fix(#58): Timeline data now accurately reflects usage patterns by GalDayan · Pull Request #74 · GENWAY-AI/clawatch

GalDayan · 2026-03-25T03:31:09Z

Problem

The "Total Usage Over Time" chart showed misleading spikes because all of a session's cost/tokens were bucketed at the session's startedAt time. A session that starts at 05:00 but runs until 12:00 would show $X at 05:00 and nothing for 06:00-12:00.

Root Cause

getBucketKey(s.startedAt) was the sole bucket assignment — 100% of cost went to the start hour.

Fix

For hourly grouping, sessions now distribute their cost and tokens proportionally across all active hours (from startedAt to lastActivityAt).

A 4-hour session with $1.00 total → $0.25 in each of the 4 hourly buckets
Session count is only incremented in the start bucket (no double-counting)
Safety cap at 168 hours (1 week) to prevent runaway loops
Daily/weekly views unchanged (sessions rarely span multiple days)

Testing

Build clean (tsc --noEmit)
All 23 existing tests passing
Manual verification: sessions now show activity distributed across their actual working hours

Closes #58

…API endpoints - modelClassifier.ts: session complexity classifier (simple/moderate/complex) based on message length, tool usage patterns, and feature analysis - modelRecommendations.ts: maps complexity to cheaper model alternatives with cost savings estimates - routes.ts: GET /api/recommendations/:sessionId and /api/recommendations/summary Implements backend for AI-powered model recommendations (issue #51)

…ge cases - 23 tests covering classifier and recommendation mapper - Edge cases: zero-cost sessions, empty models, no messages, unknown models - Savings rounding to 4 decimal places - Bulk recommendations filter completed sessions, respect limits - Added vitest as test framework - All tests passing, build clean

Prevents Express from matching 'summary' as a sessionId param.

Sessions spanning multiple hours now spread cost/tokens proportionally across all active hours instead of spiking at startedAt. Root cause: getBucketKey(s.startedAt) put 100% of a session's cost into the start hour, creating misleading spikes in the chart. Fix: for hourly grouping, generate bucket keys for every hour between startedAt and lastActivityAt, distribute cost/tokens evenly. Daily/weekly views unchanged (sessions rarely span multiple days). Safety cap at 168 hours (1 week) to prevent runaway loops. Session count only incremented in the first (start) bucket.

Copilot

Pull request overview

Adjusts backend analytics bucketing so the “Total Usage Over Time” chart reflects session activity across hours (fixing misleading spikes), and introduces model recommendation APIs with supporting recommendation/classification logic and tests.

Changes:

Distributes session cost/tokens across multiple hourly buckets between startedAt and lastActivityAt in /api/analytics.
Adds /api/recommendations/summary and /api/recommendations/:sessionId endpoints plus recommendation + classification modules.
Adds Vitest test setup and new unit tests for model classification/recommendation logic.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
backend/src/routes.ts	Updates hourly analytics aggregation logic and adds recommendation API routes.
backend/src/modelRecommendations.ts	Implements recommendation generation + savings estimation helpers.
backend/src/modelClassifier.ts	Adds heuristic session complexity classification (detail + summary).
backend/src/tests/modelRecommendations.test.ts	Unit tests for recommendation logic and edge cases.
backend/src/tests/modelClassifier.test.ts	Unit tests for classifier heuristics.
backend/package.json	Adds Vitest scripts/dependency for backend tests.
backend/package-lock.json	Locks new Vitest dependency tree.

Files not reviewed (1)

backend/package-lock.json: Language not supported

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-25T03:36:41Z

+      // Distribute cost/tokens evenly across active hours
+      const costPerBucket = s.costUsd / numBuckets;
+      const tokensPerBucket = Math.round(s.tokenCount / numBuckets);
+


tokensPerBucket uses Math.round(s.tokenCount / numBuckets), which can change the total tokenCount when summed across buckets (e.g., 1 token over 2 buckets becomes 2). Use integer distribution that preserves the total (e.g., Math.floor plus distributing the remainder), or keep fractional tokens internally and round once at the end.

Copilot · 2026-03-25T03:36:41Z

+      // Distribute cost/tokens evenly across active hours
+      const costPerBucket = s.costUsd / numBuckets;
+      const tokensPerBucket = Math.round(s.tokenCount / numBuckets);


PR description says hourly cost/tokens are distributed proportionally across active hours, but the implementation distributes them evenly per bucket (costUsd / numBuckets, tokenCount / numBuckets) and doesn’t weight partial first/last hours. Either update the PR description or change the allocation to weight buckets by actual overlap duration per hour.

Copilot · 2026-03-25T03:36:41Z

+import { SessionDetail, SessionSummary } from "./sessions.js";
+import { classifySession, classifySessionSummary, SessionComplexity } from "./modelClassifier.js";


These imports use a .js extension and are not type-only. In this repo, TS files import local modules without extensions (e.g., backend/src/index.ts:7-11), and tsconfig.json uses module: "commonjs". Using ./sessions.js / ./modelClassifier.js can break TS/tsx resolution and also forces a runtime import for what are mostly types—prefer extensionless paths and import type for SessionDetail/SessionSummary/SessionComplexity.

Copilot · 2026-03-25T03:36:42Z

@@ -0,0 +1,148 @@
+import { SessionDetail, SessionSummary } from "./sessions.js";


This file imports from ./sessions.js. The rest of the backend TS uses extensionless local imports (e.g., backend/src/index.ts:7-11) and tsconfig.json is module: "commonjs". Prefer extensionless imports and import type here to avoid resolution issues and unintended runtime imports.

Suggested change

import { SessionDetail, SessionSummary } from "./sessions.js";

import type { SessionDetail, SessionSummary } from "./sessions";

Copilot · 2026-03-25T03:36:42Z

+  // Calculate potential savings
+  const currentCost = session.costUsd;
+  const estimatedNewCost = estimateSessionCost(session.tokenCount, recommendation.model);
+  const savings = Math.max(0, currentCost - estimatedNewCost);
+  const savingsPercentage = currentCost > 0 ? (savings / currentCost) * 100 : 0;


If recommendation.model equals session.model (no change), this still computes savings by comparing actual cost to an estimated cost for the same model, which can yield misleading positive savings. Consider short-circuiting savings/savingsPercentage to 0 when no model change is recommended.

Copilot · 2026-03-25T03:36:42Z

+    const limit = parseInt(req.query.limit as string) || 20;
+


limit is taken directly from the querystring with parseInt(...) || 20, which treats 0 as “use default” and allows negatives/very large values. Clamp/validate the value (e.g., min 1, max reasonable) and return a 400 for invalid input.

Suggested change

const limit = parseInt(req.query.limit as string) || 20;

const rawLimit = req.query.limit as string | undefined;

let limit = 20;

if (rawLimit !== undefined) {

const parsedLimit = Number.parseInt(rawLimit, 10);

const MIN_LIMIT = 1;

const MAX_LIMIT = 1000;

if (Number.isNaN(parsedLimit) || parsedLimit < MIN_LIMIT || parsedLimit > MAX_LIMIT) {

res.status(400).json({ error: "Invalid 'limit' parameter" });

return;

}

limit = parsedLimit;

}

Copilot · 2026-03-25T03:36:42Z

+        keys.push(cursor.toISOString().slice(0, 13) + ":00");
+        cursor.setUTCHours(cursor.getUTCHours() + 1);
+        // Safety: cap at 168 hours (1 week) to avoid runaway loops
+        if (keys.length > 168) break;


The “168 hours” safety cap is off by one: the loop breaks only when keys.length > 168, so it can return 169 buckets. If the intent is a strict 168-hour cap, break when keys.length === 168 (or check before pushing).

Suggested change

if (keys.length > 168) break;

if (keys.length >= 168) break;

Copilot · 2026-03-25T03:36:42Z

+      const endKey = getBucketKey(s.lastActivityAt);
+      while (cursor.toISOString().slice(0, 13) + ":00" <= endKey) {
+        keys.push(cursor.toISOString().slice(0, 13) + ":00");


The while (... <= endKey) loop will include the bucket for lastActivityAt even when the session ends exactly on an hour boundary (e.g., 10:30→11:00 yields both 10:00 and 11:00), over-distributing cost/tokens into an extra hour. Prefer looping with Date comparisons (e.g., push while cursor < end) or make the end boundary exclusive when aligned to :00.

Suggested change

const endKey = getBucketKey(s.lastActivityAt);

while (cursor.toISOString().slice(0, 13) + ":00" <= endKey) {

keys.push(cursor.toISOString().slice(0, 13) + ":00");

while (cursor < end) {

const key = cursor.toISOString().slice(0, 13) + ":00";

keys.push(key);

Copilot · 2026-03-25T03:36:42Z

+      const estimatedNewCost = estimateSessionCost(session.tokenCount, recommendation.model);
+      const savings = Math.max(0, currentCost - estimatedNewCost);
+      const savingsPercentage = currentCost > 0 ? (savings / currentCost) * 100 : 0;
+


In bulk recommendations, if recommendation.model equals session.model, the computed savings is just “actual vs estimated” for the same model and can be misleading. Consider forcing savings to 0 when there’s no model change and/or filtering out entries where recommendedModel === currentModel.

Suggested change

const estimatedNewCost = estimateSessionCost(session.tokenCount, recommendation.model);

const savings = Math.max(0, currentCost - estimatedNewCost);

const savingsPercentage = currentCost > 0 ? (savings / currentCost) * 100 : 0;

const isSameModel = recommendation.model === session.model;

let savings = 0;

let savingsPercentage = 0;

if (!isSameModel) {

const estimatedNewCost = estimateSessionCost(session.tokenCount, recommendation.model);

savings = Math.max(0, currentCost - estimatedNewCost);

savingsPercentage = currentCost > 0 ? (savings / currentCost) * 100 : 0;

}

GalDayan · 2026-04-15T23:57:29Z

Closing — this PR is mis-scoped: labeled as fix for #58 but diff includes modelClassifier/modelRecommendations files that belong to #51 (covered by #73). Keeping #70 as the clean fix for #58 (focused on backend/src/routes.ts + sessions.ts).

GalDayan added 4 commits March 25, 2026 04:45

fix(#51): reorder routes — static /summary before param /:sessionId

5b6a431

Prevents Express from matching 'summary' as a sessionId param.

Copilot AI review requested due to automatic review settings March 25, 2026 03:31

Copilot started reviewing on behalf of GalDayan March 25, 2026 03:31 View session

Copilot AI reviewed Mar 25, 2026

View reviewed changes

GalDayan closed this Apr 15, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(#58): Timeline data now accurately reflects usage patterns#74

fix(#58): Timeline data now accurately reflects usage patterns#74
GalDayan wants to merge 4 commits intomainfrom
fix/issue-58-timeline-accuracy

GalDayan commented Mar 25, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

GalDayan commented Apr 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		import { SessionDetail, SessionSummary } from "./sessions.js";
		import { classifySession, classifySessionSummary, SessionComplexity } from "./modelClassifier.js";

		@@ -0,0 +1,148 @@
		import { SessionDetail, SessionSummary } from "./sessions.js";

	import { SessionDetail, SessionSummary } from "./sessions.js";
	import type { SessionDetail, SessionSummary } from "./sessions";

-    const limit = parseInt(req.query.limit as string) || 20;
+    const rawLimit = req.query.limit as string | undefined;
+    let limit = 20;
+    if (rawLimit !== undefined) {
+      const parsedLimit = Number.parseInt(rawLimit, 10);
+      const MIN_LIMIT = 1;
+      const MAX_LIMIT = 1000;
+      if (Number.isNaN(parsedLimit) || parsedLimit < MIN_LIMIT || parsedLimit > MAX_LIMIT) {
+        res.status(400).json({ error: "Invalid 'limit' parameter" });
+        return;
+      }
+      limit = parsedLimit;
+    }

Conversation

GalDayan commented Mar 25, 2026

Problem

Root Cause

Fix

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

GalDayan commented Apr 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants