add checkpoint token comparison · Entire
add checkpoint token comparison
bfd2fe2·
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
229 unmodified lines
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
436
437
438
439
440
441
442
443
444
445
446
447
32 unmodified lines
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
522
523
524
525
526
527
528
529
530
531
532
533
4 unmodified lines
"encoding/json"
"fmt"
"io"
"strconv"
"strings"
"github.com/entireio/cli/cmd/entire/cli/agent"
16 unmodified lines
Context *sessionTokensContext `json:"context,omitempty"`
Contributors []sessionTokensContributor `json:"contributors,omitempty"`
Recommendations []sessionTokensRecommendation `json:"recommendations,omitempty"`
Comparison *checkpointTokensComparison `json:"comparison,omitempty"`
Limitations []string `json:"limitations,omitempty"`
}
type checkpointTokensComparison struct {
BaselineCheckpointID string `json:"baseline_checkpoint_id"`
TargetCheckpointID string `json:"target_checkpoint_id"`
Status string `json:"status"`
Total *checkpointTokensMetricDelta `json:"total,omitempty"`
CacheRead *checkpointTokensMetricDelta `json:"cache_read,omitempty"`
APICalls *checkpointTokensMetricDelta `json:"api_calls,omitempty"`
Qualification string `json:"qualification"`
Limitations []string `json:"limitations,omitempty"`
}
type checkpointTokensMetricDelta struct {
Baseline int `json:"baseline"`
Current int `json:"current"`
Change int `json:"change"`
ChangePercent *float64 `json:"change_percent,omitempty"`
Direction string `json:"direction"`
}
const (
checkpointComparisonStatusUnavailable = "unavailable"
checkpointComparisonStatusObservedReduction = "observed_reduction"
checkpointComparisonStatusObservedIncrease = "observed_increase"
checkpointComparisonStatusObservedNoChange = "observed_no_change"
)
func newCheckpointTokensCmd() *cobra.Command {
var jsonFlag bool
var compareFlag string
cmd := &cobra.Command{
Use: "tokens <checkpoint-id>",
4 unmodified lines
resolution 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.
`,
from the checkpoint remote.
Use --compare <checkpoint-id> to compare this checkpoint against a previous
checkpoint and qualify observed token reduction or increase.
`,
Args: cobra.ExactArgs(1),
RunE: func(cmd *cobra.Command, args []string) error {
return runCheckpointTokens(cmd.Context(), cmd, args[0], jsonFlag)
return runCheckpointTokens(cmd.Context(), cmd, args[0], jsonFlag, compareFlag)
},
}
cmd.Flags().BoolVar(&jsonFlag, "json", false, "Output as JSON")
cmd.Flags().StringVar(&compareFlag, "compare", "", "Compare against a baseline checkpoint ID")
return cmd
}
func runCheckpointTokens(ctx context.Context, cmd *cobra.Command, checkpointIDPrefix string, jsonOutput bool) error {
cpID, lookup, err := resolveExplainCheckpointID(ctx, cmd.ErrOrStderr(), explainExportOptions{target: checkpointIDPrefix})
func runCheckpointTokens(ctx context.Context, cmd *cobra.Command, checkpointIDPrefix string, jsonOutput bool, comparePrefix string) error {
report, lookup, err := loadCheckpointTokensReport(ctx, cmd, checkpointIDPrefix)
if lookup != nil {
defer lookup.Close()
}
1 unmodified line
return tokenCommandError(err)
}
if comparePrefix != "" {
baselineReport, baselineLookup, err := loadCheckpointTokensReport(ctx, cmd, comparePrefix)
if baselineLookup != nil {
defer baselineLookup.Close()
}
if err != nil {
return tokenCommandError(err)
}
report.Comparison = buildCheckpointTokensComparison(report, baselineReport)
}
if jsonOutput {
return writeCheckpointTokensJSON(cmd.OutOrStdout(), report)
}
writeCheckpointTokensText(cmd.OutOrStdout(), report)
return nil
}
func loadCheckpointTokensReport(ctx context.Context, cmd *cobra.Command, checkpointIDPrefix string) (checkpointTokensReport, *explainCheckpointLookup, error) {
cpID, lookup, err := resolveExplainCheckpointID(ctx, cmd.ErrOrStderr(), explainExportOptions{target: checkpointIDPrefix})
if err != nil {
return checkpointTokensReport{}, lookup, err
}
summary, err := lookup.store.ReadCommitted(ctx, cpID)
if err != nil {
return tokenCommandError(fmt.Errorf("failed to read checkpoint: %w", err))
return checkpointTokensReport{}, lookup, fmt.Errorf("failed to read checkpoint: %w", err)
}
if summary == nil || len(summary.Sessions) == 0 {
cmd.SilenceUsage = true
fmt.Fprintln(cmd.ErrOrStderr(), "Checkpoint not found.")
return NewSilentError(fmt.Errorf("%w: %s", checkpoint.ErrCheckpointNotFound, checkpointIDPrefix))
return checkpointTokensReport{}, lookup, NewSilentError(fmt.Errorf("%w: %s", checkpoint.ErrCheckpointNotFound, checkpointIDPrefix))
}
metas, metadataWarnings, err := readCheckpointTokenSessionMetadata(ctx, lookup.store, cpID, len(summary.Sessions))
if err != nil {
return tokenCommandError(err)
return checkpointTokensReport{}, lookup, err
}
report := buildCheckpointTokensReport(cpID, summary, metas, metadataWarnings)
if jsonOutput {
return writeCheckpointTokensJSON(cmd.OutOrStdout(), report)
}
writeCheckpointTokensText(cmd.OutOrStdout(), report)
return nil
return buildCheckpointTokensReport(cpID, summary, metas, metadataWarnings), lookup, nil
}
func readCheckpointTokenSessionMetadata(ctx context.Context, store checkpoint.CommittedListReader, cpID id.CheckpointID, sessionCount int) ([]*checkpoint.CommittedMetadata, int, error) {
229 unmodified lines
return "s"
}
func buildCheckpointTokensComparison(target, baseline checkpointTokensReport) *checkpointTokensComparison {
comparison := &checkpointTokensComparison{
BaselineCheckpointID: baseline.CheckpointID,
TargetCheckpointID: target.CheckpointID,
}
if target.Tokens == nil || baseline.Tokens == nil {
comparison.Status = checkpointComparisonStatusUnavailable
comparison.Qualification = "Comparison unavailable because token usage is missing for one checkpoint."
comparison.Limitations = append(comparison.Limitations, comparison.Qualification)
return 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 {
delta := &checkpointTokensMetricDelta{
Baseline: baseline,
Current: current,
Change: current - baseline,
Direction: checkpointDeltaDirection(current - baseline),
}
if baseline != 0 {
percent := float64(delta.Change) * 100 / float64(baseline)
delta.ChangePercent = &percent
}
return delta
}
func checkpointDeltaDirection(change int) string {
switch {
case change < 0:
return "down"
case change > 0:
return "up"
default:
return "unchanged"
}
}
func checkpointComparisonStatus(total *checkpointTokensMetricDelta) string {
if total == nil {
return checkpointComparisonStatusUnavailable
}
switch total.Direction {
case "down":
return checkpointComparisonStatusObservedReduction
case "up":
return checkpointComparisonStatusObservedIncrease
default:
return checkpointComparisonStatusObservedNoChange
}
}
func checkpointComparisonQualification(status string) string {
switch status {
case checkpointComparisonStatusObservedReduction:
return "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."
case checkpointComparisonStatusObservedIncrease:
return "Observed token use increased for this checkpoint comparison. Check whether the extra context was necessary before treating it as waste."
case checkpointComparisonStatusObservedNoChange:
return "Observed token use was unchanged for this checkpoint comparison. Quality still depends on the task outcome, not token totals alone."
default:
return "Comparison unavailable because token usage is missing for one checkpoint."
}
}
func writeCheckpointTokensJSON(w io.Writer, report checkpointTokensReport) error {
enc := json.NewEncoder(w)
enc.SetIndent("", " ")
}
writeTokenUsageSection(w, report.Tokens)
writeCheckpointTokenComparison(w, report.Comparison)
if len(report.Recommendations) > 0 {
writeTokenRecommendations(w, report.Recommendations)
}
writeTokenContributors(w, report.Contributors, report.Context)
writeTokenLimitations(w, report.Limitations)
}
func writeCheckpointTokenComparison(w io.Writer, comparison *checkpointTokensComparison) {
if comparison == nil {
return
}
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 {
if delta == nil {
return "unavailable"
}
from := formatValue(delta.Baseline)
to := formatValue(delta.Current)
if delta.Direction == "unchanged" {
return fmt.Sprintf("unchanged (%%s -> %%s)", from, to)
}
if delta.ChangePercent == nil {
return fmt.Sprintf("%%s (%%s -> %%s)", delta.Direction, from, to)
}
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
}