Day 29
Day 29 - May 30, 2026: Troubleshooting Hardware, Grounding Architecture, and Breaking Phase 14 Open
Completing a methodical home AC diagnosis while grounding the Lingua Core Platform delivery boundary and opening Phase 14 with warranted route contracts.
Day 29 continued two investigations that looked unrelated on the surface.
The first was physical: finishing the technical troubleshooting of an older Amana Prestige 2-ton air-conditioning condenser paired with an Amana 80 SSE gas furnace.
The second was architectural: breaking a Phase 14 deadlock in
lingua-core-platform without bypassing the governance rules that had exposed
the gap in the first place.
Both required the same engineering posture:
Isolate the failure -> Verify assumptions -> Respect safety boundaries
-> Ground the next step -> Change only what the evidence warrants
The work was not about forcing a quick answer. It was about reducing uncertainty until the responsible next action became clear.
Continuing The AC Diagnosis
Day 28 restored furnace control power, replaced questionable thermostat wiring, and identified an open high-pressure safety switch. Day 29 continued from that checkpoint.
The diagnostic path had already crossed several layers:
- the fried 3-amp furnace control-board fuse was replaced
- the thermostat’s 2-amp slow-blow glass fuse was tested and showed healthy resistance around 0.2 to 0.5 ohms
- voltage-drop behavior exposed compromised low-voltage control wiring
- new 18/5 solid copper thermostat wire was pulled through the basement joists
- the high-pressure safety loop tested
OL, or open line
That last result mattered. The high-pressure cutout switch had failed open. Because the switch belongs to the equipment’s safety circuit, it could not be treated as a disposable obstacle. A temporary control-wiring bypass was used strictly to gather diagnostic evidence, not as a permanent repair.
With the control path temporarily isolated for testing, the contactor engaged,
the condenser fan ran, and the compressor drew power. Measurements across the
capacitor’s C and HERM terminals showed roughly 325V AC under load.
The electrical behavior was increasingly coherent.
Moving Beyond The Control Circuit
The next question was whether the compressor itself showed an obvious electrical winding failure.
Resistance testing across the compressor winding combinations found no open windings. That did not prove the compressor was mechanically healthy, but it removed another major branch from the fault tree.
Thermal evaluation then provided the more important signal. Even while the compressor was drawing power, the indoor evaporator casing remained around 79 degrees Fahrenheit. The system was electrically active without producing a meaningful cooling effect.
By that point, the remaining likely failure domain had shifted away from low-voltage controls and toward refrigeration or mechanical behavior:
- total refrigerant loss
- a major refrigerant leak
- internal compressor valve failure
That boundary matters. Refrigerant systems and mechanical compressor failures require appropriate professional evaluation. The responsible outcome was not to keep bypassing safety controls or replacing parts speculatively. It was to document the evidence, narrow the fault domain, and stop at the point where the next work requires specialized service.
Engineering Outside Software
The AC troubleshooting reinforced a lesson from Day 28: engineering habits travel well.
The process remained methodical:
- inspect the system before changing it
- identify the first point where expected behavior stops
- test one assumption at a time
- distinguish a diagnostic bypass from a repair
- respect the protective controls
- move to a new fault domain when the evidence requires it
There was also a useful humility in learning an unfamiliar system. Software engineers are comfortable being experts inside familiar abstractions. A physical system removes that comfort quickly. The multimeter results, wiring diagram, and thermal behavior matter more than confidence.
The goal is not to pretend unfamiliarity does not exist. The goal is to work carefully enough that unfamiliarity becomes structured learning rather than reckless improvisation.
Phase 14 Architectural Deadlock
The lingua-core-platform work began from a different kind of stalled system.
Phase 14 had reached an architectural deadlock. Multiple implementation assessments found no warranted slice to build. Delivery-layer concepts had been named, but they had not been grounded strongly enough in the architecture for implementation to proceed without speculation.
The conflict involved three legitimate constraints:
- Documentary Derivation
- No Speculative Extensibility
- delivery-boundary scope
The repository’s governance model was doing its job. It prevented the implementation from inventing a route layer merely because Phase 14 expected one. The absence of a warranted slice was not a productivity failure. It was evidence that the architecture had not yet carried enough information.
That shifted the work from implementation pressure to architectural grounding.
Grounding The Delivery Boundary
ARCHITECTURE.md was updated with a Delivery Boundary Layer. The work also
grounded Static Content Address as a legitimate architectural concept and
aligned ADR-0013 with the clarified model.
The key clarification was that delivery route contracts could derive directly from the Phase 13 projections. That made the Phase 14 boundary concrete without widening it speculatively.
The change was modest in the right way. It did not invent a new runtime, transport framework, or general-purpose extensibility mechanism. It documented the architectural relationship that the next implementation slice needed to follow.
Once that grounding existed, the first warranted route contract became visible:
ReadingPrimitiveSearchProjectionRouteDeliveryContract
The implementation added 18 tests, updated the schema literal and barrel
exports, and refreshed SESSION_STATE.md and CLAUDE.md. Validation remained
green at a baseline of roughly 788 tests across 58 files with 92.67% coverage.
The important part was not simply that a contract landed. The contract could now explain why it was allowed to exist.
Opening The Parallel Route Slices
With the first route contract grounded, the second parallel slice could be assessed against evidence rather than analogy alone:
WritingPrimitiveSearchProjectionRouteDeliveryContract
That implementation also added 18 tests, made an append-only barrel export update, refreshed documentation, and passed validation. The resulting baseline reached roughly 806 tests across 59 files with 92.7% coverage.
A third route contract was then assessed and authorized:
SpellingEntrySearchProjectionRouteDeliveryContract
The spelling route raised a useful identity-field question. That concern was
resolved by checking the derivation boundary carefully. The route contract
derives from SpellingEntrySearchProjection, not directly from
SpellingEntry. The existing projection already defines the relevant
delivery shape.
That distinction is exactly why governance-driven implementation matters. Similar-looking slices should still be assessed individually. Structural parallelism is useful evidence, but it is not permission to skip derivation.
Governance As A Debugging Tool
The Phase 14 deadlock initially looked like blocked implementation. In practice, it was an architectural diagnostic result.
The governance constraints exposed the actual failure:
Named delivery concept
-> insufficient architectural grounding
-> no warranted implementation slice
-> architecture clarification
-> documentary derivation restored
-> route contracts authorized
That is closely related to the AC investigation.
When the outdoor unit failed to start, the right move was not to keep replacing components. It was to measure the control path and locate the open circuit. When Phase 14 could not produce a justified slice, the right move was not to force code into the repository. It was to inspect the architecture and locate the missing derivation boundary.
In both cases, the constraint was useful. Safety controls protected the HVAC equipment. Governance controls protected the platform architecture.
Bypassing either one casually would have made the system harder to trust.
Outcome
Day 29 closed with more clarity in both domains.
The AC troubleshooting reached a responsible technical boundary. The low-voltage path was restored, the failed safety switch was identified, the compressor windings showed no open combinations, and the system’s inability to cool was narrowed to likely mechanical or refrigerant-related causes.
lingua-core-platform moved from Phase 14 deadlock into grounded delivery
work. The architecture now defines the delivery boundary more clearly. Static
Content Address has an explicit place in the model. Two warranted route
contracts landed with validation green, and a third parallel spelling route
contract was assessed and authorized.
The shared lesson was simple: progress is not the same as motion.
Sometimes the fastest responsible way forward is to stop trying to implement the next thing and identify the missing piece of the model. Once the system is understood more accurately, the next step usually becomes smaller, safer, and easier to defend.
Definition Of Done
Day 29 reached a meaningful completion point:
- the 3-amp furnace board fuse replacement remained verified
- the thermostat’s 2-amp slow-blow fuse tested healthy
- the compromised thermostat wiring was replaced with 18/5 solid copper cable
- the high-pressure safety loop tested open
- the contactor, condenser fan, and compressor behavior were isolated
- capacitor
CandHERMterminals showed roughly 325V AC under load - compressor winding combinations showed no open circuits
- the indoor evaporator casing remained around 79 degrees Fahrenheit
- remaining AC failure modes were narrowed to mechanical or refrigerant causes requiring appropriate service
- the Delivery Boundary Layer was grounded in
ARCHITECTURE.md - Static Content Address and
ADR-0013were aligned with the clarified model - reading and writing route delivery contracts landed with 18 tests each
- the spelling route delivery contract was assessed and authorized
- the platform validation baseline remained green at roughly 806 tests across 59 files with approximately 92.7% coverage
The day was a practical reminder that quality gates are not friction to work around. They are instruments for finding the real problem.