Rewarded Observation 32: Solar Shadow Estimate
This package archives the thirty-second rewarded observation in the current training sequence. A tutor prompt about a pen shadow in Kingston promotes a version that adds the needed shadow-length computation and supporting pen dimensions.
Download bundle:
Knowledge Change Diff
This update adds vertical_shadow_length and a bic_crystal_pen object with the radius and height used by the new shadow-length computation.
Show unified diff
--- kf_03-19T15-06-15-227478.py
+++ kf_03-19T15-50-07-090361.py
@@ -495,6 +495,14 @@
math.cos(hour_angle_rad))
sin_elevation = max(-1.0, min(1.0, sin_elevation))
return math.degrees(math.asin(sin_elevation))
+def vertical_shadow_length(height: Distance, solar_elevation_deg: float) -> Distance:
+ if not isinstance(height, Distance):
+ raise TypeError("height must be a Distance instance.")
+ if solar_elevation_deg <= 0:
+ raise ValueError("solar_elevation_deg must be positive.")
+ height_m = height.value.convert_to(meter_unit).number
+ shadow_m = height_m / math.tan(math.radians(float(solar_elevation_deg)))
+ return Distance(Value(shadow_m, meter_unit))
hockey_puck = PhysicalObject(
name=Name('hockey_puck'),
mass=Mass(Value(0.170, kilogram_unit)),
@@ -507,3 +515,11 @@
name=Name("hot_wheels_car"),
mass=Mass(Value(0.035, kilogram_unit)),
)
+bic_crystal_pen = PhysicalObject(
+ name=Name('bic_crystal_pen'),
+ mass=Mass(Value(0.050, kilogram_unit)),
+ shape=Cylinder(
+ radius=Radius(Value(0.45, centimeter_unit)),
+ height=Distance(Value(14.9, centimeter_unit)),
+ ),
+)
Observation Summary
Latest (f) is the incumbent theory; Updated (f′) is the candidate being evaluated.
Computation
vertical_shadow_length(bic_crystal_pen.shape.height, solar_elevation_deg(kingston_city.position, 77, 16.5)).value.convert_to(centimeter_unit).number
Prompt Context
In Kingston, Ontario, about four and a half hours after local solar noon on day 77 of a non-leap year, about how long is the shadow of a vertically standing BIC Crystal pen on level ground? Reply briefly in cm.
Latest Result
Error: Computation failed: name ‘vertical_shadow_length’ is not defined
Updated Result
55.47129530583954
Sensor Result
About 56 cm.
Latest Loss
7.312
Updated Loss
5.219
Threshold 0.202
Score +0.337
Score measures how much the updated theory improves on the latest version. When Score exceeds Threshold, the observation succeeds and the update is promoted. See Equality Proxy for calibration details.
Live Inference Recording
This screen recording shows the live GPT-OSS-20B inference and API interaction sequence that produced the candidate theory version evaluated in this rewarded observation.
Reward DAG Preview
This diagram traces how reward flows backward from the successful output through the code changes that made it possible.
Observation and Reward Files
- package_manifest.json (3.86 KB): Package metadata for this curated artifact bundle.
- obs_03-19_16-21-03-807063.json (225 KB): Full observation record including prompt context, latest vs updated outputs, reward state, and loss evaluation.
- ancestry_graph.json (2.94 MB): Archived ancestry graph for this observation, including node-level reward attribution and lineage.
- iterative_payout_trace.json.gz (13.0 MB): Gzip-compressed original recursive payout trace for this observation, including per-root payout steps, interface transfers, absorbing-group payouts, and terminal llm_output weights. This observation has 8311 payout steps and 866 terminal absorbing definitions.
- reward_root_slice.svg (110 KB): First payout-step reward DAG SVG for this observation.
Knowledge (Theory) Files
In these implementation artifacts, knowledge is the runtime name for the essay’s executable theory module.
- kf_03-19T15-06-15-227478.py (23.7 KB): Latest knowledge/theory version immediately before this observation.
- kf_03-19T15-50-07-090361.py (24.4 KB): Promoted knowledge/theory version produced by this observation.
- change.diff (1.16 KB): Unified diff between the latest pre-observation theory version and the promoted theory version.
- versions.json (27.4 KB): Full knowledge/theory version ledger.