# The Black dragon: what the runs show Written by `analysis/black/evidence.py` from `data/public/runs_clean.csv`. Counts only; nothing is fitted. "Gated tree runs" investigated the scream, helped the squadmate and climbed the tree, the two gates the developer has confirmed (`planner/knowledge.yaml`). Form submissions over-report the shadow, so their percentages are too high; comparisons inside the form (first attempt against later, one route against another) are still fair if the over-reporting is the same on both sides. ## 1. The shadow by attempt number | Attempt | Form submissions | Bulk logs | Bot | |---|---|---|---| | 1 | 90 / 892 (10.1%) | 0 / 28 (0.0%) | 0 / 0 | | 2 | 0 / 290 (0.0%) | 0 / 21 (0.0%) | 0 / 4 (0.0%) | | 3 | 0 / 187 (0.0%) | 0 / 14 (0.0%) | 0 / 5 (0.0%) | | 4-10 | 0 / 510 (0.0%) | 0 / 102 (0.0%) | 0 / 36 (0.0%) | | 11-30 | 0 / 574 (0.0%) | 0 / 57 (0.0%) | 0 / 137 (0.0%) | | 31 or more | 1 / 621 (0.2%) | 0 / 0 | 0 / 54 (0.0%) | | not given | 3 / 1522 (0.2%) | 0 / 112 (0.0%) | 0 / 0 | Complete logs (bot and bulk) have 28 first attempts at the tree and 430 later ones, with 0 shadows. Zero in 430 later attempts rules out a rate above 0.7% for them; zero in 28 first attempts only rules out more than 10%. ## 2. The same route on a first attempt and on a later one If the shadow depended only on the choices the form records (company, first direction, first move in the fight, the dagger, the bullies' fate, the fork), a route would do as well on a later attempt as on a first. It does not. | Route | First attempt | Later attempts | |---|---|---| | Alone / West / Assess the situation / Engage the smaller cadet / Squadmate's revenge / West | 22 / 92 (23.9%) | 0 / 90 (0.0%) | | Alone / West / Assess the situation / Engage the smaller cadet / Squadmate's revenge / Forest | 5 / 29 (17.2%) | 0 / 93 (0.0%) | | Alone / Center / Assess the situation / Engage the smaller cadet / Squadmate's revenge / Forest | 4 / 28 (14.3%) | 0 / 76 (0.0%) | | 1 Squadmate / East / Assess the situation / Engage the smaller cadet / Squadmate's revenge / East | 2 / 32 (6.2%) | 0 / 64 (0.0%) | | Alone / Center / Assess the situation / Daggers or not (protect squadmate) / Left bullies to the dragons / Forest | 1 / 28 (3.6%) | 0 / 64 (0.0%) | | Alone / Center / Assess the situation / Engage the smaller cadet / Squadmate's revenge / West | 1 / 8 (12.5%) | 1 / 82 (1.2%) | | 2 Squadmates / East / Assess the situation / Engage the smaller cadet / Squadmate's revenge / East | 2 / 43 (4.7%) | 0 / 45 (0.0%) | | Alone / East / Assess the situation / Engage the smaller cadet / Squadmate's revenge / East | 1 / 33 (3.0%) | 0 / 53 (0.0%) | Across all 325 recorded routes, comparing first and later attempts on the same route: 86 first-attempt shadows where 36.5 would be expected if the attempt did not matter (odds ratio 637.7, z = 12.0). First and later attempts choose almost the same routes, so the difference is not in the recorded choices: it is something about the account, or a choice the form does not record. ## 3. On one route, is the shadow decided by the cadet's wing, section and squad? The developer says the Black is decided by choices and by what the site knows about the account, with no dice. If the recorded route and the cadet assignment were all of it, every first attempt with the same route and assignment would agree. On the best-recorded route (Alone / West / Assess the situation / Engage the smaller cadet / Squadmate's revenge / West; 22 / 92 (23.9%) on first attempts), 25 assignments have two or more runs and 11 of them disagree (some runs with the shadow, some without). So something else decides it: a choice the form does not record, or a fact about the account other than its assignment. No wing, section or squad stands out: | | shadow / runs | |---|---| | wing | Fourth: 5/28, Third: 7/27, First: 6/21, Second: 3/13 | | section | Claw: 9/31, Flame: 5/30, Tail: 7/28, (blank): 1/3 | | squad | Second: 11/34, First: 6/29, Third: 4/27 | | first_sighting | Brown: 11/48, Green: 9/42 | | sighting_action | Hold still: 10/50, Respect / face it: 4/21, Unknown: 8/14, Weapon: 0/5 | ## 4. There is no single route The 94 form runs with a shadow (all attempts) took 45 different recorded routes: - **Company:** Alone 70, 1 Squadmate 12, 2 Squadmates 12 - **First direction:** West 59, Center 20, East 15 - **First move in the fight:** Assess the situation 83, (not recorded) 9, Even the numbers (smaller cadet) 2 - **The dagger:** Engage the smaller cadet 61, (not recorded) 22, Daggers or not (protect squadmate) 11 - **The bullies' fate:** Squadmate's revenge 59, Left bullies to the dragons 24, Killed the bullies yourself 9, Unknown 2 - **The fork:** West 45, Forest 36, East 12, Unknown 1 Shadows reported without both gates: 3 (likely mis-entered). ## 5. After the shadow: the Blue, then the Black The riders' notes describe one sequence: follow the shadow, a Blue lands and burns another cadet, the Blue leaves, then the Black appears. The Blue is a trial on the way, not an alternative to the Black. - Runs that reached the Black: 77, bonded 55. - Runs that died at the Blue: 15. What their notes say they did: approached the Blue 10, warned the other cadet / intervened 4. - The answer given to the Black, where the note says (bonded / runs): held its gaze 30/42; lowered eyes 0/7; ignored the sound 0/1. Holding its gaze is the only answer recorded with a bond, and it still fails about a quarter of the time. The developer says a rider who knows the system bonds with certainty, so those failures are not explained yet: the notes of the riders who died are too short to show what else they did differently. ## 6. What the bot has tried 415 attempts on 14 accounts; the earliest attempt logged on any account is number 2. Gated tree runs: 236, shadows 0, on 6 recorded routes (the commonest 70 times). The bot has not played a first attempt, so it has not yet tested the one situation where riders report the shadow. The most times any account has repeated one route in a row is 11.