Repository navigation
Connect real pill captures to /pill and remove the mock path - #27
Open
viktorinkov wants to merge 19 commits into
Open
viktorinkov wants to merge 19 commits into
viktorinkov wants to merge 19 commits into
Conversation
hardware/peel_app moves into mobile/ (lib/hardware, packages/usb_serial, test/hardware). DeviceScreen connects to the real instrument over USB or the simulator (PEEL_SIM), runs blank -> t=0 -> stream -> stop and sends the raw absorbance trace as the scan's hardware payload; 'Continue without device' keeps the POST /pill fallback. The bench screen is a long press on the device step or PEEL_START=debug.
A self-contained package that turns 17_stream output into a verdict on the tablet and maps it onto the backend's PillHardwareResult. Nothing imports it yet: this adds files and modifies none, so no existing behaviour changes. The backend keeps its mock until someone chooses to wire this in. Relocatable by construction. Every intra-package import is relative and data files resolve from the module that reads them, so the directory works at the repo root, nested in another package, or renamed, and its generically named modules (models, noise, library, server) cannot collide with the backend's. tests/test_relocatable.py copies it elsewhere and proves it on every run. - classify_capture(): firmware lines in (raw text, dicts, or the envelopes in hardware/data), ClassificationResult out. Reads red / yellow / green by name, so main's six-key sweep works unchanged. - backend_bridge: the five PillHardwareResult fields, validated against the backend's real model. A reading the rig could not take maps to "unknown", never "fake". - hardware.py: the rig's measured profile. On three all-positive channels the simulator's 0.935 cosine threshold rejected nothing (a grey filter scored "pepto, high confidence"); the rig's is 0.9995, validated on a held-out recording. INVALID_READING refuses what the optics cannot have produced. - The default (simulator) path is bit-exact against the original code. Left out on purpose: the published NIST / PhotochemCAD reference spectra and the module that reads them. Their licence terms are unchecked and this repository is public. The tuning is provisional and the docs say so: the rig was faulted when measured (slow detector, unstable light level), and the evidence is 45 independent noise pairs and two library entries. See HARDWARE_TUNING.md.
Path-scoped like hardware.yml and mobile.yml, so it runs for nothing else. Python 3.10 (the package's floor) and 3.13 (what the backend runs).
…in/truepill-classification' into codex/real-pill-integration
at350
added a commit
that referenced
this pull request
Sep 20, 2026
…n main Parts, power, the wire protocol and commands, the fault engine, a run, the simulator, captures and tools. The classification layer is described through truepill (#26) and its /pill integration (#27) rather than hardware/classify, which never reached main. Two boards, as the firmware has it: an ESP32-S3 sensor board and the ESP32-S3-BOX-3 face over ESP-NOW.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This branch includes all PR 25 commits through
1199d67and PR 26 through7107184. Both are ancestors of this branch. The latest merge is 3ee44b3. Nothing was merged into main.The mobile app uses captured instrument readings. It has no simulator button,
PEEL_SIMstartup mode, or generated-result fallback. It sends five water sweeps and five dissolved-pill sweeps to the real/pillclassifier, then submits the classification and aligned raw readings to research. Skip omits hardware. Request failures stay on Retry. The app rejects responses from the old mock backend.PR 25's synthetic reference matching remains in the pipeline, as requested. It compares real time traces with generated B12, acetaminophen, vitamin C, and caffeine curves. These comparisons are separate from the measured Advil/Pepto classifier. Synthetic matches retain their provenance and ambiguity limits. They are not confirmed chemical identities. The three-color spectrum is never treated as a time trace.
Deepgram leads with the verdict and ResearchReport headline. It gives a closest synthetic match when no measured classifier identity is available. The short prompt preserves recall information, sourced medication facts, unknown results, and report drafting. Raw telemetry stays out of the voice prompt. Research retains bounded sensor evidence.
The backend retains sensor storage fields and additive Elasticsearch migrations from PR 25. The old mock pill request is rejected. New simulated scans, graph demo requests, and graph sample-data fallbacks are disabled. Historical simulated records stay labeled.
Final local checks: 103 existing backend checks passed, Flutter analysis passed, and six existing mobile checks passed. These checks do not establish physical instrument accuracy. The measured library still has only two products and requires physical acceptance on the current instrument.
No Runpod deployment or resource changes. No new APK was built during this merge. Deploy a full repository checkout and rebuild the matching mobile app. The backend needs the sibling
truepill/package. Capture contract and limitations.