Game Replay Analyzer

Project roadmap

1. 01

   Define the replay format

   ~2h

   Write FORMAT.md for a JSON Lines replay: a header (match id, map name and size, players and teams) and events with a time in milliseconds and a type (spawn, move with x and y, shot, hit with damage, kill, objective and match end). Write a generator that creates realistic replays of any length with a seed, so you have test data. Set up a CMake project with Boost and OpenCV. Done when: FORMAT.md is written, and the generator creates sample replays.

2. 02

   Parse and validate replays

   ~2.5h

   Parse the header and events with Boost.JSON, reading line by line so large files don't need to fit in memory. Validate: times never go backwards, players exist, positions are on the map, and a dead player doesn't act before respawning. Report problems with line numbers, and decide which ones stop the analysis and which are only warnings. Done when: a valid replay parses, and a damaged one reports its problems.

3. 03

   Build the timeline

   ~2h

   Store events in a timeline structure with fast lookup by time and player, and track each player's state as events are applied (alive or dead, position and health). Detect key moments, such as a player getting several kills within 10 seconds, or a team losing most of its players in a short time. Done when: the tool prints the key moments of a sample match with their times.

4. 04

   Compute player and match statistics

   ~2h

   For each player, compute kills, deaths, assists (damage to a player killed by someone else within a few seconds), accuracy (hits divided by shots), damage done and taken, and time alive; for the match, the score over time and each team's totals. Print them as aligned tables and write them as JSON with --json. Done when: the statistics for a small hand-written replay match the numbers calculated by hand.

5. 05

   Draw heatmaps with OpenCV

   ~2h

   Accumulate player positions over time into a grid covering the map, blur it with GaussianBlur, normalize it, color it with applyColorMap, and save it as a PNG, optionally per team or per player, and for kills only. Done when: the heatmap images clearly show where players spent their time.

6. 06

   Handle large and damaged files

   ~1.5h

   Measure memory and time on a very long generated replay and improve anything slow (for example storing positions compactly). Let the analysis continue past warnings, report skipped events, and use Boost.Program_options for the command line (input, output folder, which heatmaps and output format). Done when: a long replay is analyzed within reasonable time and memory, noted in README.md.

7. 07

   Test it

   ~2h

   Write tests with small hand-written replays covering each validation rule, assists within and outside the time window, accuracy with no shots, key moment detection, and consistency checks (total kills equal total deaths). Done when: all tests pass.

Resources

• DocsBoost.JSON documentation
• DocsBoost.Program_options tutorial
• DocsOpenCV: color maps
• DocsOpenCV: smoothing images
• DocsJSON Lines format