoint_id"`
\t\t\t\t\t\t\t\t\t\t\tTargetCheckpointID string `json:"target_checkpoint_id"`
\t\t\t\t\t\t\t\t\t\t\tStatus string `json:"status"`
\t\t\t\t\t\t\t\t\t\t\tTotal *checkpointTokensMetricDelta `json:"total,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tCacheRead *checkpointTokensMetricDelta `json:"cache_read,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tAPICalls *checkpointTokensMetricDelta `json:"api_calls,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tQualification string `json:"qualification"`
\t\t\t\t\t\t\t\t\t\t\tLimitations []string `json:"limitations,omitempty"`
\t\t\t\t\t\t\t\t\t\t\t}
\ttype checkpointTokensMetricDelta struct {
\t\t\t\t\t\t\t\t\t\t\tBaseline int `json:"baseline"`
\t\t\t\t\t\t\t\t\t\t\tCurrent int `json:"current"`
\t\t\t\t\t\t\t\t\t\t\tChange int `json:"change"`
\t\t\t\t\t\t\t\t\t\t\tChangePercent *float64 `json:"change_percent,omitempty"`
\t\t\t\t\t\t\t\t\t\t\tDirection string `json:"direction"`
\t\t\t\t\t\t\t\t\t\t\t}
const (
\tcheckpointComparisonStatusUnavailable = "unavailable"
\tcheckpointComparisonStatusObservedReduction = "observed_reduction"
\tcheckpointComparisonStatusObservedIncrease = "observed_increase"
\tcheckpointComparisonStatusObservedNoChange = "observed_no_change"
)
func newCheckpointTokensCmd() *cobra.Command {
\tvar jsonFlag bool
\tvar compareFlag string
\tcmd := &cobra.Command{
\t\tUse: "tokens <checkpoint-id>",
\t \tresolution path as 'entire checkpoint explain'. Checkpoint IDs may be abbreviated as long as the prefix is unambiguous; positional targets may also resolve from a commit ref with an Entire-Checkpoint trailer, and missing metadata may be fetched from the checkpoint remote.
\t\t\t\t\tArgs: cobra.ExactArgs(1),
\t\t\t\t\tRunE: func(cmd *cobra.Command, args []string) error {
\t\t\t\t\t\treturn runCheckpointTokens(cmd.Context(), cmd, args[0], jsonFlag)
\t\t\t\t\t\treturn runCheckpointTokens(cmd.Context(), cmd, args[0], jsonFlag, compareFlag)
\t\t\t\t},
\t}
\tcmd.Flags().BoolVar(&jsonFlag, "json", false, "Output as JSON")
\tcmd.Flags().StringVar(&compareFlag, "compare", "", "Compare against a baseline checkpoint ID")
\treturn cmd
}
func runCheckpointTokens(ctx context.Context, cmd *cobra.Command, checkpointIDPrefix string, jsonOutput bool) error {
\tcpID, lookup, err := resolveExplainCheckpointID(ctx, cmd.ErrOrStderr(), explainExportOptions{target: checkpointIDPrefix})
\tif lookup != nil {
\t defer lookup.Close()
\t}
\treturn tokenCommandError(err)
}
if comparePrefix != "" {
\tbaselineReport, baselineLookup, err := loadCheckpointTokensReport(ctx, cmd, comparePrefix)
\tif baselineLookup != nil {
\t defer baselineLookup.Close()
\t}
\tif err != nil {
\t return tokenCommandError(err)
}\n\treport.Comparison = buildCheckpointTokensComparison(report, baselineReport)
\t}
if jsonOutput {
\treturn writeCheckpointTokensJSON(cmd.OutOrStdout(), report)
}
writeCheckpointTokensText(cmd.OutOrStdout(), report)
return nil
}
func loadCheckpointTokensReport(ctx context.Context, cmd *cobra.Command, checkpointIDPrefix string) (checkpointTokensReport, *explainCheckpointLookup, error) {
\tcpID, lookup, err := resolveExplainCheckpointID(ctx, cmd.ErrOrStderr(), explainExportOptions{target: checkpointIDPrefix})
\tif err != nil {
\treturn checkpointTokensReport{}, lookup, err
}\n\tsummary, err := lookup.store.ReadCommitted(ctx, cpID)
\tif err != nil {
\treturn tokenCommandError(fmt.Errorf("failed to read checkpoint: %w", err))
}
if summary == nil || len(summary.Sessions) == 0 {
\tcmd.SilenceUsage = true
\tfmt.Fprintln(cmd.ErrOrStderr(), "Checkpoint not found.")
\treturn NewSilentError(fmt.Errorf("%w: %s", checkpoint.ErrCheckpointNotFound, checkpointIDPrefix))
}
metas, metadataWarnings, err := readCheckpointTokenSessionMetadata(ctx, lookup.store, cpID, len(summary.Sessions))
if err != nil {
\treturn tokenCommandError(err)
}
report := buildCheckpointTokensReport(cpID, summary, metas, metadataWarnings)
if jsonOutput {
\treturn writeCheckpointTokensJSON(cmd.OutOrStdout(), report)
}
writeCheckpointTokensText(cmd.OutOrStdout(), report)
return nil
}
func readCheckpointTokenSessionMetadata(ctx context.Context, store checkpoint.CommittedListReader, cpID id.CheckpointID, sessionCount int) ([]*checkpoint.CommittedMetadata, int, error) {
return "s"
}
func buildCheckpointTokensComparison(target, baseline checkpointTokensReport) *checkpointTokensComparison {
\tcomparison := &checkpointTokensComparison{
\t\tBaselineCheckpointID: baseline.CheckpointID,
\t\tTargetCheckpointID: target.CheckpointID,
\t}
\tif target.Tokens == nil || baseline.Tokens == nil {
\tcomparison.Status = checkpointComparisonStatusUnavailable
\tcomparison.Qualification = "Comparison unavailable because token usage is missing for one checkpoint."
\tcomparison.Limitations = append(comparison.Limitations, comparison.Qualification)
\treturn comparison
}
comparison.Total = buildCheckpointMetricDelta(baseline.Tokens.Total, target.Tokens.Total)
comparison.CacheRead = buildCheckpointMetricDelta(baseline.Tokens.CacheRead, target.Tokens.CacheRead)
comparison.APICalls = buildCheckpointMetricDelta(baseline.Tokens.APICalls, target.Tokens.APICalls)
comparison.Status = checkpointComparisonStatus(comparison.Total)
comparison.Qualification = checkpointComparisonQualification(comparison.Status)
return comparison
}
func buildCheckpointMetricDelta(baseline, current int) *checkpointTokensMetricDelta {
\tdelta := &checkpointTokensMetricDelta{
\t\tBaseline: baseline,
\t\tCurrent: current,
\t\tChange: current - baseline,
\t\tDirection: checkpointDeltaDirection(current - baseline),
\t}
\tif baseline != 0 {
\t percent := float64(delta.Change) * 100 / float64(baseline)
\t delta.ChangePercent = &percent
}
\treturn delta
}
func checkpointDeltaDirection(change int) string {
\tswitch {
\tcase change < 0:
\t\treturn "down"
\tcase change > 0:
\t\treturn "up"
\tdefault:
\t\treturn "unchanged"
\t}
}
func checkpointComparisonStatus(total *checkpointTokensMetricDelta) string {
\tif total == nil {
\t\treturn checkpointComparisonStatusUnavailable
\t}
\tswitch total.Direction {
\tcase "down":
\t\treturn checkpointComparisonStatusObservedReduction
\tcase "up":
\t\treturn checkpointComparisonStatusObservedIncrease
\tdefault:
\t\treturn checkpointComparisonStatusObservedNoChange
\t}
}
func checkpointComparisonQualification(status string) string {
\tswitch status {
\tcase checkpointComparisonStatusObservedReduction:
\t\treturn "Observed token use decreased for this checkpoint comparison. This does not prove quality was preserved; verify the task outcome or tests before treating it as a successful optimization."
\tcase checkpointComparisonStatusObservedIncrease:
\t\treturn "Observed token use increased for this checkpoint comparison. Check whether the extra context was necessary before treating it as waste."
\tcase checkpointComparisonStatusObservedNoChange:
\t\treturn "Observed token use was unchanged for this checkpoint comparison. Quality still depends on the task outcome, not token totals alone."
\tdefault:
\t\treturn "Comparison unavailable because token usage is missing for one checkpoint."
}\n}
func writeCheckpointTokensJSON(w io.Writer, report checkpointTokensReport) error {
\tenc := json.NewEncoder(w)
\tenc.SetIndent("", " ")
}
writeTokenUsageSection(w, report.Tokens)
writeCheckpointTokenComparison(w, report.Comparison)
if len(report.Recommendations) > 0 {
\twriteTokenRecommendations(w, report.Recommendations)
}
writeTokenContributors(w, report.Contributors, report.Context)
writeTokenLimitations(w, report.Limitations)
}
func writeCheckpointTokenComparison(w io.Writer, comparison *checkpointTokensComparison) {
\tif comparison == nil {
\t\treturn
\t}
fmt.Fprintln(w)
fmt.Fprintln(w, "Comparison")
fmt.Fprintf(w, "Baseline: %s\n", comparison.BaselineCheckpointID)
if comparison.Status != checkpointComparisonStatusUnavailable {
fmt.Fprintf(w, "Total tokens: %s\n", formatCheckpointMetricDelta(comparison.Total, formatTokenCount))
fmt.Fprintf(w, "Cache/context replay: %s\n", formatCheckpointMetricDelta(comparison.CacheRead, formatTokenCount))
fmt.Fprintf(w, "API calls: %s\n", formatCheckpointMetricDelta(comparison.APICalls, formatPlainCount))
}
fmt.Fprintln(w)
fmt.Fprintln(w, "Qualification")
fmt.Fprintln(w, comparison.Qualification)
}
func formatCheckpointMetricDelta(delta *checkpointTokensMetricDelta, formatValue func(int) string) string {
\tif delta == nil {
\t\treturn "unavailable"
\t}
\tfrom := formatValue(delta.Baseline)
\tto := formatValue(delta.Current)
\tif delta.Direction == "unchanged" {
\t\treturn fmt.Sprintf("unchanged (%%s -> %%s)", from, to)
\t}
\tif delta.ChangePercent == nil {
\t\treturn fmt.Sprintf("%%s (%%s -> %%s)", delta.Direction, from, to)
\t}
return fmt.Sprintf("%%s %%s (%%s -> %%s)", delta.Direction, formatPercent(absFloat(*delta.ChangePercent)), from, to)
}
func formatPlainCount(value int) string {
return strconv.Itoa(value)
}
func absFloat(value float64) float64 {
if value < 0 {
return -value
}
return value
}