knowledge/README.md

Knowledge base: real Raspberry Pi boards, modules and hobby projects

Built by make kb (scripts/build_kb.py) from cached sources in data/kb/raw/ (git-ignored). Every fact carries its source. Nothing here was written by a language model: facts come from official source files, parsed code, or datasheet passages confirmed by at least two sources.

File	Contents	Main sources
boards.json	32 boards (16 in scope: every Linux Pi with a 40-pin header), SoC, header type; per GPIO family (BCM2835/6/7 + RP3A0, BCM2711, RP1): alternate functions GPIO0–27, default pulls, drive strengths, input thresholds, hardware-PWM channels	raspberrypi/documentation source (line-level links), raspberrypi/utils pinctrl, BCM2711/BCM2835 datasheets
sensors_src.yaml	19 modules (PIR, HC-SR04, relay boards, DHT22, DS18B20, SG90, buzzers, L298N, IR, MQ-2, soil, WS2812, button, LED, LDR, line sensor, encoder, DC motor) with signal levels, v1 action mapping, caveats; 39 source references (29 unique URLs)	datasheets, GPIO Zero docstrings, vendor/tutorial pages found with Firecrawl
projects.json	122 real project scenes (GPIO Zero recipes + 30 Raspberry Pi Foundation projects): device → BCM pin, operations, conflicts; 88 usable under spec 1.0	gpiozero examples (AST), raspberrypilearning repos (AST + regex)
sources.json	Repo commits, licences, datasheet and module URLs

Verification

• Alternate functions: the pinctrl tables were compared cell by cell with independent datasheets.
  • RP1: 252/252 cells equal to the frozen spec's datasheet-verified table.
  • BCM2711: 168/168 equal in substance vs the CM4 datasheet (4 cells differ only as BSCSL vs I2CSL naming).
  • BCM2835: 36/36 GPIO-function cells equal vs BCM2835 ARM Peripherals §6.2.
• Module facts: each needs a primary source (datasheet or library docs) or two independent sources.
• LLM extraction is not trusted. Firecrawl's JSON extraction mis-stated facts on the first docs page, and on one project page it fabricated an entire project (a "Smart Home Automation System" with a DHT11 on GPIO 4) when the page did not render. Firecrawl is used for discovery and quoted passages only; project facts come from the source repositories.

Board differences that matter for a future spec 2.0

	BCM2835/6/7, RP3A0 (Pi 1 B+/A+, 2, 3, Zero, Zero 2 W)	BCM2711 (Pi 4, 400, CM4)	RP1 (Pi 5, 500, CM5)
Hardware PWM	2 channels: GPIO12+18 share PWM0 ch0, GPIO13+19 share ch1	same as left	4 channels: 12, 13, 18 (ch2, shared with 14), 19 (ch3, shared with 15)
Drive strength	2–16 mA in 2 mA steps, default 8 mA	same register, half the current; default 4, max 8 mA	2/4/8/12 mA, reset 4 mA
Input thresholds	V_IL ≤ 0.9 V, V_IH ≥ 1.6 V	V_IL ≤ 0.8 V, V_IH ≥ 2.0 V	(not in docs)
Default pulls	GPIO0–8 up, 9–27 down	GPIO0–8 up, 9–27 down	not stated per pin (D-05)
Alt functions	pinctrl table	pinctrl table (adds SPI3–6, UART2–5, I2C3–6)	frozen spec table

Spec 1.0 labels stay Pi 5/RP1-only. These rows are what a board-aware spec 2.0 would need to encode.

Conflicts for human review

• see-like-a-bat (official project) wires a vibration motor straight to GPIO 14 and drives it with PWM. This contradicts the official GPIO docs ("Do not connect motors directly to the GPIO pins"), spec F-NO-DIRECT-MOTOR and D-01 (no PWM on GPIO14/15). Excluded from positive examples; the spec rule wins.
• interactive-traffic-lights-python uses pin 28 (a board pin number mixed into BCM code) and led_builtin.py uses the Pi's internal activity/power LEDs (GPIO 35/47). Both excluded.
• Two module label mappings are marked needs_review and unused: HC-SR04 "measure distance", because the spec doesn't say which reason applies; and LDR "light level", which comes from RC timing rather than a true analog read.