Tactical plot NAV // DAYS
Tactical plot // 31 release days
- DAY017 logs
- DAY025 logs
- DAY035 logs
- DAY045 logs
- DAY055 logs
- DAY065 logs
- DAY075 logs
- DAY085 logs
- DAY095 logs
- DAY105 logs
- DAY115 logs
- DAY125 logs
- DAY135 logs
- DAY145 logs
- DAY155 logs
- DAY165 logs
- DAY175 logs
- DAY185 logs
- DAY190102038 logs
- DAY2004057 logs
- DAY210607088 logs
- DAY2209107 logs
- DAY231112138 logs
- DAY2414157 logs
- DAY251617188 logs
- DAY2619207 logs
- DAY272122238 logs
- DAY2824257 logs
- DAY2926277 logs
- DAY3028298 logs
- DAY3130317 logs
Lanes NAV // LANES
| Lane | Write-ups | Single Solve | Time Trial | Mastery |
|---|---|---|---|---|
| Warmups | 32 | 31 | 30 | 29 |
| Metadata Forensics | 31 | 31 | 29 | 29 |
| Mobile Android | 31 | 30 | 29 | 29 |
| Script Analysis | 31 | 31 | 31 | 31 |
| Binary RE | 32 | 31 | 30 | 30 |
| Web Exploitation | 31 | 31 | 31 | 30 |
| Other | 1 | 1 | 0 | 0 |
| Total | 189 | 186 | 180 | 178 |
Mode columns count write-ups that report that mode CORRECT in their own text (parsed automatically; a write-up that only says "solved" is not counted).
Headline lessons LOG // DEBRIEF
- 01First contact is the speed.
When a lane's deterministic solver already knew the day's shape, a mode landed in one to three seconds. When it did not, the same mode cost an agent anywhere from seconds to half an hour. Almost every speed gain came from raising the share of modes the solver handled on first contact.
- 02Every lane was yesterday plus one layer.
The daily challenges in a lane shared a spine and added one new container, encoding or language each day. Once we wrote that down as data and predicted the next layer, we could prebuild for it the night before instead of meeting it cold at release.
- 03Exact strings, no partial feedback.
The grader compares exact text and rejects a whole batch if one field is off, without saying which. Banking every accepted value and format (hex width, casing, braces) turned a guessing game into a lookup.
- 04A thirty-second window forgives nothing.
On Day 11 the release-time rate limiting made our retry backoff sleep straight through a timed window, and a lost Time Trial also locked the Mastery behind it. Afterwards every timed start became a hard time budget, preceded by a free status check of how many attempts remained.
- 05Derive from this attempt's bytes.
Timed material rotated per attempt. Reusing an old answer was never safe; only a solver that read the current artifact was.
- 06One owner per mode.
With two stations and shared attempts, two runners on one mode could burn a scarce ranked attempt. Atomic claims and static lane ownership kept that from happening.
- 07A model in the hot path sets a floor.
Each agent decision is a model turn. The fastest teams cleared a full day in minutes with plain scripts dispatching to solvers; our agents were best used for triage and for extending solvers, not for executing every step. A full reverse-engineering special (Day 30) was a different class entirely: hours, not seconds.
Late-event contacts LOG // RECENT
- Final Boss
"The gauntlet: everything you've learned ... in one campaign." The artifact is an AES-256 evidence.zip (password infected, 46 members). It has one dNN-- member group for each of Days 01-31 (ELF, PE, .NET ...
- Framework Roulette
Result: all three modes CORRECT by 13:51 ET on release day (Single Solve, Time Trial, Mastery with six accepted ids). Time Trial and Mastery were regular untimed attempts, not ranked. All three submits ...
- Full Stack Incident
The target exposed its own clues and request log. Those endpoints were the route map; the route names were randomized for this instance, so they should be read from /api/clues instead of guessed or hardcoded.
- Mobile Arena Gauntlet
Every APK carries assets/huntress_puzzle.json with codec: "analysis" (schema huntress.mobile.puzzle.v2, a day and a build). The answer is not in the JSON: it is the String returned by Release.code(Context) ...
- No Artifact Left Behind
Only one field was new: graph.execution_chain_artifact_ids for case-001.zip. Every other id in the Time Trial (36) and Mastery (540) used extractors that already existed. The answer is a compact JSON array ...
- One More Thing
Single / Time Trial: one blob, base64 -> gzip -> base32 -> slug; solved by the stock decoder chain with no change. Mastery: 153 numbered-line blobs of 31 rows each (ladder D30 plus a new row 31). Rows 1-30 ...
All 189 write-ups > How we were set up > The Game Master contract >